NVIDIA is arranging half a trillion dollars to finance GPUs. Stocks of the neoclouds, the specialist clouds that borrow to buy GPUs and rent them out, are ripping. Etched just hit a ten billion dollar valuation for an ASIC built to take GPUs’ steadiest work, and OLIX raised $312 million at $3.3 billion eleven days behind it.

Start with the bullish case, because I hold it. Demand is enormous and still growing: about half of US adults now use AI chatbots and a quarter use them daily. Nobody halves the token budget, and a better model can create its own market. That is why I am bullish in a way I never was on mining, and why custom inference silicon looks like one of the better hardware bets of the decade: the volumes are finally large enough to pay for a chip that does one thing. Even so, that is a quarter of one rich country, and the demand story runs well ahead of the paying base, which is precisely the condition in which build-outs overshoot.

None of that saves the machine.

I started mining Bitcoin at fourteen. On 1 September 2014, Bitmain posted an announcement on the Bitcointalk forum: computing power from its Antminer S2 miners, sold through Hashnest at 0.0016 BTC per GH/s, maintenance fee deducted daily. I was on the other side of that page, buying not the coin but the machinery under it: Antminers, steel boxes built to perform one calculation, and Hashnest claims on somebody else’s machines. Later I bought GPU rigs for Ethereum. Demand never killed any of them. Newer chips killed them on a spreadsheet while Bitcoin was booming: a computer can work perfectly and still be a terrible asset.

That was my first data-center asset class. The data center was smaller.

Twelve years later, NVIDIA started talking like a bank. On 10 August 2026 it announced plans with six of the largest names in capital markets, Apollo and BlackRock and Blackstone among them, to mobilise more than $500 billion of outside capital. The agreements were not final. Jensen Huang then said NVIDIA might, project by project, support the value left in the machines for up to 25% of one financing opportunity. The heroic number matters less than the shape of the deal: the firm selling the chips is helping to organise the money, and residual value has become part of the sales pitch before the current build-out is finished.

The bust will not begin when people stop using AI. It will begin when the predictable inference jobs that were supposed to pay for a GPU fleet leave before the debt does.

Bitcoin mining is cruel in one useful way: its economics are hard to hide.

The network pays out a roughly bounded pot, miners split it by the computing power they contribute, and every machine that joins shrinks everybody else’s slice. Then electricity sends the bill. Every generation was a race between revenue per unit of computing power, which miners call hashprice, and energy cost, and a newer application-specific chip, or ASIC, ran the same calculation far more efficiently. The old box did not break. It crossed a line on a spreadsheet and became irrational to run.

Energy per hash fell about 984-fold between two sparse endpoints, a 2013 Avalon at 9,351 joules per terahash and 9.5 for Bitmain’s 2025 S23 Hyd. That is what the old machine was up against.

Using April 2026’s average Bitcoin network revenue and electricity at five cents per kilowatt-hour, a 29.5 J/TH Antminer S19 Pro failed to cover electricity alone. A 13.5 J/TH S21 XP still cleared it. Both worked; only one paid its power bill, and cooling, labour, financing and the price of the machine were all still waiting outside the calculation.

Miners have a name for a machine on the wrong side of its line: a space heater. The name is unfair, and the unfairness is the best objection to everything I argue after it. A rig underwater at one power price clears at another. Operators undervolt them, trade speed for a smaller bill and move them to power nobody else wants, which is why rigs written off at five cents were still hashing years later. My chart is drawn at one price. Cheap power buys the old machine time, which is not the same as buying it a job.

The resale market moved before the machines wore out. Hashrate Index reported that three ASIC price indexes fell between 82% and 87% during 2022. Bitcoin and its miners carried on, the boxes now worth much less.

Ethereum’s Merge then sent its machines looking for another chain. Across equal thirty-day windows, Ethereum Classic’s estimated computing power rose 264% while revenue per unit fell 79%, and the revenue pool itself fell only 23%, so most of the per-machine collapse was dilution. Compatibility found the machines another calculation without manufacturing another revenue pool.

The data are entirely within Ethereum Classic and do not identify which GPUs or ASICs moved, where they came from, or how much equipment shut down. The revenue estimate excludes unobserved miner income and every cost.

An old GPU has more possible lives than an old Antminer, which gives its owner a convenient answer whenever residual value comes up: we can redeploy it. I always want the next sentence. To what? Which model fits in this card’s memory, which software supports it, and after networking, support, electricity and debt, does the work still pay? “It can run inference” describes compatibility, not residual value. The seller makes the spec-sheet case officially: NVIDIA’s financing post calls the factory “flexible and fungible” and stretches the A100’s economic life “toward a decade”. An old GPU gets no residual value in my model until somebody names the workload.

Mining had hashprice; AI has the dollar-per-GPU-hour rate every rental contract is built on.

Hashnest sold me a claim on a machine somebody else operated; the neoclouds sell that product at institutional scale, financed with debt against the machines. Even the hiring rhymes: the scarce people are not model researchers but power engineers, site developers and whoever can get a grid connection energised.

Bitcoin at least published its executions: one difficulty number, one hashprice, everyone repriced in the open. AI reprices in private, contract by contract, queue by queue, so the owner of an undifferentiated GPU learns the price of obsolescence later, and alone. The absence of a halving is not the absence of a cliff; it is a cliff without a published schedule.

The chip wins, the vendor usually does not

Being right about a chip is not the same as owning the company that sells it. Mining ran that experiment, and nobody in inference hardware brings up the result.

Demand for mining chips beat every forecast. Computing power on the Bitcoin network rose about 351 times between January 2017 and April 2026. Knowing that in 2013, you would have bought the vendors.

You would have lost almost all of it. Revenue per unit of that computing power fell 95% across the same months, out of the same file. Record capacity, earnings per machine down to almost nothing, and only the second line pays for anything.

The buyers went first, diluted by every machine that shipped after theirs. Then the vendors went. KnCMiner, a Swedish maker of Bitcoin ASICs, raised $32 million from investors including Accel and Creandum, shipped real machines to real customers, and declared bankruptcy in May 2016. Its chips worked. Its chief executive blamed the coming halving: “Effectively our cost of coin...will be over the market price”.

The reason is market structure, not engineering. A merchant vendor sells a machine whose value is set by a wage it does not control and cannot see, commits to a wafer months before delivery, and has to win the next generation on the last one’s cash. Selling the machine is the worst seat in a business where the machine is the thing that depreciates.

Bitmain won, and it did not only sell machines. The forum post at the top of this essay is a manufacturer renting out the output of its own Antminers, so a stale generation had somewhere to work: its own hashrate business, while the fresh one went to everybody else. It was the only seat where a short-lived chip never had to be sold to a stranger.

The platforms designing their own accelerators sit in that seat. They own the queue, the model, the software and the customer, so a chip welded to their workload is welded to something they control, and when the model shape moves they are the ones moving it. OpenAI has signed for ten gigawatts of accelerators it designs itself, racks targeted from late 2026 through 2029, inside every six-year book now being written. Anthropic, which still buys NVIDIA systems at scale, signed for multiple gigawatts of next-generation TPUs expected from 2027 and runs Claude across Trainium, TPUs and NVIDIA GPUs to match workloads “to the chips best suited for them”.

The merchant startups sit in KnCMiner’s. Etched, Fractile and OLIX raised more than $800 million this year, and between them they have first silicon, chips still described as future, and first deliveries targeted for the second half of 2027. Samsung SDS says it launched a subscribable service on FuriosaAI hardware, the furthest public step in the group. I found no proof that any of them has a sustained paid workload. That is not a verdict on their engineering, which may well be better than anyone else’s. It is that they must guess which model shape somebody else will still be serving when the wafer starts. Bitmain never had to guess. The test is the first invoice attached to a named workload that keeps paying.

Three clocks on one machine

Alphabet’s disclosed token throughput went from seven billion a minute in October 2025 to about twenty-two billion by July, though what the company says that number counts changed along the way, and its undefined Gemini “serving unit costs” fell 78% during 2025. Neither can value a fleet: price, profit, model mix and utilisation are all missing. Both show why the boom stays loud, because more use can arrive beside a lower cost to serve it.

Microsoft reported $41 billion of company-wide capital spending in its fourth financial quarter of 2026, roughly two thirds of it in what it called short-lived assets, primarily CPUs and GPUs. Yet major owners generally spread broad server and network costs across five or six accounting years. The books move slowly while chips and workloads move quickly. Mining companies converged on roughly three years from opposite directions while the selected AI equipment categories mostly lengthened. None of those policies isolates a GPU, but the mismatch is hard to miss.

Amazon extended server lives from five years to six in 2024, then shortened a subset to five a year later because AI and machine-learning technology was changing faster. The reversal added $1.4 billion to 2025 depreciation and amortisation expense. CoreWeave spreads technology-equipment cost across six years while its weighted-average committed customer-contract term was about five years at the end of 2025.

The filings do not show a cemetery of formally impaired GPUs, and I am not claiming one exists. Only that the accounting life, contract life and earning life of one machine already run on three different clocks. There is no line item for space heaters.

The pot is the one place the analogy genuinely breaks, and I would rather break it myself than have it broken for me. Bitcoin pays its machines from a pot designed to shrink: issuance halves on a schedule, the two step-downs in the first chart, and every machine that joins dilutes the rest within two weeks. AI has no such ceiling. But an open-ended pot secures the aggregate, not the machine, because what killed my Antminers was never the size of the pot; it was a better machine taking the same job, in rising markets as often as falling ones. Underneath the pot sit three different machines.

Bitcoin ASICAI GPUinference ASICcomputesone fixed calculationany parallel jobone model familythe wagehashprice, public$ per GPU-hour, private$ per token, contractedbook lifeabout three yearsfive to six yearstoo new for a booksecond lifespace heaterwe can redeploy itnone, welded to the jobdies ofa better ASICcustom chips take the steady workthe model moves

Who gets paid when the chip changes

One rule sorts this industry: prefer the business that gets paid when the chip changes, and distrust the business that needs the chip not to change before the debt is repaid. Bitmain passed that test; its customers did not. Three positions are left, the merchant vendor having been sorted above.

The chip seller is the safest seat, because it is also selling the system around the chip. NVIDIA reported $75.246 billion of Data Center revenue in its latest quarter and roughly $14.8 billion of networking revenue, and it is working to make that network useful around accelerators other companies design. Microsoft already says Maia 200 uses its own transport over standard Ethernet, which reads to me like a polite way of routing around it, and NVIDIA’s own filing warns that customer-designed chips may not need every feature in its systems. Nobody discloses how much a vendor keeps when the accelerator belongs to the customer. The purest version of the seat is Broadcom’s: it builds custom accelerators for the platforms, booked $10.8 billion of AI semiconductor revenue in its latest quarter and guided the next quarter to $16 billion. It gets paid when the chip changes, which is also why the clean short on this thesis does not exist.

The borrower who owns the fleet is where the analogy bites hardest. CoreWeave is the cleanest public expression: at the end of June it disclosed $35.6 billion of debt principal and $16.3 billion of recognised operating-lease liabilities, and it spent $14.1 billion on equipment in six months. The defence is serious: $103.7 billion of work still owed under contracts, with 98% of quarterly revenue tied to customer commitments. The other side is that three customers supplied 72% of Q2 revenue and quarterly interest expense was $640 million. A backlog does not pay interest by itself. Contract cash has to cover interest, leases, replacement chips and the cost of finding the next job, and that arithmetic belongs to every leveraged fleet in the industry, listed or not.

In March CoreWeave closed an $8.5 billion facility at investment grade, SOFR plus 2.25, against GPUs wrapped in a customer contract. In August it closed a $2.6 billion facility at SOFR plus 5.50, below investment grade, a roughly five-year loan against contracts averaging about three years, with no residual-value guarantee anywhere in the filed agreement. The 325 basis points between those two loans is a first public price for the years a GPU must earn after its contract ends.

Nebius has a direct answer: large contracts bring material prepayments, expected payback is about twenty-two months, and pricing for older GPUs improved more than 30% from the prior quarter. I want that to be true. It has not met a chip refresh yet, and that is where such claims are settled.

The landlord is a different animal, and the market keeps pricing it as the same one. Applied Digital’s core leases sell the powered building, cooling and electrical system rather than the accelerator inside. A mine can leave behind a power connection; it does not leave behind an inference customer, which is why this is the position I would rather hold through any hardware transition, with one condition: distance from the GPU only helps when the product survives the next GPU. Only 100 of 1,410 contracted megawatts was operating and earning revenue at year-end. Contracts still have to become buildings.

What happens next

Predictions, so this essay can be graded. The footnotes arrive quarterly, and I keep score on a ledger that logs the prints cutting against me too.

The six-year GPU book will not survive the decade. Amazon’s cut from six back to five was the first crack; before 2030 a hyperscaler or major neocloud discloses four or fewer for accelerators, and the sell side will call it prudence when it is a confession. Miners tried five and two and met at three.

AI will get its hashprice. The H100 rental series charted above fell 57% in under three years, and the sceptic’s answer rides in the same series: one-year contract rates up 38% from their October low, on-demand capacity sold out since February. Hashprice rallied before every collapse too. None of it is yet a number a lender marks a fleet against, which is what made the difficulty adjustment lethal rather than merely interesting. One of these series hardens into that number before 2030: quoted continuously, for a named generation of hardware, cited in a credit agreement. The moment it exists the private cliff becomes a public one, and last-generation GPU-hours grind toward power cost the way the electricity line ground down my Antminers.

Custom silicon takes the payroll and the GPU keeps the frontier. By 2030 the majority of the largest platforms’ own stable, high-volume inference, the feeds, the recommendations, the search-shaped work, runs on chips they control. The aggregate GPU market can grow while this happens, which is why almost nobody will notice.

The first liquid secondary market in accelerators prints the number everyone is avoiding, the way Hashrate Index printed ASIC prices in 2022, and the first large GPU impairment lands in a quarter of record demand, blamed on a customer event rather than the curve. That is when the accounting life and the earning life meet in public.

Most neoclouds will not die as clouds. The survivors will quietly become power and shell companies, the way the surviving miners did, because the durable asset underneath was always the grid connection and the building. The rest get absorbed for their contracts and their megawatts, and their GPUs go to the secondary market.

The bill arrives before the boom ends

At fourteen I thought I was buying computing power. What I had bought was a temporary spread between mining revenue and electricity, wrapped in a machine. Turner painted the asset class in 1839: a working ship, nothing wrong with her, under tow behind the machine that took her job.

Today’s version comes with hyperscale customers, long contracts, financing syndicates and scarce grid connections, every one of which can be real while an individual GPU investment fails. The bust arrives loud: record electricity demand, full pipelines, more tokens every quarter, and rental rates falling anyway. The phrase “AI demand” tells me none of that.

The AI data-center boom is a repricing event dressed as a building programme: hundreds of billions of dollars a year underwritten on the belief that a general-purpose chip keeps its best job for five or six accounting years, when everything mining taught me says the best job leaves first, quietly, while the buildings are still full. Somebody ends up holding the one asset this industry has never learned to price: a space heater with a six-year book.

Before I underwrite another GPU rack, I want one answer:

Which named customer and workload will still pay for this machine if custom chips take more of the predictable work?

If the answer is “AI demand”, I have seen this asset before. It was a mining rig with better branding.

Originally published at josef.cn/blog/asic. Every prediction above is graded in public on the space-heater ledger; when a print lands, for or against me, I will note it here.

Research cutoff: 16 August 2026. This is a research view, not investment advice.

In 1770 Wolfgang von Kempelen, a Hungarian civil servant working at the court of Empress Maria Theresa in Vienna, showed off a most impressive machine to amaze and astound the empress and her court: a cabinet with a person inside that played chess. The cabinet, with a wooden figure of a Turkish puff-chested gentleman dressed in Ottoman robes sitting inside, played chess for eighty-four years and finally came to an end in 1854 when it was burned in the Chinese Museum in Philadelphia.

The Turk was not the first cabinet of wonders. On 30 May 1739, another remarkable automaton was unveiled to the public in Paris. The Digesting Duck, as it became known, was created by Jacques de Vaucanson and appeared to eat grains of corn and then pass them through its system to be excreted as perfectly formed duck pellets. The ingested grains entered the bird via a container in its back end, whilst the pre-loaded pellets that were excreted came from a second container. The flute player and the tabor player of Vaucanson, as well as the automated silk loom, were later to be created for the French state.

It was not to be the first of a long line of chess playing machines by a very long margin. As he would go on to do with so many other inventions, Kempelen spent the rest of his life refining them but the chess playing Turk was a momentary diversion and he was extremely unwilling to take the thing on tour. It would be over 80 years before he got his wish.

The Turk was never a mechanical chess player. The operator, sitting on a sliding seat, would follow the game on a pegboard covered by a magnet for each piece. Above the board the operator would use magnets on hung pegs to follow the game. The arm of the Turk would be moved by the operator via a pantograph. The real gears shown to the audience by candlelight were simply connected to nothing. The audience was always shown the wrong compartment. This was a matter of stage management.

For 84 years the cabinet was unbeaten by human opponents. It traveled throughout Europe and America. Ben Franklin challenged the Turk to a number of games while living in Paris. Napoleon Bonaparte on several occasions played the automaton at Schönbrunn Palace. Although legend has it that Napoleon attempted to cheat by making an illegal move the Turk simply removed the piece from the board. On the third game however the Turk took a different tack – it swept all the chess pieces off the table with its arm. According to reports the Emperor was thrilled.

Maelzel who also promoted the use of metronomes for orchestras (the reason why conductors today use them) was also a great inventor in his own right and amongst his inventions were also ear trumpets for Beethoven, a battle symphony that Beethoven wrote for Maelzel and even a canon. Beethoven however sued Maelzel over the battle symphony and also because Maelzel did not fulfill his promises as laid out in the contract for the canon. The Turk was just one of many items in Maelzel’s traveling show of novel inventions. He also had mechanical trumpeters and a very large and impressive diorama of the burning of Moscow.

The months were Whit, Jun, Jul, Aug in 1826 when the famous Chess-Playing Machine was wheeled into the assembly rooms of the National Hotel, No. 112 Broadway, New York City on April 13th. After a few months here it was moved to Boston, then to Philadelphia, and next to Baltimore where it remained for several months, finally beating Mlle. SPAHN but losing to the signer of the Declaration of Independence, Charles CARROLL. Some years later, after Maelzel’s death, the famous Turk was found in the Chinese Museum in the old Rogers Building, Philadelphia. It was chained up and burning to the ground on July 7, 1854.

There followed a long list of people who had operated the Turk in its many tours of the chess halls of Europe and America. They included Johann Allgaier, Boncourt, Aaron Alexandre, William Lewis, and finally, in his old age, Jacques Mouret who sold a full account of the thing for a drink to a Hungarian journalist and which appeared in Le Magasin Pittoresque in 1834 under the title An attempt to analyze the automaton chess-player of M. Kempelen. Almost no one believed a word of it. The American writer Edgar Allan Poe, for example, was able to see the Turk playing in Richmond in 1835. A year later he published Maelzel’s Chess-Player in which he had diagnosised the thing correctly as a man in a cabinet. The Turk, after all, lost occasional games of chess.

After over eighty-four years of exciting European and American audiences, the famous chess automaton finally fell into the hands of the Chinese Museum in Philadelphia where unfortunately it was burnt down in July 1854 by an unknown person.

In February Will Manidis launched a website called "Tool Shaped Objects". He writes there that the "market for feeling productive" is "orders of magnitude larger" than the "market for being productive".

This spring, a collection of 67 frontier models beaten by their best counterpart, failing together on identical input in identical ways. Last autumn, some 90% of the roboticists attending the main conference on the field agreed that humanoids will not replace the bulk of human workers in 25 years or so. In experiments run at Figure’s factory in kvin topsia, humanoids work at a rate of about $25 per robot-operating-hour. They work in 10-hour shifts, loading 90,000 parts onto real cars.

1819: After watching the Turk play at various London establishments for a total of 14 hours Charles Babbage, in a mood perhaps a bit supported by the losses he had incurred, exposed the robotic entity for the clever hoax that it was. The charade that Kempelen had created, of course, had another property, though. It posed a problem to which Babbage would devote the remainder of his life: the construction of a machine that could play chess.

Seventeen years later, Poe was still struggling with this problem, coming to a realization in his classic essay “Maelzel’s Chess Player” (1836): “Arithmetical or algebraical calculations are, from their very nature, fixed and determinate. Certain data being given, certain results necessarily and inevitably follow. But the case is widely different with the Chess-Player. With him there is no determinate progression. No one move in chess necessarily follows upon any one other.

For arithmetic and for algebra, the operations are of fixed and determinate type, and it follows from the nature of the given data that the operations determined by these will also be of fixed and determinable character. With chess-playing automatons, however, the case is radically otherwise. In a game of chess there is no principle of movement which necessitates any given move from a given position of the men; and it follows, of course, that no given position of the men at any given stage of the game permits any determination as to how the men shall be placed at any succeeding stage of the game. — Edgar Allan Poe, "Maelzel's Chess-Player", 1836

Determinate work can easily be translated into machinery. It is the work that is based on chance, that is variable and non-deterministic, that cannot be translated into a mechanical system. The person in the cabinet had to be a human being, a Chess-Player.

When Charles Hooper, a cabinet maker, built a automat called Ajeeb in 1868 for some 30,000 visits and games of chess with the likes of Harry Houdini, Theodore Roosevelt and O. Henry a man was sitting in the compartment. Later he was supplied by others, among them Constant Ferdinand Burille, Charles Moehle, Charles Francis Barker and the former US champion Albert Hodges. It was Hodges who in 1898 introduced the Ajeeb to Harry Nelson Pillsbury, then one of the strongest chess players in the world.

For decades, various operators, including Constant Ferdinand Burille, Charles Moehle, Charles Francis Barker, and the ex-US chess champion Albert Hodges, hid inside Ajeeb compartments and operated it for thousands of shows. In 1898, Hodges brought the Ajeeb to the USA and it was introduced to Harry Nelson Pillsbury, then one of the world’s strongest players, who operated it ‘on and off’ for ten years until 1904. He was, reportedly, one of the finest operators that ever sat in the box.

Of greater interest is the Mephisto chess playing machine constructed by Charles Godfrey Gumpel, an Alsatian manufacturer of artificial limbs, who took some six or seven years to build the robot and first displayed it at his home in Leicester Square in 1878. Gumpel’s remarkable machine was to play chess against Human opponents remotely, its compartments being empty of any human presence, whilst being operated by a chess master, Isidor Gunsberg, in another room, via electromechanical transmission.

A prototype for a marketplace for invisible human work was launched by Amazon in 2005 under the name Mechanical Turk. The company was marketing it as “artificial artificial intelligence”. In 2018 Google demoed a phone call with a human to book a hair cut with Duplex. It would be revealed later that some of the calls would be completed by human operators. A large share of work completed by Engineer.ai’s app-building assistant, reported the Wall Street Journal in 2019, was done by human engineers.

Builder.ai, now known as 1X, had previously raised around $445m from investors including Microsoft. Its chatbot assistant, “Natasha”, had been sending the vast majority of work to some 700 or so human engineers, and the various administrators of the company had filed their papers as a public company in 2025.

The just reported extent of human involvement in tasks touted as facilitated by Just Walk Out technology at Amazon’s Fresh supermarkets—reviewing a near 700 per 1,000 transactions carried out remotely by approximately 1,000 contract workers in India via an internal 50% benchmark—was vigorously disputed by the retail giant who shortly thereafter disengaged from relying on this form of invisible labor for the purposes of facilitating in-store Fresh shopping. The gap between what is demoed using AI and what actually goes on in a hidden call center has been and continues to be exposed by regulators and is actionable under law, described pejoratively as AI-washing.

This year 1X have started taking pre-orders for the Neo home humanoid, which in its “Expert Mode” allows a human operator in a VR headset to see what the robot sees and fold your laundry in a room you will never have seen. This is staging ‘fake videos with human teleoperators in next room’ as Figure’s Brett Adcock describes. Mephisto, from 1878, was similar.

For a century or so the illusion was to continue; but in 1989 the show was brought up to date by illusion designer John Gaughan who built a perfect replica of Kempelen’s Cabinet of Automata and displayed it at a conference of the Society of Ally Magic held in New York in November of that year. In Gaughan’s version the space previously occupied by the chess player was now home to a computer, and a chess program, which in demonstrations, played acceptable chess against humans.

Kempelen’s six-of-the-last-seven decades were consumed by his attempt to produce a speaking machine: a device with bellows, reeds and a rubber mouth which could laboriously utter simple child-like words and short phrases, which he published in full in 1791 with all the cabinets open as if to prove a point. The line of work that he had started here would stretch through Bell’s telephone to the voice models which today are pesky to hang up on you.

The AI-washing by the SEC of Delphia (personal finance and data for machine learning) and Global Predictions (financial advisior, “first AI-regulated” – no, they did not have the capacity to deliver on their promises) followed in March 2024 with fines of $225,000 and $175,000 respectively. The high-profile case of Presto, a restaurant-technology company listed via a SPAC in September 2022 and trading on the Nasdaq until September 2024, as AI-washing followed in January 2025.

More recently, evidence supporting the need for human involvement in physical robotic systems was amassed during the eleven-month Spartanburg, SC trial. Two Figure 02 robots, operating under the direction of the trial’s articulate counsel, were placed to work in ten hour days, five days a week, for over a year logging over 1,250 hours of runtime. During that time, the two units loaded in excess of 90,000 parts, directly or indirectly enabled the production of over 30,000 X3s, prior to the project’s cessation in 2025. Today, BMW is working with Figure 03 to implement parts sequencing in the plant’s logistics department and has established a Center of Competence for Physical AI at its Leipzig campus. Just last week, robotics darling Unitree filed for a USD 4.2 billion listing on the Shanghai STAR Market. Retail book was wildly over-subscribed and, on the issue’s first trading day, the stock more than quadrupled from the issue price to close at 460.

Many robotics researchers, including those who work for robotics companies, share the intuition that while some robots may replace some human workers in the near future, most humans will not lose their jobs to robots in the near future. In a recent debate held at IROS 2025 in Zhejiang province, China, organizer Wang Xingxing of Unitree (which is listed to go public on Shanghai STAR Market for about 4.2 billion Chinese yuan) said that based on the audience poll before his speech, 80% of the people in the audience agreed with him that humanoids will not replace most human workers in the near future. After his speech, the poll result changed to 90%. Hans Moravec wrote about the inverse relationship between progress in a robot’s solving dancing and its solving hands in 1988, and the same holds for 38 years since.

Independent errors only occur when there are other errors in the system (i.e. in aircraft, in RAID arrays, in juries, other members make errors). This error is external to our control and is currently out of our sampled distribution. All of us are working off the same feeds off the net. None of us currently have good contact with matter (for example, grippers which slip and recover from the slip). What we have are samples from a process that hasn’t been covered in anyone’s training set and cannot be deduced from larger models of the data. In other words, the data you really need to make robots better cannot be faked.

In edge cases old cloud behavior fails. The robot runs at batch size 1. There is no amortization of prior computation to speak of. All the latency that you demonstrate is all the latency that you will ever have to deal with, as opposed to some average and then some overhead for communication. Surprisingly to many the primary bottleneck in use cases here is not the bandwidth to the robot, but the overhead to launch work in the first place. Memory ceases to be an abstract concept. The flash wearing out in a one-write-per-second use case at some point introduces a cost to each write, which in turn demands new policies around how to commit things to memory in order to learn to maximize action value in that single write. This is to say that even the redundancy of having multiple policies, of having a stronger model watching a weaker model’s internal state for slight wobbles that hint at impending fumbles and then taking over in time, is not somehow eradicated at the edge but rather becomes even more important. A chatbot that pauses for 10 seconds to think is engaged in deep thought. An arm that pauses for 10 seconds to think before dropping what it was holding is, by contrast, inconsistent and therefore will have to be controlled by a stronger policy in real time.

There is nothing to be gained from rebooting the cabinetry of the last era: the new orchestration layer is the cabinet itself scaled globally, and it opens doors one at a time. The committee of correlated guessers is just one guesser in a trench coat. This is Turks all the way down.

If the last era’s big idea was more compute per model, I predict that this era will be more world per model. Build the bellows.

Build the bellows.

The Genghis Khan Operating System

Foxconn and TSMC are running an 800-year-old operating system. It conquered the world once. Here is how the West takes it back.

The world’s richest empire in 1220 was destroyed by a group of semi-literate herders. They were operating on an 800-year-old (much improved) operating system which had served them well when they were just a loose collection of herders. That same operating system (in some form) runs today at Foxconn and TSMC – and is the reason that the West has so far failed to counter Asian success.

Start with the herders.

The Khwarazmian Empire, with its capital in the Caspian region, stretched from the Hindu Kush to the southern shores of the great sea. Its major cities were Samarkand and Bukhara, rich centers of culture, full of beautiful libraries, skilled astronomers, and enormous riches – far more than the men, driven by their thirst for power, needed. Its massive army dwarfed that of the Mongols.

Within two years the Empire had crumbled. Many of its cities were burnt to the ground, Samarkand, Bukhara and Khiva being among them. When the Shah finally sought to escape his Mongol persecutors he was murdered on an island in the Caspian, hunted down by Genghis and his men who had traveled half across the world to defeat him.

Although there probably had been as many as a million men of the Khan’s army, none had ever lived in cities and most did not even speak a written language until it was provided by their leader, Genghis. It would have been impossible for such people to take a single walled town, let alone the largest empire ever heard of, twice as large as that of the Romans, within a single generation.

While always fighting against greatly superior numbers and never having any considerable funds to his name, Genghis Khan had something which proved to be adequate: an OS written by a single man and functioning to this day – albeit outside of the West.

How he actually did it

Genghis Khan did something really important. It wasn’t his inventions, that he had none. It wasn’t his fighting tactics. Rather, he built a very efficient organization. This organization was made out of the least promising material on earth (scattered and feuding nomads). But he organized them so well that they in the end out-thought and out-ran all the settled empires on earth.

He set up the organization by cutting out inheritance. You wouldn’t rise through a Mongolian tribe as you would in a traditional company. Instead, Genghis broke up the tribes into groups of ten, hundred, thousand and ten thousand men. The highest ranked man in each of these groups would have been promoted on his merits, regardless of his family.

His general Subutai was a blacksmith's son from a forest tribe. He is recorded to have conquered more ground than any other general in history. Unlike in other armies of the time where the best born son of a noble was given the command of an army, the Mongols gave the command to the best person for the job.

All members of a division of ten are to obey their leader without hesitation. If one member of a group of ten runs away during battle the whole group of ten is to be put to death.

A modern western army is composed of "coalition" units made up of nobles who are often sent to fight abroad against their will, and who return home as they please. They are a patchwork of disparate units which are often at odds with one another. The herder on the steppe was a single unit, and that was how he functioned.

But the greatest tool he had was speed. Within a week of being sent, a message reached its destination. Twenty-five miles of riding were followed by twenty-five miles of rest, to keep the horses fresh. Europe’s news traveled at the pace of a walking man; the Mongols decided upon matters before their enemies even knew that a decision had been reached.

He wasn’t interested in having what he didn’t have, like the ability to take walled cities. The Mongols’ horsemen couldn’t breach stone walls, so he went to China for the best engineers in the land, the men who could design and launch catapults to set cities on fire. Four thousand miles from the battlefields in which he had just defeated the Chinese, he carried these same engineers with him. It wasn’t long before, instead of capturing these at the head of his army, he would be commissioning his own. The best capacity of his enemies was to become his own.

Mass annihilation of whole towns. Yes, that was the most ‘cost-effective’ way of breaking a town or city: causing sufficient losses to deter others down the line. Clearly a chilling and very dark account of human history as well.

That is how a million herders beat the world.

The system never died. It moved.

For a glimpse of what life is like on the floor of a Foxconn factory in Shenzhen making your iPhone, click here to read the Khan’s army in high-visibility vests: new workers go through a rigorous boot camp before being put on the production line and managed in strict hierarchical structures of thirteen levels deep, each worker admonished to remember the founder’s principles on the walls of the factory including “Growth, thy name is suffering” (Workers as Machines: Inside China’s Supplier Fabricthem, 2013; Lewis 2009:527).

Foxconn is studied most comprehensively in a book that calls the workers in Foxconn’s plants “Machines”. The founder of Foxconn, Terry Gou, says in the quote above that he wants to “grow” by having “suffering” as his name for “growth”.

As human beings are also animals, to manage one million animals gives me a headache.

By 2012 however Gou was saying something rather different in a TV interview, apologising for the comment a day later: ‘Workers as animals’ in Foxconn village.

His legendary vision and management philosophy have often been referred to. He is a great fan of Genghis Khan and the Khan’s Temple beads are among his favorite accessories.

If I decide to do it, I will break through. If I fail, I will die on the battlefield like Genghis Khan.

Terry Gou

But then there is the special that Khan set up for himself: the line item for the systems dark underbelly. He set up the dorms at the Longhua complex outside of Shenzhen and in 2010 a stream of suicides occurred there with more than a dozen jumping from the roofs of their dorms. Instead of slowing down the production line, the company in charge of the facility set up nets under the roof eaves and had the employees sign a pledge not to take their own lives. Two years later, during a brutal lockdown, tens of thousands of the workers at another Foxconn facility collectively decided to walk home along the Shenzhen Stateown motorway. Whether the attempted suicide rate at Foxconn is higher than for the rest of the Chinese population or not (it is not), a company that has to put up nets to catch people jumping from the roofs of their dorms and then require them to sign a pledge to not attempt to take their own lives within a few years is a huge machine that treats its workers as part of the production line.

The very same discipline that built mass amounts of chips on Foxconn production lines also enabled the country to build more than half the ships in the world, most of the commercial drones, and a third of the world’s EV batteries. Much of that is produced by BYD which has actually managed to outsell Tesla on many models.

TSMC runs the system out of super clean factories, pocketing 40% of every dollar spent on chips while TSMC staff are constantly monitored by security. Until recently, the same kind of culture was used to build out the two biggest companies in China’s hi-tech industry: Huawei under Ren Zhengfei, the company’s founder and a former army engineer, who named the company culture himself.

We win against the lions. We have a strong desire to win and we are not afraid of losing. In the end, we wear the lion out. (Ren Zhengfei, founder of Huawei)

Cost mattered once. It built the first factories, and it stopped being the reason a long time ago. While the West congratulated itself on having moved beyond all of this, the Khan's operating system took the physical economy of the twenty-first century. The West is Khwarazmia.

“Cost mattered at some point in time. In the US you could have a meeting of tooling engineers and not fill the room. In China you could fill several football fields. The vocational skill is just much deeper. — Tim Cook, CEO of Apple (Tim Cook, CEO of Apple)

Earlier, cost had been the main driver for setting up the first factories. By now, however, that reason for change has long been eclipsed. While the West has long since moved on from this type of production, the operating system of the Khan has in the meantime taken control of the physical economy of the 21st century.

The West is Khwarazmia

The West is Khwarazmia

We have the libraries and the brands and the software, and we hold meetings about psychological safety while the people who will inherit the century bolt robots to a factory floor. Like every doomed civilisation before us we are ruled by an aristocracy, only ours runs on credentials rather than blood, and it promotes the well-schooled over the dangerous. Somewhere along the way we forgot how to build anything physical at scale. We even diagnosed it ourselves, and then did nothing.

Richard Hanania

@RichardHanania

·

Aug 8, 2024

TSMC is having trouble replicating its success in Arizona because Americans won’t work as hard. They won’t get up in the middle of the night to help with emergencies or do work outside their job descriptions.

The world benefits from Asian workaholism. We have other strengths.

TSMC can’t replicate their success in Arizona because Americans won’t work hard enough to support a 24/7 operation. They won’t do work outside of their normal job duties in the middle of the night to support an emergency. The world benefits from Asian workaholism. We have other strengths. (Richard Hanania, @RichardHanania, Aug 8, 2024)

Once we do get to try our hand at a project of this scale, we will learn some painful lessons. For example, TSMC’s Taiwanese managers at the Arizona fab were quite explicit in writing in a lawsuit that the American workers there lacked commitment to their work and did not understand how to get work done within restrictions of job description. In fact, the founder of TSMC, Morris Chang, wrote many years ago that an effort like this by the Americans to make their own chips would be “a very expensive exercise in futility.”

For a start, a machine breaks down at 1:00 AM and is fixed the next morning in the United States. In Taiwan, it is fixed at 2:00 AM. (Morris Chang, founder of TSMC)

Everyone in politics wants to “bring back” the operating system by building the fab here and thereby having the factories for making chips on American soil and having Americans work 12 hours a night to complete the work as in Taiwan.

But this would be to mistake the whole purpose of gaining supremacy over the Khan’s operating system in the first place. Importing all the Taiwanese to run all the new facilities on the West Coast is not going to solve anything for the West. We can’t even reproduce the culture of hard work and good management that the Taiwanese already possess. And if, against all hope, it did work and we became as hard working as the Taiwanese, then the West would have lost its soul.

How the West actually wins

And, thus, you must run the half of the enemy’s system that you have already abandoned, in novel and competitive ways.

Run the full meritocracy as coldheartedly as you would any other system. Right now, the deepest weakness of the West is a manager class who give undue weight to credentials over results and will not promote a ‘below-the-line’ (i.e. non-engineer) genius (like a blacksmith) no matter how wonderful. But there are already many companies operating as first class meritocracies. Anduril makes and deploys high-end defensive weapons faster than the century-old ‘defence primes’ (like Boeing and Lockheed) have ever managed. Hadrian’s Chris Power builds robots in a matter of weeks to turn baristas and bus drivers into first-class machinists. All these companies run full meritocracy and suffer no problems with ‘lazy Americans’. They run hard and decide things quickly. The Yam and the meritocracy have already been re-implemented in the best parts of the West – in Texas and in California.

What is unique to our side of the world, the Asian operating system, is being preached by the founder of NVIDIA (most valuable company on earth and driving force behind AI revolution) to students at Stanford University in 2024. Jensen Huang says that it is not intelligence that is necessary to be great at something but character, and that suffering builds character.

For all of you, I wish upon you ample doses of pain and suffering.

Jensen Huang, to Stanford, 2024

He also tells Stanford MBA students that he got where he is by being of very strong character and having suffered greatly in his life. Here is a photo of Gou with a big painting he commissioned at the TSMC plant in Shenzhen of that same profound saying. Note that same word ‘suffered’ and that the following would be to unleash a ton of intensity at something that no one can copy. But yes, we do have the creed to drive our hard working people to their fullest potential. The question is whether or not we have the nerve to implement such a culture and to build a factory that runs (and therefore does not require) nets of HR people.

Copy what works and bring the very best to the job (Genghis took huge numbers of Chinese engineers with him when he left China). All that is required is to hire the best and the smartest and allow them to run with the operating system as fast as they can go. Don’t reconstitute the entire human army to fight the East, just automate the assembly lines of Foxconn and the very floors of the two Indian manufacturing sites where bodies and nets are used in their production of electronics. This is the human-reducing, labor-intensive-software-enabled operating system that someone recently described (I don’t know who to credit) as The Suffering-Engineered Operating System (SEOS). It’s been developed on a massive scale by the West’s greatest competitor, and with one factory Chris Power is planning to build in Arizona (his company Hadrian is already manufacturing automated factories for rockets and for production of satellite components as well as for converting baristas and bus drivers into capable machinists within about a month) all of that can be made obsolete.

This West-based advantage is the merit-based way of building an Operation that can be as efficient, as fast and as disciplined as any of the Asian players, but with a key difference: to build the most demanding machine ever created for human beings and then not to run it on the suffering of those same human beings, but on the capital and code that now constitute the principal means of production and that, more and more, have replaced labor entirely.

It boils down to this: when you decide to run as hot as the other half of the world, you had better build a gage to see how hot you are running! The Khan’s system for developing talent and the 40-year or so TSMC engineers have to develop the stuff that no one can copy to build the stuff that FOXCONN can build in the most horrible of conditions, in numbers that you can’t even imagine in addition to the horrible wages for body parts that they are – in a literal sense – selling to the highest bidder for two cents on the dollar… That is the stuff you can build for!

Suffering is to be heeded. The Khan’s system and the Foxconn assembly lines lack one critical tool to gauge the limits of what human beings can endure to do. That tool would be enough to turn their collective sense of hardship into a manageable thing to track on a big screen in front of you. For the West, with all its empirical warts, this is a distinguishing advantage: it can create the most demanding thing on earth and refuse to run it on human suffering.

As for rebuilding American industry, people are putting their money where their mouths are and that’s not about margin either.

Reindustrialisation is bringing critical functions back to the US, not because it is cheaper, but because otherwise the country would be a vassal state of those that currently make everything for it. (Palmer Luckey, founder of Anduril)

The choice

He wouldn’t even recognize a single chip. What he would recognize is a system, with numbered ranks, with code memorized and recited, with absorbed enemy engineers. Speed, will, readiness to do things that the soft and rich will not do — that is the system of the Khan.

That vast, efficient system that took the whole world from a mere start which was almost zero and ran the whole of physical economy for centuries in Asia against insurmountable odds, while the West is busy studying, writing and negotiating in a steady pursuit of a semblance of ‘balance’, is once again taking over the physical economy of Asia, and it is Khwarazm.

We are Khwarazmia. A 800 year old civilization. We can keep going on as a wealthy and isolatedKhwarazmia, but we know how that ends for us.

Or we can try to be different, to build a different system, harder and more unknown, but which could be worthy of the world’s first superpower. We could take the operating system that is being used against us, strip it of human bodies, and use it to win the only war a free and wealthy people can win.

The Khan is eight centuries dead.

His machine is still running. Who will run it next?

William did not make any plans. He built the center part of the canopy, the fabric work for the roof, and the lifting gear needed to erect it. He made wooden templates for the sculptor to cut the stone for the Angels above the screens. He even went on to design and make the stone recessed “seats” for the monks inside the choir.

In 1178 William of Sens fell 50 feet from the scaffold. Trying to conduct the construction from his sickbed as master builder (maître d’œuvre) by means of his deputy was not to work. He returned to France where he died a short time later. He had no drawings for the others to follow. The body of the master builder was taken back to France. The knowledge with it.

Until recently, English was not even spoken at site of Évreux Cathedral, where Gautier de Varinfroy, Werkmeister, worked from 1253 for salary plus one condition: he had to live in Évreux, and was not allowed to leave the town for more than two months. This amazing man must have known the great abbe Suger of St. Denis very well. As presensence of Emperor Charles IV, the master builder Werkmeister, his equivalent in German speaking countries, was the contract.

At twenty-three years of age the young Peter Parler was summoned by the Emperor Charles IV. to complete the St. Vitus Cathedral in Prague. The family name of the architect originated from his occupation as a master builder – his family name Parler translates into Foreman or Master Builder in French. The young Parler had learned his trade from his father, who worked as a builder in Cologne before he went to Prague to revolutionize stone-vaulting there by introducing the net vaulting technique.

His brother Johann became the Master of the Freiburg Minster. Johann’s son Michael succeeded him as Master at Strasbourg. Peter’s own sons followed in his footsteps as his assistants at the St. Vitus Cathedral in Prague, after which they built other churches in Vienna and in Kutná Hora. The knowledge of the geometric innovations of the Parlers was thus transmitted from father to son. Each new cathedral in turn became a school for the sons. The Parlers had established a lucrative franchise.

The builder’s lodge (Bauhütte) was the building institution. In the course of the construction of a cathedral, it was set up as a kind of self-governing community of all masons, apprentices and masters working on the building site. The Bauhütte at Strasbourg, for example, worked for over four hundred years.

The sites varied greatly and presented their own specific problems in respect of stones, soil, clergy and towns. Chartres was built on Lutetian limestone, which is found beneath the Paris basin. Strasbourg, on the other hand, made use of red Buntsandstein from the Vosges. Every solution that was found to problems was in turn added to the knowledge of the Bauhütte. Thus the sculptor of the Beau Dieu at Amiens trained up a group of apprentices who then spread his techniques to Reims, Bourges, etc.

Villard de Honnecourt was a Picard builder active around 1230. He left thirty-three sheets of parchment, his techniques for building gathered from the sites of Chartres Cathedral, Laon Cathedral, Reims Cathedral, and others in Hungary. It is the only body of information which survives to give us a glimpse of how the medieval builder’s mind moved from site to site accumulating and then redeploying knowledge.

So that’s what looked like an unnecessary expense – sending the best people to remote locations to work for years on a single project. All the time, they were learning on the job. The single task of building a cathedral was not merely the completion of said edifice but also the growth of the builder’s skills, knowledge and capacity to handle more complex projects down the line.

The Bauhütte was their core platform. Deploying practitioners to sites across Europe and letting them accumulate expertise and tools to then be redeployed to new sites at lower cost and higher capability was how the lodges grew and scaled. So Strasbourg innovations would turn up at Cologne. The innovations and knowledge that the Cologne masters accumulated at Cologne would then get transferred to the Prague site and so on to Vienna.

This is especially true for all the cathedrals they had already built - these monasteries and churches have outlasted every single empire that had existed until then.

I work as an engineer in Europe, and this model has been successful for hundreds of years in Europe – it is how the greatest engineering structures of all time, such as the Gothic cathedrals of Chartres, Strasbourg, Amiens, Rouen and others were built (the best in Christendom). I will explain why I think this model, of engineers on the front line, adapting as they go, solving problems as they arise, and letting knowledge and experience build up in their collective bodies, is coming back to Europe.

It is the model Europe is currently recreating on an industrial scale in some of the most challenging and valuable real estate on Earth.

Many still cling to the Silicon Valley delusion that Europe can’t build winners, that it has great universities but then proceeds to employ most of its engineering talent in large corporations. While it’s true that the vast majority of European venture-backed companies exit for significantly less than their American counterparts (roughly a third over the last decade), that tells only half the story.

The vast majority of this is true. European venture-backed companies have only managed to generate some third of the exit value of their US peers over the last decade or so. If we compare Europe to the Silicon Valley then essentially the question is: can you build the next neolab, the next consumer app or the next vertical SaaS? The honest answer is: probably not.

What cannot be built in Silicon Valley?

But what Europeans can build? Their largest, under-digitised operational surface is a €2.5 trillion manufacturing base of 30.2M industrial workers, operated off 347,000Km of 19th century pipe to serve 40% hand-sorted waste streams. Their energy systems are presently undergoing radical re-wiring across 27 or so individual states within a single economic bloc, and are manned by the deepest STEM talent pool outside of China, which is to say that there is an enormous amount of ‘policy capital’ currently being deployed here, of the order of €43Bn for chip making alone, €200Bn for infrastructure to host AI software, and £104Bn to sort out water in the UK alone.

This is not to say that European countries don’t have plenty of talented young engineers, hoping to change the world. They do. This graph from the UK’s largest survey of graduate employers is headed “Top 10 graduate employers by number of graduate placements in the last 3 years” and it makes for bleak reading: 6,700 graduate placements in consulting and in business (over twice as many as in engineering). Of those places in engineering and industry, 2,900 graduate placements (again, the majority are in consulting over industry). (More on this elsewhere).

William of Sens didn’t send to Canterbury a treatise on French Gothic cathedral architecture. He went there and built the thing with his own tools. That was the medieval system: all knowledge was embodied in people’s bodies, not written down or applied remotely. For centuries the most successful European companies have been those that deployed their practitioners to the forward edges of their operating terrain. From this century’s viewpoint, there’s nothing but ugliness in that terrain and a blizzard of complex, fraught, regulating, highly sophisticated, operationally dense and often crumbling, brick-and-mortar (or concrete-and-asphalt) physical structures. But exactly this kind of terrain needs to be captured by forward-deployed engineers for a new generation of European industrial and processing-scale companies.

It’s already happening. The terms have simply transmuted. Forward-deployed engineering, or FDE, is the de facto model that is rapidly gaining traction in the European market and has roots going back to the work of Palantir in enterprise software in the US.

PhysicsX - a new company based in London (founded by a group of former Formula 1 engineers) has been silently building something for the last 3 years and just raised $32 million to deploy what they call "industrial-grade" physics to help make simulations run MUCH faster (up to a million times) using deep learning. Their Chairman is Robin Tuluie, the former Chief Scientist at Mercedes AMG Petronas where he helped win 4 F1 World Championships. Their CEO, Jacomo Corbo, was the Chief Race Strategist at Renault F1 and one of the co-founders of QuantumBlack, which was later acquired by McKinsey & Company. As I described in PhysicsX - Building a Bauhütte for Industrial Physics, they are building the equivalent of a Bauhütte for industrial physics and will be deploying their engineers as Forward Deployed Engineers (FDEs) across industries to help large corporations such as SpaceX.

Helsing, founded in Munich in 2021 by Torsten Reil, Gundbert Scherf, and Niklas Köhler, is the continental-scale variant. Reil previously founded NaturalMotion, pioneering physics-based animation and sold to Zynga. Scherf was a special advisor in the German Federal Ministry of Defence. Together they built a defence AI company that has raised over €1.3 billion, chaired by Spotify founder Daniel Ek. It deploys engineers and AI systems directly into the operational environment: its personnel went to Ukraine's front lines, where battlefield feedback loops shaped the HX-2 strike drone now in mass production. It acquired Grob Aircraft to build its own autonomous combat vehicle, the CA-1 Europa, at a factory in Bavaria. It is opening a production facility in Plymouth and investing €70 million in Estonia to manufacture at NATO's eastern flank. Helsing operates across Munich, London, Paris, and Tallinn the way the Parler family operated across Cologne, Prague, Freiburg, and Strasbourg.

Matta was spun out of the University of Cambridge’s Institute for Manufacturing (IFM) in 2022. On factory production lines, Matta is deploying AI-powered eyes to help check for faults and defects. The company’s co-founder, Doug Brion, is a PhD-holding expert in deep learning-enabled control. Camera-equipped on the shop floor of factories, in mere hours unsupervised AI is trained up on ten minutes’ worth of data to be able to detect near-to-100% levels of faults and defects on production lines. matta.com hires Forward Deployed Engineers to deploy its software and hardware on factories’ production lines in a short space of time – and, in time, will be dispatching hundreds of engineers all over the world to maturity on three hundred or so factories in due course. Importantly for our analysis, Matta’s job postings define work that its Forward Deployed Engineers will be doing on factories’ production lines across the entire customer journey – ‘installing & running software & hardware on factory floors: Haribo one day, aerospace the next’.

Rowden in Bristol is a defense variant of this model. Founded by Rob Harper MBE, a former Army corporal, the company builds advanced sensing, communications and electronic warfare systems to deploy into environments where normal IT just won’t function. Largely bootstrapped with under a million dollars in external funding, Rowden is a $20.4m revenue business with a 130% compounding annual growth rate over the last few years. They ranked 13 on the Sunday Times 100 Tech Startups list last year. They deliver complex integration programs for military customers around the world. Their products are hardware-manufactured in a sovereign UK supply chain. A decoy platform that confuses enemy radar travel differently than white box software.

The Company is working with leading architects & designers to deliver Not Just Walls: a new generation of houses & homes created by robotic micro-factories, using bespoke Timber Structures & well-designed Living Spaces to produce truly beautiful dwellings, that are not only environmentally sustainable but genuinely outstanding buildings. First completed for Belgian construction group Vandenbussche (Ghent), then shipped to the US and in operation on the Rival Holdings sites in the Midwest for 180 houses per year each, the full Timber Structure of a house is manufactured on site within 12 hours (40% cheaper than local construction & 75% less on-site labor than traditional building). AUAR’s “Field Ops” (a term borrowed from the video games) are the master builders who arrive at a site in a container with the lodge and within days are producing beautiful buildings – outstanding Timber framed dwellings, of the very highest quality.

Isembard - industrial manufacturing, based in London. Isembard are building out their own modular factories, running the first of these, called a "prototype" factory, out of a large warehouse in London, right next to the River Thames. The company has a team of ex-Palantir engineers, and the CEO is also ex-Rolls Royce. In a change from normal start-ups, they are building a massive industrial asset first, and then franchising out the software (MasonOS). Isembard are building every single part they make in their first factory. The team are constantly hitting all sorts of technical problems on the factory floor, and every single one of these gets fixed in MasonOS, the software that runs the factory, meaning that each subsequent factory will get better and faster at making things. The first of these prototype factories opened in January 2025 and they already have a number of franchisees running their own Isembard factories, with the heavy industrial / manufacturing oriented company Rowden being one of their customers.

Monumental are based in Amsterdam and are deploying autonomous bricklaying robots on active construction sites. Salar al Khafaji co-founded Silk which was then acquired by Palantir in 2016 to form part of the core of their enterprise AI platform. His new venture, Monumental, don’t sell the robots, they are a sub-contractor, charged by the brick laid. So far they have worked on house façades, canal retaining walls and social housing for some of the top 25 general constructors in the Netherlands. The company has grown in the last 12 months by 148% and have recently raised $25 million in funding. They have built over 100 homes, including façades, and their robot is on site where the bricks are.

Waste is a particularly exemplar domain. Recycleye, a company based near Brighton in London, have built complex robotic waste-picking systems which are capable of sorting waste into individual, ‘end-pure’ streams with a rate greater than 99% in some cases and at half the manual sorting price or less. They have clients in the UK, Germany, France and around the world. Conversely Greyparrot, also based in London, put cameras on the ends of conveyor belts in waste processing plants and are able to track over twenty five billion individual waste objects per year across twenty countries — sixty percent of all European waste management is done by companies who Greyparrot count amongst their clients. In both cases complex engineering tasks are being performed on waste processing lines all around Europe, and with both companies complex physics are being translated into reality on the factory floor by teams of forward-deployed engineers and technicians.

The approach of EthonAI in Zürich for instance combines quality inspection with so called causal analysis of the root-cause for equipment failures at customers like Siemens, Lindt and Bosch in their production lines. Cognite from Oslo ($1.6Bn market cap) implements large projects (up to hundreds of implementations) and has ‘fellows’ embedded into oil platforms and power plants around the world for TotalEnergies, Saudi Aramco and BP. As a European provider of industrial AI, GridBeyond is also embedding in multiple sites in UK, Ireland and US and deal with 27 or so different EU and national regulatory environments. Daedalus in Karlsruhe was founded by OpenAI’s first Engineering hire and are themselves a manufacturing company with a focus on making CNC machining easier with AI for customers in aerospace and semiconductor industries.

We keep returning to a single pattern: send engineering talent to a particular geographic region and let an infrastructure compound of their actions there.

The technologist solves complex problems in Silicon Valley by reducing to code and removing the human from the loop in order to scale. The larger industrial opportunity in Europe is the reverse: to deploy a talented Engineer to operate and solve problems at the operational surface of the industrial complex, to compound value incrementally at the point where the physical and digital meet.

The medieval master builder, a most wonderful practitioner, was an expert in his field – a builder. Working in the field, or on-site as we say today, this most wonderful practitioner had a vast amount of knowledge at his fingertips. This practitioner worked on a site and another, and as he worked, his knowledge of buildings was constantly expanded. The things he learned on one building, were applied to the next. And so, this wonderful practitioner’s knowledge was constantly increasing. It was as if he were reading a book on buildings, and then, suddenly, he was the book. He was the accumulated knowledge of all buildings, past and present. The practices he developed, in building a church, were used to build another. He would go from building to building, and as he worked on each one, his knowledge would be increased. This knowledge would be then be applied to another building, and so on. And so, this wonderful practitioner’s knowledge of buildings, increased with each new building. The skills he acquired in building a church, were then used to build another. His knowledge of buildings was so great, that he was able to build not just one church, but many. And the value of what he created, lasted for eight centuries.

The forward-deployed engineers in Europe building the industrial future of that continent are his direct descendants. They know that the big gap in performance between the currently available and developing AI software and the rather tough operational realities can only be bridged by some human(s) physically and operationally sitting at the point of contact.

The operational surface area for these industrious endeavors amounts to some $3 trillion of manufacturing output and $12 trillion of construction across the developed world – the largest, under-digitised operational surface area in the space.

The cathedrals are still standing.

Tesla's cars are delivered to customers with huge gaps in between the panel elements – big enough to stick a finger into. And while their market capitalization is at an all-time high of around $800 billion, BMW is «worth» only about $70 billion.

Meanwhile, SpaceX is rocketing to the Mars and beyond by blowing up their rockets in public and iterating at an incredible pace, outdoing Boeing who can’t even certify a single system in that same timeframe.

Their lack of adherence to ‘norms of European engineering’ and the fact that things worked magnificently was quite illuminating for me. A lesson that in engineering, speed over perfection every time.

As for us, at KAIKAKU, we managed to ship a 7000+ components robot with only 3 people in less than 2 years. We managed to file 3 patents and 25 design rights for our restaurant operating system made of a full restaurant computerized with a real-time computer vision dashboard.

Perfectionism is also reigning in institutional terms. Last week the UK government announced it is to abolish the independence of the UK Space Agency.

How Universities Train Us to Move Like Molasses

The problem starts at University. A comparative review of engineering degree programs found that European University engineering degree programs concentrated on ‘theoretical foundations first’, whereas the primary focus of US engineering degree programs are ‘on the job learning’ via ‘tens of mini projects’.

In the US you are often trained to complete a project and then fix any errors. In Europe you are expected to get it right the first time.

I have myself been through Imperial College London’s engineering program, one of the top in Europe. We were certified to kill by passing a series of rigid exams, each testing a specific range of knowledge with no room for error and a single shot at each of them. Meantime, in MIT, many students were failing fast in a continuous stream of “assignments”, eventually normalizing to the speed of iteration to fixing errors in engineering designs that had gone wrong in their early mini-projects in just days, as opposed to ours spending years perfecting a single implement, erring on the side of excess caution in our trying to design around all the “what-ifs” that seldom if ever would actually occur in practice.

We're not born this way. We're trained this way.

EUR-ACE (which accredits engineering programs across Europe) emphasizes input-focused criteria and curriculum content requirements. In contrast, ABET (which accredits engineering programs across the United States) looks for programs to ensure students are able to meet Student Learning Outcomes and continue to improve their learning once they have graduated (Taylor & Francis, 2013).

My Cofounder and CTO commented thoroughly on the matter for his own blog: What the Education System is Getting Wrong about Engineering.

Jolla: A €50 Million Cautionary Tale

Ex-Nokia engineers from Helsinki create Sailfish OS for smartphones. Jolla the company raised €50 million in funding to create revolutionary OS for mobile phones. By 2024 however the OS and the company will be dead.

Jolla’s first phone ran on a dual-core processor while others used quad-core processors as early as a year prior. The screen ran at 540x960 while others already were at 1080p. Jolla spent years to perfect gesture-based navigation which in the end didn’t get much appreciation from users or press (Wikipedia, 2024). The phone was described by TechCrunch as “plodding” and with “interface lag” (TechCrunch, 2014).

So European companies with technically excellent hardware often fail commercially because it’s released too late.

The sense of improving edge cases for rarely occurring situations ends up adding too much complexity in the end for European engineers, workiing in hardware.

Europeans Must Break Rules

When European companies abandon perfectionism, they dominate globally.

Spend 12 months in development before cash-burn must get your hard-ware venture to Series A? This ‘handicap’ afflicts hardware startups originating in Germany above all, but also Belgium and the UK, as opposed to fluid software companies hailing from the U.S. (McKinsey & Company, 2024).

Supercell’s Helsinki-based mobile game developers produce prototypes in weeks, shoot down 99% of their games, and launch the remainder with bare-bones graphics and gameplay. They manage to reach $2.3 billion in revenue this year with those ‘unfinished’ games.

The author of this post previously wrote about how Revolut started off as a prepaid card for currency exchange, and has grown to a multinational company with $33B in value in just 7 years after 100+ rapid feature updates.

TransferWise (now Wise), the online money transfer service for personal and business customers, started out with manual bank transfers and a basic website to open and manage currency accounts online. It was a long way from the full range of services for private and business customers that the company offers today, including accounts in over 40 currencies and real-time foreign exchange transfers.

BioNTech, a pioneer in European pharmaceuticals, only followed a very different path. The COVID-19 vaccine was designed in two days and could be shipped for trials a few months later.

European hardware companies are especially disadvantaged when they try to innovate like Americans.

The 12-Month Handicap

Hofstede’s analysis of culture reveals an interesting insight into how the Europeans’ psyche functions in relation to uncertainty avoidance: Germany scores 65 on the scale, Belgium an enormous 94 and the US a mere 46 (IMF, 2012). European hardware companies also take significantly longer to travel the road from seed to Series A funding compared to their US counterparts (McKinsey, 2024).

Twelve. Months.

By Silicon Valley standards, this translates to three product iterations, three rounds of customer feedback, and three major pivots before a European hardware startup can finally launch version 1.0 of their product to be canibalized by their American competitor.

The data actually supports our hypothesis of additional scrutiny. European hardware startups validate their ideas 4.2 times on average before they launch a product. In contrast to their American counterparts (2.8), however. And as we have mentioned before, each of these validation cycles typically take the European founder 4-6 weeks on average to get feedback (as opposed to 2-3 weeks for their American counterparts).

As for the feedback from potential customers, it is well known that in the U.S. startups can get such feedback within 3 months while in Europe it takes a whole 9 months on average (Taylor & Francis, 2023). So by the time a Munich-based founder finally gets to talk to his or her first customers, the same startup in Silicon Valley has already gone through two iterations.

The data also reveals why European startups fail to reach the same levels of successful outcomes as their American hardware counterparts – a meager 27.5% vs. 37.5% according to McKinsey (2024).

Why? Because perfectionism doesn't scale.

This market also needs to be served with products designed for 28 countries with different languages, local regulations and cultures. This is exactly the opposite of what American hardware startups have to deal with: 330 million people speaking one language and using one currency.

Meanwhile, imperfection is flushed out to 330 million people of one language speaking, and with one currency only, in the U.S. alone.

MVP: The Concept Europe Needs to Grasp

These ideas were even more revolutionary, developed by one of the founders of Silicon Valley, Steve Blank (a professor at Stanford University at the time) and subsequently validated by another founder, Eric Ries, author of the bestseller The Lean Startup.

The most common mistake among European founders is to misunderstand the term MVP. Many people associate it with barely functional prototypes. This is also how most European founders look at an MVP. In reality, MVP stands for the smallest thing that can be used to validate the biggest assumption about your product. As an example, Airbnb started out with a website and a few air mattresses. The core product that every business later developed wasn’t there yet. When Dropbox was first launched, all users saw was a 3-minute video. The actual cloud storage followed later.

An example of such crowdsourcing on Kickstarter and Indiegogo in Europe are campaigns launched with near-finished products, offered for pre-order to potential customers. Such an approach to crowdfunding is completely opposite to that of their American counterparts (Crowdfunding.io, 2024).

The Quality Paradox

All that perfectionism? It doesn't even deliver quality.

A very European problem – European product recalls are at an all-time high (Sedgwick, 2024) of 14,484 total (an increase of 32.9% over previous year’s numbers for consumer products). Your long attempts at quality guarantee will only get you to market late.

R&D = Research & Development. The number of R&D investments in Europe’s high-margin industries is dwindling. Instead, most of the global R&D investments in high-margin sectors are made by American companies – a staggering 42.3% as opposed to 18.7% by European companies. What’s more, the industries that European companies are investing in are mostly ‘sunset’ industries, like the automotive sector.

What Actually Works

Their approach to building revolutionary aircraft in record time has been passed down through generations of engineers, and includes the XP-80 jet fighter built in 143 days, the U-2, SR-71 Blackbird, and F-117 Nighthawk, all of which were described as “impossible” to build at the time by Lockheed Martin.

Johnson's 14 rules read like heresy to European engineers:

• Rule 3: Work in small, strong teams. These would represent 10-25% of the normal size of an aerospace program.

• 5. Limit written reports to a single monthly cost report.

• Rule 7: Trust them to get it right. If they make a mistake, Kill Them. But don’t give themtime to do it.

• Rule 13: Any member’s participation in a project must be strictly on a “must know” basis. There is no tourism on a Skunk Works project!

These programs have an average of 500+ aircraft delivered to customers with a budget 30% below the average for a conventional development program and delivered 50% faster than normal. The biggest secret to Lockheed’s continued success is there is only one layer of management between the engineer and the product.

SpaceX: The Algorithm That Breaks Every European Rule

The famous 5 step algorithm by Elon Musk for producing innovative products while preventing over-engineering by European standards (Valispace, 2024) in the following order:

1. 1. Make less dumb the requirements as to what the product is to do

2. The requirements are definitely dumb; it's particularly dangerous if an intelligent person gave them to you. 2. Delete a part or a process

3. Note: If you remove something, make sure you add back 10% of what you just removed! 3. Simplify or Optimize – Once you started deleting, simplify or optimize whatever is left. The most common error of a smart engineer is to optimize something that shouldn’t exist. 4. Accelerate cycle time – But not until you have simplified. 5. Automate

4. The last step, not the first.

SpaceX blow up their Starships from time to time. Each explosion teaches them so much more than a decade of complex and expensive simulations by Boeing would. They have completed foundational work for crewed flights in 10 years and launched over 400 flights so far. They can launch rockets and even land them on a boat while Boeing struggles to land even a few contracts.

Tesla: 27 Changes Per Day While Germans Plan

Tesla for example makes 27 hardware and software changes to its Model S every day, that is 5,400 times a year (Valispace, 2024). We are talking about hardware changes to cars that are rolling off the production line.

Engineering is Poetry at BMW’s 9,000 man strong Research & Innovation Centre in Munich, where 1 mm fits inside another, superior raw materials are selected and a 5 year product cycle is sufficient to get every single detail just right.

For now, most German automotive executives still reject the obviously unacceptable panel gaps on new Teslas. Meanwhile, their manufacturers are selling only about 320,000 cars at an average of merely 4% profit margin (Bloomberg 2023/2024). In 2023, Tesla delivered an astonishing 1.81 million vehicles to customers all over the world.

The Path Forward

At KAIKAKU we had a small team of three engineers, and so we didn’t have the luxury of perfectionism, hence we opted for the ‘American’ way of development. Our first robot was held together with zip ties and dreams. The 547th version of the robot now serves thousands of meals every month.

Many founders are unsure if they should invest so much time in the validation of their ideas or if they should just start to build and improve their product along the way. The only thing that is for sure is that European founders will be undercut by their American and Chinese competitors while they are trying to perfect all the details. The numbers for advanced manufacturing from the European Commission (2024) support this: Only 1 in €10 of global investment is destined for European companies (errors in the original).

Want to win? Here's the playbook:

Kill your validation cycles. Ship at 70% completion. The customers will fill in the remaining 30% and tell you all about it.

So ignore your professors (unless they can launch a product). Europeans tend to be academics rather than practitioners so have optimized for publications rather than real products that customers want.

Let misfits join your team, e.g. that engineer who got fired from a previous job because he was shipping too fast to the customers. And yes, children who were always breaking things as a toddler playing with their parents’ appliances are great to hire.

Action > Thoughts. Their thoughts will not get them into trouble but their inaction sure will. Hard stuff require discomfort.

And no future-proofing. You make the best plans, but the products you are developing will be obsolete (or you run out of money) when your competitors have been executing on their products for the market in the meantime.

European Commission forecasts that only 1 in €10 of global advanced manufacturing investment is going to end up in the accounts of European firms (European Commission, 2024).

However, there is a complete change of culture required. Anything less than a revolution is unacceptable.

Which is exactly why they work.

The Uncomfortable Truth

All these highly skilled and very talented engineers are being imprisoned in very suboptimal engineering systems. Every month, I get to see a bunch of very talented engineers in London, Berlin, and Munich struggling with PowerPoint presentations to develop stuff that could change the future. Meanwhile, someone in a garage in Palo Alto is fixing real problems.

So the reality is that the market is Darwinistic. Things that do not function are chucked in the bin. Documentation is not read by users and financiers invest in things that already exist and are of immediate value to them.

Perfect is the enemy of good. In hardware, perfect is the enemy of everything.

No future-proofing.

The Black Death killed 100 million people in the 14th century. It’s happening again. Just like back then, none of us seem to know why.

Another dinner.

Another room full of men in overpriced jackets and underdeveloped convictions, debating how they want to die.

“Happy,” answered one.

“Without regrets,” mumbled another. (He definitely regrets that £5,000 jacket.)

Then they turned to me.

How do I want to die?

I don’t.

Tech Bro's Longevity Obsession

I know “not dying” has become a tech bro punchline. If past generations had given up so easily, we’d still be in horse-drawn coaches, not hypersonic jets.

Most of the tech world chases gene editing, biohacking, and blood transfusions. But while everyone else is looking for answers in labs, one of the most respected longitudinal studies in the world, Harvard’s 75-year Study of Adult Development, quietly found something else.

And let’s be honest. The current longevity game is mostly smoke.

• Gene editing? Risky, still stuck in the lab.

• Biohacking? Mostly almond butter and definitely botox.

• Blood-swapping with your teenage son? Weird flex, Bryan.

We spend billions chasing longevity, while malaria, which kills a child every minute (UNICHEF, 2024) gets just $4 billion globally, mostly from overstretched public funds.

A £2 net saves a life (AMF, 2025). A £300 supplement makes your pee more expensive.

Genetics only explain 20–30% of your lifespan (NIH, 2016).

So what does?

Harvard’s longest-running study followed individuals from Boston’s working-class neighbourhoods and Ivy League campuses (including president John F. Kennedy) to uncover what truly shapes long-term health and fulfilment. Is it wealth, status, or access to elite healthcare?

It Is Human Connection.

Loneliness chronically activates the HPA axis and sympathetic nervous system, flooding the body with cortisol. Over time, this disrupts immune regulation, fuels systemic inflammation, and suppresses both T-cell and natural killer (NK) cell activity, raising the risk of infection, cardiovascular disease, and premature death (Cole et al., 2007; Hawkley & Cacioppo, 2010). It's as detrimental as alcohol or smoking.

This is the new Black Death.

Over the next decades, we will be together alone. We will still inhabit shared spaces, yet experience them as emotionally vacant. Proximity will no longer guarantee presence.

AI will not only displace labour, it will erode social interaction. Increasingly, we will interact with machines rather than people, because our infrastructure will allow nothing else

During the Black Death, people died without understanding how or why. That was medical helplessness - a visible crisis with invisible causes. Today, we face the opposite: our health crisis is invisible, but the cause is everywhere. Chronic loneliness, social isolation, and emotional detachment are eroding our bodies from within.

Yet we fail to see it. That’s psychosocial blindness: the inability to recognise how disconnection is quietly killing us, and we’ve normalised it.

Remember: The Black Death didn’t end because it was cured.

It ended because it ran out of hosts.

When I started building, I understood the risk:

A future of automation, ambient despair, and hollowed-out communities too numb to know they’re grieving. Every time we automate or outsource connection through AI, touchscreens, or soulless UX, we chip away at the signals our brains evolved to crave.

I’m not against (embodied) AI. Let it handle the soul-crushing tasks. Let it even run our politics!

But not this. Not the part where someone looks you in the eye and makes you feel like you exist.

So I Build The Future Differently.

That’s why in our robotics-powered restaurant a real person greets you at the door.

That was always part of my vision. I believe human connection matters even in the most high-tech environment. But belief is one thing. What surprised me was how many people told us, explicitly, that it mattered to them too.

And what really shocked me?

How many of our regulars turned out to be the elderly. Were they not scared by the robotics? The touchscreens?

So we asked and the answer stayed the same:

“Here, it still feels like someone cares about me.”

(81-year-old Anne added: “Also, your bowls are better than sex.”)

They’re right. This is not just good manners or a “nice touch.” It’s neuroscience.

fMRI research from Elizabeth Redcay (2010) shows that live conversation lights up far more of your brain than passive video. It engages emotion, attention, memory, and decision-making.

But connection alone isn’t enough. You can’t out-socialise crap food or bad health.

I care about connection. But I’m also obsessed with everything else that keeps a body alive. Sleep, light, micronutrients, air quality, muscle mass, psychological load, hormonal rhythms, protein timing, plastics in your bloodstream, and yes, food.

There are extraordinary projects shaping this future:

Neko Health is one of them, delivering next-generation healthcare experiences at scale. Their human operators and GPs make you feel like their only client.

(I witnessed their obsession over experience and connections firsthand from Hjalmar and Shak.)

Another figure I admire is George who's building ZOE: pushing the boundaries of personalised health to feel like a community rather than just a blood sugar test.

Food just happens to be one variable I believe to have an unfair advantage at through what I'm building. One of our PhDs is literally decoding taste, so you can eat what your body needs before you even know you need it and actually love how it tastes.

Here’s what no one tells you: if you want to not die, you don’t need optimism. You need standards. Unreasonable ones.

For me I'm sacrificing all my standards temporarily to solve this problem. I’m building a cult (aka a startup), which means I already get zero sleep, zero sunlight, and probably inhale a worrying amount of 3D-printing fumes.

Honestly? What keeps me is the people. I’d rather share lunch with cracked engineers and obsessive PhDs than sit through another dinner party pretending to care how someone wants to die.

Longevity, as per research, isn’t about avoiding death. It’s about refusing a life not worth living.

So yeah, I’ll take your injections, your supplements, your son’s blood.

But I’d rather take a shared table, a warm meal, and someone who remembers my name.

Let’s hack death - not with AI, but with each other.

Most of these problems affect not only “startups” but a broad spectrum of individuals and groups including inventors, researchers, scientists and academics as well as industry innovators.

Here’s the reality:

• In the US the provisional patent application can be filed for a low fee of $75 or so to fix a priority date. This is a low cost means by which to test the market, gather feedback, complete additional design development, go for funding and work with others for up to 12 months whilst preserving intellectual property.

• In China the fast-track utility model patents are granted within less than a year and are even partly subsidized by the government. Utility model patents protect the technical implementation of an invention and are granted faster than inventions patents.

• In Europe, mere public discussion of an invention by an innovator can lead to patent rights being lost forever. In addition, high up-front filing fees of €2,000 to €6,000 and the lack of any form of flexibility or “grace period” of any kind for second changer innovations mean that very often such high risks have to be avoided at all costs.

That’s why, for example, China’s fast and flexible patent system for technical innovations – including drones and robotics (DJI, Unitree Robotics etc.) – has enabled it to become a world leader.

What does this mean in practice?

• Academics & Researchers: European universities are among the best research institutions in the world, but the scientists often have to decide between publishing their work and applying for a patent. If the research has been made public before filing a patent it can no longer be patented. As a result scientists are put under pressure to delay sharing their knowledge for scientific progress.

• Small Inventors: Independent inventors are mostly working on their own and have very little budget to spend on enforcing intellectual property. The possibility to file a low cost provisional patent in the US (as described above) and to secure their rights whilst looking for funding makes it easier for independent inventors to develop and bring innovations to the market in the US than in Europe. The risk of losing out to bigger players is too high to even consider the option of filing full patent applications at big cost.

• Startups & Industry Innovators: Many emerging technologies are subject to significant iteration and are tested on many different occasions. A number of businesses require the inventor to demo their invention to potential investors, customers or partners as part of their business development activities. If the public disclosure of their work puts their patent application at risk, startups and industry innovators will be severely impaired.

All patent systems are geography-bound filing regimes. This means that if an invention is made in a researcher’s laboratory in Europe then that researcher can have no workarounds and must therefore consider moving to somewhere with a patent system that is more favorable to innovators.

Comparing Patent Systems

A Founder’s Experience: The Cost of a Demo Day

I actually had to rush a £6,000 patent filing so that I could legaly present at a demo day to investors. That was really scary as I had risked destroying our patent position by pitching to investors.

A few months later that product had evolved in ways we could not have anticipated. The £6,000 patent was now worthless. We could file another £6,000 patent a few months later.

Total cost of £12,000 plus numerous hours.

If we had set up shop in the US then I would have filed a $75 provisional patent application and we could have iterated freely until 12 months later when I would have filed the full patent application.

This goes to the heart of a major problem facing the world of innovation today.

How to Fix It: Three Simple Reforms

These are not radical suggestions. These already exist as models for innovation in the US and in China. All Europe’s national IP offices and the EPO’s search and examination activities need to be set up to deliver these ends without the need for subsidy or tax break.

1. Introduce a Provisional Patent System

• Introduce a Provisional Patent System

• allow simple and cheap (£100) provisional patent filings which can be developed into full patents within 12 months.

• Allow 12 months of flexibility.

Impact: More innovation, less financial risk—empowering startups, researchers, and independent inventors alike.

2. Implement a 6-Month Grace Period

• A 6-Month “Grace Period” Implementation

• Covered entities: academic publications, investor presentations, crowdfunding platforms, early-stage products and services about to be published, presented or offered for sale on the web.

Impact: Encourages open collaboration, removes the fear of killing IP by simply talking about an idea.

3. Fast-Track Patents

• Lower initial search and examination fees for startups.

• Expedited review for first-time inventors (e.g. as in the UKIPO’s Green Channel).

Impact: Startups get faster, cheaper protection—helping them attract investors and secure commercial deals.

The Time for Change is Now

So let’s get moving on improving our patent system to level the playing field with regions such as the US and China to enable innovation to flourish in Europe.

The solutions are simple: no need for subsidies or handouts – just a patent system that works for startups, researchers and inventors alike.

Let us hope policymakers recognize these straightforward solutions to unlock the current patent straightjacket before it is too late. Innovation will not wait!

Imagine a silent pact so powerful it can forge empires or crush dreams. In families worldwide, particularly immigrant ones, an invisible contract often binds children to their parents' sacrifices and expectations.

Break it, and you're ungrateful. Honour it, and you might suffocate your true potential. But those who dare to shatter these unspoken expectations? They often emerge as the most formidable founders in the startup world, driven by a relentless urge to prove their worth on their own terms.

The Anatomy of a Covert Contract

A covert contract, a concept popularised by psychologist Robert Glover, is an unspoken agreement where one party silently expects something in return without explicitly communicating these expectations. It's "covert" because it's never openly discussed, and "invisible" because it operates beneath the surface of conscious interactions.

While present in many family dynamics, in immigrant families, it often intensifies to:

We sacrificed everything. You owe us prestige.

For children, this invisible pact creates:

1. Suffocating pressure to pursue "prestigious" careers.

2. Guilt when choosing passion over stability.

3. Constant reminders of parental sacrifice.

4. Expectation to honour roots while conquering a new world.

Breaking this contract feels like betrayal. It's not just saying no to business school or a prestigious job offer — it's shattering your parents' dreams.

But for the best founders, it's not about escaping the debt. It's about proving that unorthodox path yields greater returns than any "safe" career ever could. It becomes rocket fuel.

They're not running from obligation; they're sprinting towards success so massive it will justify every risk, every parental tear, every sleepless night.

This creates extreme urgency. As parents age, the clock ticks louder. Each year demands bigger wins to validate their divergent path and compensate for the immediate prestige and stability they've foregone for their families.

Cultural Pressure Cookers

The intensity of covert contracts varies across cultures, ranging from subtle expectations to overwhelming pressure. East Asian cultures, deeply influenced by Confucian values of filial piety, often exemplify the extreme end of this spectrum:

• Korea: Quitting a stable job is family disgrace.

• Japan: "Resignation agencies" thrive to avoid job-leaving shame.

• China: Children's achievements are family social currency in a perpetual status game.

Western cultures aren't immune, though the pressure often manifests more subtly as expectations of loyalty—to family, mentors, employers, and others who've invested in one's success.

This cultural crucible forges founders with distinct characteristics:

1. Guilt-Forged Resilience: Each setback isn't merely a business challenge, but a test of life choices and familial loyalty.

2. Hyperdrive Time Sensitivity: An acute awareness of time pushes these founders to operate at warp speed.

3. Chip-on-Shoulder Motivation: The need to prove their worth fuels an insatiable hunger for success.

4. Cultural Arbitrage Skills: They leverage their unique position between old-world values and new market opportunities.

The result? Ambition that is matched only by their resilience — a potent combination in the high-stakes world of entrepreneurship.

For VCs, these "pact-breakers" are invaluable. Their drive often surpasses that of founders from less pressurised backgrounds. They're not just building companies; they're on a mission to redefine success and validate their life choices on a grand scale.

A Message to My Parents, and to Those Who Understand

This essay isn't about showcasing success. It's about breaking the silence on the complex emotions and drives that shape many of us who've grown up under the weight of parental sacrifice and expectation.

Growing up in a Chinese restaurant family in Austria, I felt the unspoken pressure daily. Every good grade was a silent nod to the long hours my parents worked. Their sacrifice demanded excellence, and I delivered—not just for myself, but for them.

This drive led me to the UK, to a top university, to prestigious internships. I chased the white-collar jobs that would make my parents proud. The satisfaction in their eyes when I wore a suit, when I mentioned the corporate names, was palpable. For a while, it felt like enough.

But many of us reach a point where we question if fulfilling our parents' dreams truly honours their sacrifice. Is following a prescribed path really the best way to repay their investment in us?

For me, and for many others I know, the answer was no. Leaving the security of a conform life wasn't about rejection—it was about taking the drive they instilled in us and applying it on our own terms.

This decision comes with immense pressure. Every day becomes a test of our choice. We work relentlessly, not just for success, but to prove that our path is valid, that we haven't squandered the opportunities they gave us.

To my parents, and to all the other parents who've sacrificed: We haven't forgotten. We're not running from your expectations, but striving to exceed them in ways you might not have imagined. Our drive, our work ethic, our ambition—these are your gifts to us. We're using them to forge new paths, to create value in new ways.

The urgency we feel, the relentless push for success — it's all rooted in a deep desire to make you proud, to ensure your sacrifices weren't in vain. We may have broken the traditional pact, but we're working tirelessly to repay it tenfold.

This journey isn't unique to me. It's a silent struggle shared by many children of immigrants, of small business owners, of parents who dreamed of better lives for their kids. 

We carry your hopes with us, always.

• Top London hardware engineer graduates: £30,000-£50,000

• Silicon Valley equivalent: $150,000+

The reality for most graduates is even grimmer:

• £25,000 starting salaries at traditional engineering firms

• Exodus to consulting or finance just because it's compensated better

Meanwhile computer science graduates land lucrative jobs in big tech or quant trading, often starting at £100,000+

Examples of wasted potential:

• Sarah: Built a fusion reactor at 16. Now? Debugging fintech payment systems.

• James: 3D-printed prosthetic limbs for A-levels. Today? Writing credit risk reports.

• Alex: Developed AI drone swarms for disaster relief at 18. Graduated with top honours from Imperial. His job? Tweaking a single button's ergonomics on home appliances.

These aren't outliers. They're a generation of engineering prodigies whose talents are being squandered.

This isn't just wage disparity. It's misallocation of human capital on a national scale.

As a hardware founder in London, I've witnessed this firsthand. We have the talent for groundbreaking innovation, but lack the means to realise it.

Root Causes:

1. Geographical Constraints: Unlike lucrative software jobs, hardware engineering demands physical presence.

2. Venture Capital: European VCs, mostly bullish on fintech and SaaS, remain wary of hardware. Result? A feedback loop of underinvestment and missed opportunities.

3. Industrial Stagnation: Traditional engineering firms fail to innovate in talent strategies and match compensation, accelerating brain drain.

Consequences:

1. Innovation Stagnation: We're not just losing salary differences; we're missing out on the next ARM or Tesla.

2. Economic Ripple Effects: One successful hardware company can spawn dozens of ancillary businesses. We're losing these compounding effects.

3. National Security Implications: In an era where technological edge equals geopolitical power, can we afford to let our best hardware talent languish?

4. Brain Drain Acceleration: We risk losing our top talent permanently to overseas markets.

Debunking Common Myths

"London's lower living costs justify lower salaries."

False. London is around the same as NYC and more expensive than most parts of California and definitely Texas. This also ignores:

• Wealth Creation and Ecosystem Acceleration: High salaries and successful exits compound dramatically over time, that's why the US has so much more VC and angel capital.

• Talent attraction: Top jobs draw global talent. Example: Google's entry into London with competitive salaries reshaped the entire tech ecosystem.

"UK's small market limits growth."

Outdated thinking. Consider:

• Dyson: From a Wiltshire barn to a global technology powerhouse, now innovating in Singapore and Malaysia.

• Ocado: Online grocer turned global automation technology provider, with robotics solutions deployed across Europe and North America.

• ARM: Powering 95% of smartphones globally.

"Hardware is riskier than software."

No longer true:

• Development speed: 3D prints and PCB prototypes now available in 24 hours, rivalling software iteration speeds.

• Moat strength: Apple's hardware-software ecosystem is far more defensible than most pure software plays.

• Massive hardware exits:

  • ARM: Sold to SoftBank for $32B in 2016, now worth $140B+.

  • CSR (Cambridge Silicon Radio): Acquired by Qualcomm for $2.5B in 2015.

  • Dyson: While not an exit, it's valued at over £20B as of 2023.

The Arbitrage Play:

It isn't about costs. It's about ambition.

While software talent flows freely globally, ambitious UK hardware startups can exclusively tap into a world-class, locally-bound talent pool.

1. The Software Brain Drain:

   • US tech giants easily poach UK software talent

   • Remote work erases geographical boundaries

   • Result: Constant outflow of top software engineers

2. The Hardware Opportunity:

   • Physical presence matters - can't build rockets remotely

   • UK hardware talent largely untapped by global competition

   • Build something ambitious, attract local engineering superstars

3. The Talent Trap:

   • Brilliant minds wasting away in soul-crushing corporate jobs

   • Your future "10x engineer" is someone else's bored employee

4. The Hardware Advantage:

   • Forget software. Hardware is the new frontier.

   • Build the next ARM or Dyson, not another fintech app

   • Leverage UK's world-class research institutions

5. Why Now:

   • Incumbents are unambitious, startups are few (for now)

   • Top-tier VCs awakening to UK hardware potential

   • First movers will have pick of the talent pool

The Window is Closing

This arbitrage won't last forever. As you read this, others are waking up to the opportunity. The first movers will reap the rewards. The followers will wonder why they didn't see it sooner.

The Hardware Revolution Starts Now

Wake up, UK. Our engineering talent is our nuclear fusion. Ignite it or lose the future.

VCs:

Your next unicorn isn't code. It's cobalt and circuits. Back the tangible.

Founders:

Stop fleeing to the US. London can be the hardware capital of the world. We have the talent. We have the creativity. What we need is your audacity.

Engineers:

Your brain's worth billions. Build empires, not apps.

While the world obsesses over the next GPT wrapper, we'll forge the next industrial revolution.

This isn't a pipe dream. It's an imperative.

UK, it's time to build.