But price first proves a shift in risk appetite. It doesn’t prove the bins are empty.

Whether grain markets have crossed from “risk repricing” into “structural scarcity” can’t be read off the chart. It requires tracing a longer causal chain:

Right now the early links in that chain have tightened noticeably. The later links still lack consistent confirmation.

Three Crops

Bulk agriculture gets traded as a single macro basket, but wheat, corn, and soybeans sit at different points on stock levels, substitution elasticity, and pricing tension. Using USDA’s August 2026/27 WASDE balance sheet:

• Global wheat production is 819.30 million tonnes against consumption of 826.27 million tonnes, a production deficit of about 6.97 million tonnes, with ending stocks at 273.25 million tonnes, up 0.41 million tonnes from July.

• Global corn production is 1,298.88 million tonnes against consumption of 1,323.05 million tonnes, a deficit of about 24.17 million tonnes, with ending stocks at 274.66 million tonnes, down 0.60 million tonnes from July and down about 24.17 million tonnes year over year.

• Global soybean production is 442.25 million tonnes against consumption of 442.02 million tonnes, a small surplus of 0.23 million tonnes, ending stocks at 124.21 million tonnes, up 0.19 million tonnes from July.

Global stocks-to-use ratios:

• wheat 33.1%,

• corn 20.8%,

• soybeans 28.1%.

Strip out China and the picture changes:

• wheat 22.6%,

• corn 11.0%,

• soybeans 26.1%.

Narrow further to major exporter stocks-to-use: the actual cushion available to the global trading system, and it is:

• wheat 12.2%,

• corn 9.1%,

• soybeans 23.9%.

Note: Two different stock metrics are used here and shouldn’t be confused. USDA’s official stocks-to-use ratio is ending stocks divided by domestic total use, as USDA standardly defines it. A second, derived metric, trade-available stock ratio, is calculated in this piece as ending stocks divided by domestic total use plus exports, meant to measure the thickness of the cushion actually available to international trade.

The corn numbers break down like this. Global corn stocks-to-use of 20.8% is 274.66 divided by 1,323.05, USDA’s official figure. Ex-China stocks-to-use of 11.0% is 109.53 divided by 999.05, still USDA’s official basis — ending stocks over domestic total use. The trade-available ratio of 9.1% is 109.53 divided by the sum of 999.05 and 210.46, a derived figure measuring the buffer of ex-China global corn ending stocks against domestic use plus exports.

Three conclusions follow.

First, global grain isn’t entering shortage everywhere at once, at the same intensity. Wheat and corn production run below consumption; soybeans are roughly balanced. The real stock contraction is concentrated in corn. US 2026/27 corn ending stocks were cut 137 million bushels from the July estimate to 1.653 billion bushels. US wheat ending stocks were cut 5 million bushels to 717 million bushels, down 22% year over year. US soybean ending stocks rose 10 million bushels to 320 million bushels.

Second, aggregate global stock totals create false comfort. China holds roughly 60% of global corn ending stocks, and that grain doesn’t automatically flow into international trade. Strip China out and global corn stocks-to-use falls from 20.8% to 11.0%. Push further to the trade-available ratio, ending stocks over domestic use plus exports, and the cushion drops to about 9.1%. That’s closer to what importing countries actually face than “the world still holds 275 million tonnes of corn.”

Third, the same price rally can be trading entirely different risks. Wheat is trading whether grain can leave port on schedule. Corn is trading whether final yield covers elevated consumption. Soybeans are trading the standoff between low US stocks and South American replacement supply. Lumping all three under “food shortage” erases the relative value that actually matters.

• https://www.usda.gov/oce/commodity/wasde/wasde0826.pdf

Subscribe now

Read more

Bessent is not trying to reverse the long-term trend in Treasury yields. He is buying time.

Three Constraints

Three facts bound everything Bessent has done.

Subscribe now

Read more

By mid-August 2026, the 30-year Treasury yield hit its highest level since 2007. The same week, U.S. federal debt crossed $40 trillion for the first time. That timing was coincidental. The bigger driver was global.

• Japan’s 10-year yield touched 2.945% on August 18, the highest since September 1996. Japan’s 30-year hit 4.155%.

• Germany’s 10-year rose to 3.2%, the highest since 2011.

• France’s 10-year hit its highest since 2008.

Every major sovereign bond market moved in the same direction at the same time. This wasn’t a Treasury-specific story. It was a repricing of long-duration government debt across developed economies.

Three factors drove it.

First, fiscal anxiety: markets priced in the scale of U.S. deficits and debt issuance, plus sticky inflation expectations.

Second, AI financing crowded into credit markets. Global AI infrastructure investment exceeded $1 trillion in 2026. The Bank of England’s July financial stability report found that five AI hyperscalers accounted for over 15% of U.S. investment-grade bond issuance as of early May. The BoE added that there’s no clear evidence yet of crowding out other borrowers.

Third, oil: the U.S.-Iran ceasefire never resolved Strait of Hormuz shipping risk. Crude went back above $90 a barrel, and inflation fears reasserted themselves.

Foreign demand also thinned. June TIC data showed foreign holdings of U.S. Treasuries down about $72.1 billion from May, the steepest monthly drop since March 2026. Japan, the UK, and mainland China all sold. Japan alone cut $26.4 billion.

The Yen Intervention: A Coordinated Operation and a Deliberate Leak

Behind the bond story sat a currency crisis. USD/JPY hit 163.99 on July 23, a level not seen since December 1986. The mechanics are old and structural: a 250- to 275-basis-point rate gap between the U.S. and Japan keeps the yen carry trade alive; borrow yen, buy dollars. Japan imports almost all its energy, so a weak yen amplifies the oil shock directly. Japanese government debt sits above 240% of GDP, which rules out aggressive rate hikes as a fix. The Bank of Japan had already raised rates to 1.00% in June, a 31-year high, and it wasn’t enough.

Tokyo intervened unilaterally in April and May. It didn’t hold. On July 30, USD/JPY dropped fast, from near 164 to around 158. Japanese media reported that the Ministry of Finance instructed the BOJ to act alone that day, with estimates around 6‑7 trillion yen. Later BOJ data suggested Tokyo may have sold as much as $58.97 billion buying yen in that New York session — the largest single‑day figure ever floated for Japanese intervention. Japan’s Ministry of Finance said nothing on the record.

July 31 was different: confirmed, joint, and public. Japan’s MOF and the U.S. Treasury coordinated a yen‑buying operation during New York trading hours. The New York Fed, acting for the Treasury, sold euros, not dollars, to buy yen on its behalf, and Bessent later confirmed the FIMA repo facility was activated as part of the operation. On August 3, Finance Minister Satsuki Katayama confirmed it directly and said Japan would “not hesitate to intervene further in cooperation” going forward. This was the first Japan‑U.S. joint FX intervention since the aftermath of the 2011 earthquake, and the first time since 1998 the U.S. directly participated in buying yen. BoJ money‑market data suggests Japan committed around $34 billion on July 31 alone. Combined estimates for the two days run close to $87 billion — about ¥11 trillion — though only the July 31 joint action carries official confirmation. The July 30 figure remains a market inference.

Then the leak. At a Camp David cabinet meeting on July 31, a Reuters photographer captured a handwritten note under Treasury Secretary Scott Bessent’s nameplate: “Buy $5‑10bn yen.” Bessent later said it was placed there so reporters would see it. That’s not a mistake. That’s a signaling mechanism. Before the money moves, you plant the expectation that it will. Cheaper than actually deploying capital, and sometimes just as effective for a few days.

It worked, briefly. From 9:30 am ET on July 30, the yen moved from about 162.80 to 157.80 in 50 minutes. By August 3, it reached 155.23, a roughly three‑month high. By August 4, USD/JPY was back near 157.75. By mid‑August, back to 159. The structural drivers: the rate gap and oil prices never went away. Intervention buys time, not resolution.

The Refunding Auctions: Where the Real Test Happened

On August 5, Treasury announced its quarterly refunding: $125 billion across August 11-13, split into:

• $58 billion in 3-year notes,

• $42 billion in 10-year notes,

• $25 billion in 30-year bonds.

Treasury said it expects to hold nominal coupon and FRN auction sizes steady for the next several quarters. That statement did not survive contact with the market.

Subscribe now

Read more

The Jackson Hole

The July 28-29 FOMC meeting held rates at 3.50%-3.75%. Fifth straight hold. But three FOMC voters dissented, all pushing for a 25bp hike:

• Cleveland's Beth Hammack,

• Minneapolis's Neel Kashkari,

• Dallas's Lorie Logan.

St. Louis's Musalem then said publicly he wanted a July hike too.

Kansas City's discount rate policy also leaned hawkish.

The August 26 discount rate minutes show four of twelve regional Fed boards voted before the July meeting to raise the discount rate:

• Dallas,

• Cleveland,

• Minneapolis,

• Kansas City.

Fed Chair Warsh speaks at Jackson Hole on August 28, around 10:00 a.m. ET. First appearance there since taking the chair in May. Theme this year:

• financial innovation,

• payments

• policy.

This is the most-watched policy event before the September FOMC meeting.

Subscribe now

Read more

The Market Architect Capital Research web app is now live:

https://marketarchitectcapitalresearch.com/

UI/UX experience: feedback me.

On August 19, 2026, Treasury Secretary Bessent announced the single-operation cap on long-term Treasury liquidity support buybacks would at least double, from $2 billion to $4 billion, covering 10-to-30-year Treasuries. New rules take effect September 9, running through November 4. The next day he said buyback size “could exceed $4 billion.” Treasury framed this as liquidity support for the long-end market. The market read it differently: an attempt to suppress long-term yields and ease government funding pressure.

The effect faded fast. The 30-year yield dropped from an intraday high of 5.334% to 5.183% the day of the announcement, then bounced back to 5.276% by August 21. Meanwhile, gold and bitcoin surged. Spot gold cleared $4,600/oz on August 21, the first time since May 15, then hit $4,659.96 on August 24. Gold is up over 13% since the start of August. Spot silver broke $70/oz. The dollar index fell hard after the buyback announcement, breaking below 99.00 support, extending toward 98.50 for the first time since mid-May. Bitcoin ran from roughly $64,100 on August 19 to $79,500 on August 21.

One month. A full reversal in market posture.

Warsh’s decision on July 29 and Bessent’s move on August 19 are the same problem wearing two faces: credibility. Warsh talks more than he acts. Bessent acted more than the market expected, for less than it needed.

Subscribe now

Read more

If all goes well, we'll launch this app soon (within the next few days). Find a way to reduce the deployment cost because the dataset is around 1 GB to run. Domain name: marketarchitectcapitalresearch.com

I believe the app is still imperfect.

The app is completely free for the next 6 months. We need your input to improve quality and attract more users, which will help us reduce per-user infrastructure costs (targeting ~$2/month). Because everyone can subscribe to data and build a similar thing.

For full transparency, my envisioned full system annual data/newspaper/newsletter/propertiary data subscriptions run from $10,000 to $50,000. I’ll publish a detailed breakdown of those costs in another publication (building the system) once the market validates this project's value. It is a law of physics.

We’re small, so delivery timelines are a bit longer.

The next phase involves building the economics runtime environment and adding physical metals data to the app. This app will help you track and monitor. Substack paid subscribers get early access to this.

This project may fail. After the 6-month free period ends, please subscribe for a maximum of 3 months at a time. That way, if we shut down, you only lose 3 months of fees.

I believe in good products, fair prices, and solid service to win a small, loyal niche market.

Silver futures closed August 21 at 69.53 dollars an ounce, up 2.09 percent on the day from a previous close of 68.11, ranging 67.96 to 70.08 intraday. COMEX, London vaults, and the Shanghai Futures Exchange are all showing rising inventory at the same time. That’s the headline. It’s the wrong headline. Rising inventory doesn’t mean loose supply. The question isn’t how much silver exists. It’s where that silver sits, and at what price its holder agrees to let it go.

Subscribe now

Read more

On August 18, the U.S. 30-year Treasury yield briefly reached 5.3390%, its highest level since June 2007. By the New York close, it had fallen to 5.2850%, down 2.50 basis points from the previous session. The 10-year yield reached 4.7620% just after the New York equity open, its highest since January 2025, then fell 1.90 basis points to 4.7060%. The two-year yield fell 0.70 basis points to 4.1750%. The 10-year/2-year spread narrowed from 54.30 basis points to 53.10 basis points.

This was not an American event with foreign echoes. German 10-year and 30-year Bund yields closed at their highest levels since 2011. Kjersti Haugland, chief economist at DNB Carnegie, put the issue plainly: markets were demanding greater compensation for tying up capital for long periods.

France’s 10-year yield rose to 4.105%, the highest since November 2008. Its 30-year yield reached 4.899%, near the highest level since 2008. The UK’s 50-year gilt yield reached a record 5.392%. Canada’s 30-year government-bond yield rose to 4.164%, the highest since 2010.

Japan was no exception. Its five-year JGB yield rose to 2.18%, a record high. The 10-year yield reached 2.9494%, the highest since September 1996. The 40-year yield rose to 4.2480%.

Subscribe now

Read more

August 2026 has produced two simultaneous changes in the gold market.

The visible change is the reversal in ETF flows. Months of outflows have turned into large inflows. Gold has moved through $4,500 per ounce.

The bigger change is institutional. London is trying to turn gold from one of the world’s most important physical assets into an asset that can move through modern collateral, settlement, and liquidity systems in digital form.

These are not separate stories. They are different layers of the same market.

Subscribe now

Read more

The promised web apps will be launched within these 2 weeks probably if all run smooth.

The Topline Numbers

BLS released July CPI on August 12. Unadjusted CPI rose 3.4% year-over-year, down from 3.5% in June. Lowest since May 2026. Core CPI rose 2.5% year-over-year, down from 2.6%. Seasonally adjusted, headline CPI rose 0.1% month-over-month, versus -0.4% prior. Core CPI rose 0.2%, versus 0.0% prior. Both matched consensus exactly.

Jobs data told a parallel story. BLS’s August 7 report showed nonfarm payrolls fell by 23,000 in July. Consensus expected a gain of 80,000. May was revised down from +129,000 to +63,000. June was revised down from +57,000 to +20,000. Combined downward revision: 103,000 jobs. Unemployment fell to 4.1% from 4.2%. But the drop wasn’t driven by hiring. The labor force shrank by 264,000. The participation rate fell to 61.4%. Since January 2026, participation has cumulatively dropped by 0.7 points.

Weak jobs plus cooling inflation moved rate expectations. Equity futures rose after the release. FED Watchtool cut September hike odds below 40%. Ellen Zentner at Morgan Stanley Wealth Management: the inflation data “will reinforce the prevailing ‘no need to hike’ view that formed after last week’s jobs report.”

What’s Actually Dragging CPI Down

Energy and lodging did the work. Energy prices fell 1.5% month-over-month, versus -5.7% prior. Energy commodities fell 2.9%, versus -9.5% prior. Gasoline fell 2.9%. Hotel and motel prices fell 2.8%. The World Cup ran June 11 to July 19 in the US. Some market participants tie the post-tournament demand drop to the lodging decline. Weak travel demand and elevated travel costs may also be factors. Food prices held flat, up 0.1% versus 0.2% prior.

Subscribe now

Read more

Block Japan once. Block France next time. Block China after that. It becomes a pattern. Every country wants to convert holdings to dollar cash exactly when a crisis hits, and that’s precisely when they can’t, and have to go ask the Fed instead. An asset you could freely liquidate becomes an asset someone else controls. A hostage asset.

Subscribe now

Read more

Late July 2026. The US and Japan jointly intervene to defend the yen. Japan burns through roughly $53 billion on July 30 and $34 billion on July 31, about $87 billion over two days. Treasury Secretary Bessent, working through the New York Fed, gets photographed with a memo pad reading “buy $5-10 billion yen”. First direct dollar-buying of yen by the US since 1998. First coordinated US-Japan intervention since 2011.

The core arrangement: Japan doesn’t sell its Treasuries to raise dollars. It borrows dollars from the Fed’s Foreign and International Monetary Authorities (FIMA) Repo Facility, pledging Treasuries as collateral. Bessent publicly pushed Japan to use FIMA rather than sell, and pushed the Fed to raise the facility’s cap. On X: FIMA is “an important backstop... we should encourage it to be upsized in the coming months”.

Some economists read this as a threat to the dollar’s reserve status. Goldman Sachs disagrees, saying it’s unlikely to damage the dollar’s position as the dominant global reserve currency. Walk the logic through, and the picture gets less clean than either side wants.

Subscribe now

Read more

A hold that read like a threat

On July 29, the FOMC held the federal funds rate at 3.50%-3.75%, the fifth consecutive hold since December 11, 2025. The vote split 9-3. Cleveland’s Beth Hammack, Minneapolis’s Neel Kashkari, and Dallas’s Lorie Logan all voted to raise the range by 25 basis points immediately. June had passed unanimously. Five weeks later, three members broke ranks in the same direction.

The post-meeting statement ran 115 words, the shortest in nearly two decades. Markets read it as hawkish restraint dressed up as inaction. Bank of America flagged the vote as triggering an “emerging-market-style credibility shock”: bear steepening, dollar weakness, rising breakevens, the standard signature of a market that no longer trusts the central bank’s framework.

The S&P fell 1.52%, the Nasdaq 1.74%, and the Dow dropped over 1,100 points, down 2.19%. The 30-year yield pushed to roughly 5.23%-5.28%, the highest level since 2007.

At the press conference, Warsh called it the start of a story, not the end of one. No soft target: 2% is the only number that matters. He also pointed out that market rates had already moved up since the last meeting, meaning the market had done some of the Fed's tightening for it. Many analysts labeled the outcome a "hawkish hold," warning the dissents would fuel hike bets for months.

Subscribe now

Read more

The paid subscribers' email:

Subscribe now

Read more

Warsh has held rates at 3.50%-3.75% for two straight meetings. That’s not the story. The story is what he stopped doing.

Confirmed 54-45, the narrowest margin ever for a Fed chair, straight party lines. Sworn in at the White House by Clarence Thomas, Kavanaugh in the room, first chair sworn in there since Greenspan in 1987. Trump told him: don’t look at me, don’t look at anybody, do your own thing, do it well. Powell stayed on as a governor, no precedent for that in almost 80 years.

Killing Forward Guidance

June 17, first press conference. Warsh cuts the FOMC statement to roughly 130 words, versus 300+ in prior statements, and strips out any signal about future rate direction. He calls forward guidance unsuited to the current environment. Then he skips submitting his own dot on the quarterly projections. Nineteen participants, eighteen dots.

Subscribe now

His stated logic: the Fed’s forecasting record is bad, and once a projection goes public the committee gets stuck defending it instead of reacting to incoming data. Reuters frames it directly: Warsh wants bond markets to do the steering on where money gets priced. QNB’s read matches: less forward guidance, more flexibility to move fast when conditions shift.

It’s costing him something. After the July 29 meeting, September-hike odds dropped 20 percentage points to 41% on CME FedWatch, yet long-term Treasury yields rose at the same time. That’s the market pricing uncertainty in two directions simultaneously, not because the data is contradictory, but because the Fed stopped telling anyone which way it leans.

The Fight He Asked For

Warsh said he wanted “a good family fight” inside the committee. He got it July 29: 9-3 vote to hold, three regional presidents: Hammack, Kashkari, Logan, dissenting for a 25bp hike. Largest number of dissents on one side since 2016.

June’s dots already showed the split: nine of 19 penciling in a hike by year-end. Warsh isn’t managing this fracture away. He’s letting it show on the record before the vote happens, which is a different kind of institution than the one that spent two decades chasing unanimous statements.

Two Camps, One Ranking Problem

Dissenters: current rates aren’t restrictive enough, inflation stays sticky, the real fear is a 1970s-style entrenchment from hesitating too long.

Majority: 3.50%-3.75% is already doing work, June CPI fell 0.4% month-over-month, first monthly drop since May 2020, annual rate down to 3.5% from 4.2%. Their fear is triggering a recession by overtightening into a slowing labor market.

This isn’t a data dispute. It’s a disagreement about which tail risk to insure against first: credibility risk from letting inflation re-accelerate, or growth risk from breaking the labor market. Warsh’s own position on the target hasn’t moved: “no soft target… only one target, 2%,” restated at the July press conference. What moved is his willingness to say how he’ll act on it.

The Meeting Calendar Is Also on the Table

Separately from the five task forces: communications, the $6.7 trillion balance sheet, data methodology, a new Productivity, Labor & AI panel, and the inflation framework. Warsh is now weighing cutting the number of scheduled rate-setting meetings itself, down from the current eight a year. He raised it at his second meeting and wants something in place before the mid-September meeting. The Fed hasn’t commented. Fifteen outside co-leads on the task forces include Thomas Sargent, Mervyn King, Raghuram Rajan, and Marc Andreessen. Waller’s objection at the pre-meeting dinner still stands unanswered: tell me who’s running the review, and I’ll tell you the conclusion.

The Jobs Report Just Landed

July nonfarm payrolls: -23,000, against a consensus of 83,000. Consensus wasn’t close. May and June revised down a combined 103,000-126,000. Unemployment fell to 4.1%, but only because participation dropped and roughly 264,000 people left the labor force; the drop is a withdrawal signal, not a strength signal. Private payrolls rose 30,000; government cut 53,000, doing most of the headline damage.

This lands on the growth-risk side of the ledger, right where the three dissenters weren’t looking. September hike odds on CME FedWatch were sitting at 56.7% before the report; NBC noted futures already showed above 50% odds pre-release, with the outcome now hinging on whether the September call turns into a hike-on-hot-inflation scenario Warsh has reportedly signaled he’d accept. Citi’s out-of-consensus call is now three rate cuts by January 2027. Different data providers don’t even agree on the print. Revelio’s private index showed +79,200 for the same month, ADP showed +44,000, and Vanguard’s internal 401(k) data implied near-zero growth scaled to ~9,000. That spread is itself the story: when the official print, gig-economy data, and private payroll data diverge this much, “let the market read the data” stops being a clean alternative to forward guidance. It just relocates the guesswork from the Fed chair’s mouth to a set of conflicting spreadsheets.

What’s Actually Being Tested

Warsh is betting that talking less buys credibility: fewer forecasts, no dots, statements that only state facts already on the books. The exposure: a central bank’s credibility isn’t just adherence to 2%. It’s whether the market can predict how the bank reacts when data conflicts. Strip out forward guidance and the market loses the reaction function, not just the forecast. Rising long yields the same day hike odds fell is that mechanism already firing.

The real test isn’t whether markets can function with a quieter Fed. It’s whether an institution that explains itself less can still convince anyone it knows what it’s doing when the labor data and the inflation data start pointing in opposite directions at the same time, which is exactly where things stand right now.

August 4, 2026. Treasury Secretary Scott Bessent tells CNBC a deal with Iran to reopen the Strait of Hormuz “could happen today or tomorrow.” He adds that once the stranded ships get clearance to leave, “oil prices should fall further.” Multiple outlets report him citing “hundreds, maybe thousands of ships” waiting.

Timing: before the New York open. Brent falls 5.10%, closes at $79.49. WTI falls 5.54%, closes at $75.89. The 30-year Treasury yield pulls back from the prior week’s high of 5.28% to close at 5.254%. Dollar-yen closes at 157.18, having dipped near 155.2 intraday, extending the yen’s prior rally. Dow closes at 54,085.88, S&P 500 at 7,736.52, both record closes. Nasdaq up 2.59%.

Subscribe now

Rate futures and oil move together the same day. The market treats the statement as a credible de-risking signal. Rubio backs the line hours later: negotiations have “made progress,” no final deal yet.

The Physical Constraint Being Talked Around

Cushing, Oklahoma: the designated delivery point for WTI futures. EIA weekly data: commercial crude stocks there fell 771,000 barrels in the week to July 24, to 18.6 million barrels, lowest since October 2014. Energy Aspects flags 10+ million barrels as the point where operational strain sets in. EIA’s own explanation: tanks and pipelines require a minimum volume of crude to keep pumping, “tank bottoms” that can’t actually reach the market. So when the headline number drops below 20 million, the deliverable volume is materially lower than the reported figure.

This constraint already shows up in the WTI curve. Near-month contracts trade at a widening premium to deferred months through late July, backwardation, historically associated with tight spot supply.

Strategic Petroleum Reserve, same direction. EIA data: down 2.841 million barrels in the week to July 31, to 304.8 million barrels, lowest since 1983: 427 million barrels below maximum capacity. Stored in Gulf Coast salt caverns. DOE facility assessments have previously flagged degraded withdrawal and injection capacity at some sites from sustained heavy use and deferred maintenance. Industry consensus puts the operational floor at 250-300 million barrels. Current level sits near the top of that band; further drawdown starts constraining the reserve’s ability to respond to a shock at all.

Context for why the drawdown happened this fast: the US has been acting as the marginal global crude supplier while Hormuz and Bab-el-Mandeb were both constrained. EIA export data show US crude and product exports holding at elevated levels through the disruption, eating into the domestic buffer. IEA director Fatih Birol said in May the world was approaching the “red zone” for oil supplies by July or August, driven by Hormuz disruptions and rapid inventory drawdowns, the same red-zone language Cushing data now confirms directly.

The scale of the underlying shipping disruption: as of early August, tanker crossings through Hormuz have been running at a handful per day, down from 100-140 before the conflict began in late February. IMO and UN estimates put 1,500-2,000 vessels stranded in the Gulf, with thousands of crew stuck aboard. IEA has called it the largest supply disruption on record, exceeding the 1970s oil shocks and the post-invasion loss of Russian pipeline gas combined.

Futures Positioning: Two Benchmarks, Two Different Signals

CFTC data through July 28, before Bessent spoke. Brent: managed-money net-longs near a five-year high, speculative long positioning crowded, commercial hedging light. Historically, that setup precedes downside reversals.

WTI: managed-money net-longs rose 28,159 contracts to 106,507 — but the overall net-long level is still low by five-year standards, nowhere near crowded. The anomaly sits on the commercial side. Producer/merchant net-long positioning is at an extreme high. Commercials normally hedge short. A net-long position this large on the commercial side is rare — it reads as producers and traders declining to hedge downside and instead expressing a direct bet on physical tightness, which lines up with Cushing sitting at 18.6 million barrels.

Two markets, two different signals. Brent’s crowded speculative long reflects geopolitical premium and shipping-disruption expectations layered on a global benchmark. WTI’s commercial extreme points directly at the physical bottleneck at the delivery point. Falling prices, alongside an extreme WTI commercial long, usually mean the market is waiting for a catalyst before it commits to a direction. Bessent’s statement supplied that catalyst, and it pointed the same way the positioning data already implied: down.

How Oil Stress Became a Bond Market Problem

The transmission channel runs through Japan. Japan is heavily import-dependent on energy. When Hormuz jammed, Japanese buyers pivoted to US crude to cover the gap. METI’s petroleum statistics: June 2026 crude imports from the US around 3.24 million kiloliters, about 73 times the year-earlier volume. Japan’s Ministry of Finance trade data puts the figure at 2.722 million kiloliters, roughly 5.6 times the year-earlier volume, the highest since 1979. The gap between the two numbers is a methodology difference: METI’s petroleum statistics and MOF’s trade statistics use different scope and measurement — but the direction is unambiguous either way. MOF also shows the yen value of those June imports at ¥335.8 billion, roughly tenfold year-on-year.

Sustained import volumes at that scale generate sustained dollar-buying pressure. Yen weakens. Late July: yen briefly touches 164 to the dollar, a nearly 40-year low. Weak yen raises import costs further, feeds imported inflation, forces a reassessment of BOJ policy and Japanese fiscal discipline. JGB yields rise, volatility widens. That JGB volatility triggers rebalancing at risk-parity and multi-asset funds holding both JGBs and Treasuries; losses on the JGB leg force cuts to the Treasury leg to hold portfolio risk constant. Not a view on US fundamentals. A mechanical flow.

Result: July 30, the 30-year Treasury yield jumps 10bp to 5.21%, first break above 5.2% since June 2007. Next session, intraday high of 5.28%. Japan is the largest foreign holder of Treasuries; forced JGB-driven Treasury selling pushes US long-term funding costs up directly.

July 31: joint US-Japan FX intervention. Finance Minister Katayama Satsuki confirms it August 3 — coordinated yen-buying, the first time since the 1998 Asian financial crisis intervention has taken that direction. FT reporting, citing sources, says the New York Fed executed on Treasury’s behalf through Goldman Sachs and Morgan Stanley, selling euros to buy yen. Reuters photographs a Bessent notepad at the August 3 cabinet meeting: “To do: buy JPY, 5-10bn.” Intervention lifts the yen short-term. Bond market unease doesn’t fully clear.

Why a Treasury Secretary’s Words Moved More Than a President’s

Trump has cycled through statements on Iran for months: threats, promises of assistance, no consistent follow-through. Market reaction to his statements on this topic has visibly dulled.

Bessent’s statement lands differently for two reasons. It’s specific, “hundreds, maybe thousands of ships,” not diplomatic boilerplate, giving the market a concrete downside scenario to price rather than a sentiment cue. And it’s timed for maximum absorption, delivered ahead of the cash open so investors have the full session to digest and react.

The result: a coordinated reversal across oil, bonds, FX, and equities the same day, all consistent with one narrative: supply relief incoming. The positioning backdrop did the rest of the work: Brent’s crowded speculative longs were already primed to unwind, and WTI’s commercial extreme had already priced the tightness. The catalyst didn’t need to do much. It just needed to point.

Meanwhile, the companies sitting on the physical supply side had already booked the upside. Aramco reported a 33% jump in second-quarter profit, citing its ability to maintain business continuity through the Hormuz disruption via alternate infrastructure like the East-West pipeline. Big Oil more broadly posted record profits through the disruption period; the tightness Bessent was talking down had already been monetized upstream.

What the Statement Didn’t Settle

By August 5, Iran denies resuming direct talks with the US, says indirect discussions are proceeding through Oman. President Pezeshkian says Iran will “defend its territory” but is “not seeking to expand the war.” Rubio confirms “progress,” not a final deal. Houthi spokesman Yahya Sarea claims a ballistic missile hit the Saudi tanker Daisy in the Gulf of Aden; the ship’s management company later denies any strike occurred, says the vessel and crew are safe. Verification ongoing.

Counterpoint on supply: US crude production is running near 13.804 million bpd as of the week to July 31, up 8,000 bpd week-over-week. OPEC+ spare capacity and incremental barrels from Canada and Brazil haven’t been fully tapped. Reuters’ own analyst poll from July 31 leaned the other way on timing; normalization of Gulf flows was expected to take four to six months after a durable ceasefire, with one analyst’s base case putting full normalization at early 2027. IEA data shows global oil inventories rose 21 million barrels in June after falling 73 million barrels in May; some rebuilding was already underway before Bessent’s statement, but nowhere near the pace needed to reverse the Cushing and SPR drawdowns. API’s own weekly estimate showed a 2.69 million barrel US crude build the same week Washington was talking up the peace deal, inventories inching up even as the headline diplomatic narrative did the heavier lifting on price. Public data on the actual deployable size of OPEC+ spare capacity remains limited; how much of that can hit the market fast is unverified.

The market chose, in the short run, to price the promise of supply relief over the current state of supply. If diplomatic progress stalls in the coming weeks, the inventory constraints and the fund-flow pressure parked on August 4 don’t disappear. They come back into energy and bond pricing.

Two days, July 30–31, 2026. U.S. Treasury and Japan’s Ministry of Finance intervene together in the foreign exchange market. Bank of Japan money-market data puts the combined total near 87 billion dollars, about 11 trillion yen. The yen bounces roughly 5% off a near-forty-year low. First large-scale U.S. yen purchase since 1998. First joint action with Japan since 2011. After the dust settles, most analysts say the same thing: the forces that pushed the yen down remain.

The mechanics of two days

July 30, Dense dollar selling hits the dollar-yen market. Japanese media reports that the Ministry of Finance instructed the BOJ to act alone that day, with the amount estimated at 6-7 trillion yen. Bank of Japan data later suggests Tokyo may have sold as much as 58.97 billion dollars buying yen in that New York session.

July 31, the U.S. joins. Reuters photographs a handwritten notepad on Treasury Secretary Scott Bessent’s desk at a Friday cabinet meeting. Under “To Do”: “Buy Japanese Yen (JPY) $5-10 bil.” Never officially confirmed as an exact figure. Same day, the New York Fed, acting for the Treasury, requests yen quotes from primary dealers — a rate check, the standard prelude to intervention. One detail sharpens the picture: the Financial Times reported the New York Fed actually sold euros, not dollars, to buy yen on Treasury’s behalf. Bloomberg’s read of BOJ account data puts the joint move at roughly 5.33 trillion yen, about 34 billion dollars. On August 2, Bessent confirms publicly that the Fed’s FIMA standing repo facility was activated as part of the operation.

Subscribe now

Timing wasn’t accidental. Nomura’s Yujiro Goto points to July 30 landing on the Fed’s rate decision, a weak U.S. Q2 GDP print, and month-end rebalancing — thin liquidity, so a smaller amount of capital buys a bigger price move. CFTC data through July 28 shows speculative net short yen positions at 163,400 contracts, a five-year high, up nearly fourfold in five months. Traders say the intervention set off short covering that did most of the heavy lifting on the way up. Dollar-yen falls from around 163.90 to an intraday low of 155.20 by August 3, about 5% in three days. But BOJ account data released August 4 shows no big outflow on August 3 itself — no direct sign of official buying that day. Translation: that day’s move was the market clearing out its own positioning, not fresh official firepower.

What Washington was actually defending

Several analysts argue the U.S. motive runs past ordinary currency diplomacy. The chain gets repeated: as the yen weakens, Japanese institutions hedging currency exposure, or just chasing yield, keep trimming their U.S. Treasury holdings. Xinhua frames it directly — stabilizing the yen isn’t only about helping an ally; it’s about stopping the largest foreign creditor from dumping Treasuries to defend its own currency. U.S. Treasury data shows Japan holding about 1.14 trillion dollars in Treasuries as of May 2026, down about 96 billion dollars over three months, the lowest since April 2025. Before the intervention, 30-year yields hit 5.28%, the highest since 2007; 10-year yields rose over 30 basis points in July alone, the biggest July move since 2005.

If Japan keeps funding intervention by selling Treasuries outright, it pushes U.S. borrowing costs higher. Bessent pushing Japan toward the FIMA repo facility — foreign central banks borrow dollars against Treasury collateral instead of selling it, capped at 60 billion dollars a day — is designed to cut the need for outright Treasury dumping. Bessent called FIMA repo an “important backstop” on X on August 3 and pushed to expand it. CNBC’s framing lines up: highlighting FIMA access “tells markets Japan can raise dollar liquidity without selling Treasuries”. Japan’s currency chief Mimura pushed back a little on the emphasis, saying FIMA repo is just one tool among several and its funding limits don’t cap Japan’s broader intervention capacity.

Not everyone accepts the Treasury-defense story as the full explanation. BNP Paribas’s Calvin Tse calls it an oversimplification — the intervention also carries political solidarity with an ally, concern for market stability generally, and a judgment call on the risks of an overly strong dollar.

Japan’s own hands are tied

The flip side of U.S. willingness to co-intervene: Japan’s policy toolkit is boxed in. Government debt sits above 240% of GDP, worst in the G7. The FY2026 budget shows “national bond expenses” — servicing existing debt — at 31.28 trillion yen, breaking 30 trillion for the first time. That structure makes the BOJ wary of hiking, because higher rates directly raise the cost of rolling new issuance, squeezing an already tight budget.

Prime Minister Sanae Takaichi, after dissolving parliament and winning the election, kept pressing the BOJ to stay loose to support growth. That pressure has been lowering market expectations that the BOJ will hike hard enough to fix the yen on its own. The BOJ’s July 31 decision: rates held at 1.0%, 8-1 vote, after a June move from 0.75% to 1.0%. The intervention worked technically but didn’t touch the rate differential doing the real damage — “what the market read from it is that policymakers are still avoiding the rate-hike option.”

History backs the skepticism. April-May 2026, Japan ran a record roughly 73.2 billion dollars of intervention. The yen gave it all back within a month.

Three things to actually watch

Short-term noise is over. What’s left is checkable.

FIMA usage going forward. Finance Minister Satsuki Katayama said August 3 that authorities are ready to intervene again, without naming the funding source. Japan’s ministry has said it plans to use the FIMA facility for future interventions. Watch whether the Fed expands the facility’s scale as Bessent floated, and whether Japan actually draws on it again.lcx

BOJ policy signals. The July meeting minutes get read word for word for any softening. Given Takaichi’s pressure, most analysts expect the minutes to keep leaning dovish, even as the BOJ separately signals it’s open to further hikes if inflation risk builds.

Carry-trade unwind risk. A fast further yen rally could force cross-asset unwinds, hitting emerging-market currencies and stretched equities first. Precedent: August 2024, a surprise BOJ hike sent the yen up fast, forced carry unwinds, and the Nikkei 225 dropped 12.4% intraday — the worst single day since 1987.

What decides the rest is whether the rate gap actually narrows and whether the BOJ finds room to move under domestic political weight. Until those answers show up, the yen’s weak cycle doesn’t get reversed by intervention alone.

Quick Update:

I want to be honest about where the project stands.

My original goal was to build the first layer of a software runtime (to reason about macroeconomic conditions) driven by economic theory and live macroeconomic data. This runtime would reason over macroeconomic data and COT positioning to generate higher-level insights. I still believe that’s the real value of the software. If it only fetches data, there’s little reason for people to pay.

But I hit reality. I’m not an economist or finance expert, and building something credible requires far more knowledge, data, and resources than I have today. I also spent a lot of time testing the runtime code generated by Claude Code, but after many rounds of validation, the results weren’t reliable enough. It is my first time building such a complex runtime. I underestimated the amount of economics, finance, and quantitative modeling knowledge required to build the runtime to the standard I wanted.

The macro data pipeline is mostly done. COT data is complete. The news workflow is still incomplete because the LLM-powered reasoning layer hasn't been integrated yet.

I’ll still launch the MVP, and it will be completely free. I don’t think it’s right to charge for something that doesn’t deliver the value I originally envisioned. My brother also helped improve the workflow, since the AI-generated workflow wasn't good enough.

The MVP will include macroeconomic data, COT data, and the economic events I’ve integrated.

The probabilistic inference engine is not part of the MVP. Building it requires market microstructure data, and the data subscriptions alone cost over $2,000 per year.

• The software includes four modes:

  • Trader Mode – A workflow designed to help traders quickly monitor macroeconomic data, COT positioning, signals, watchlists, and upcoming economic events.

  • Research Mode – Planned, but not included in the MVP. This was intended to be powered by an economic runtime together with an economics/finance knowledge graph, proprietary data, task state, etc. Instead of only displaying data, it would connect macroeconomic indicators, central bank policies, inflation, employment, COT positioning, news, and market behavior into a structured research workflow. The goal was to explain why markets move and help users reason through market movements.

  • Analyst Mode – Planned, but not included in the MVP. Designed to estimate risk, evaluate reward scenarios, and generate structured trading or investment analyses.

  • Raw Data Mode – Direct access to the underlying macroeconomic data, COT reports, and economic calendar (raw sources) without additional interpretation.

  Only simplified Trader Mode is fully available in this MVP.

Thank you to everyone who has followed my journey. I'm sorry if this is disappointing.

To those who subscribed and later unsubscribed because I overpromised, I'm sorry. I appreciate your support.

If you have a strong background in economics, macroeconomic modeling, or quantitative finance and are interested in building this with me, I'd be happy to hear from you. I can't offer payment at this stage, but if the project succeeds, we'll share the profits. Feel free to reach out.

Building Systems Thinking

I am Building a Market Microscope/Eye That Thinks

My System I am Trying to Build…

Read more

3 months ago · 2 likes · Jin

The Fed held rates at 3.50%-3.75% on July 29, 2026. Not news. All 104 economists in Reuters’ survey called it. The news is the vote: 9-3. Three dissents, same direction, all wanting a 25bp hike. Hammack from Cleveland, Kashkari from Minneapolis, Logan from Dallas. First time since 2016 three directional dissents landed on one decision.

Three Dissents

Each dissenter published a full written statement. Nick Timiraos — the reporter markets read as the Fed’s unofficial channel — made a pointed observation: the three hawks explained their reasoning more completely than the majority did.

Hammack’s position is the most direct. She says she lacks confidence inflation returns to 2% on its own. Current policy is “not appropriately restrictive.” Inflation has sat above target for five years. Labor market still solid. Economy can take higher rates. Her line: the longer high inflation persists, the harder and more expensive it gets to bring down.

Kashkari runs risk management logic. Tighten gradually while collecting more data on inflation and employment. He flags a specific structural driver: the AI data center buildout is injecting fresh demand heat into the economy. His warning — if inflation stays elevated, a series of small proactive moves beats sitting still and getting forced into a violent, aggressive one later.

Logan goes furthest. Even adjusting for supply shocks and productivity gains, she sees inflation settling near 2.5%, not drifting back to 2%. That’s a claim the inflation target itself has shifted structurally higher. Her warning: without policy constraint, inflation likely stays above target until an unexpected shock forces it down.

Shared conclusion across all three: the cost of waiting outweighs the cost of acting now.

“Watch the Game, Not the Referee”

The vote split exposes one fracture. Waller’s press conference exposes a deeper one — how the Fed communicates with markets.

Since June, FOMC statements have shrunk hard. Forward guidance has been stripped down. This statement ran about 130 words. Under Powell, statements routinely ran past 300. Waller kept the pattern in July. He refused to call this a “pause,” calling it instead a “period of deliberation.” His stated philosophy: “Market participants are learning to watch the game, not the referee.” Earlier in July, in congressional testimony, he said it more bluntly: “My message is: play ball, don’t fight the Fed.”

For decades, markets tracked every Fed signal. Waller wants markets tracking fundamentals instead. Bloomberg’s framing: from wars to tariff fights to sticky inflation, markets now face another risk layered on top — a Fed that won’t tell them what it’s thinking.

Market response isn’t calm adaptation. Breakingviews called scrapping forward guidance potentially counterproductive.

Yield Curve: Dovish Front, Hawkish Back

After the decision, Treasuries moved in an unusual pattern — short rates down, long rates up. A bear steepener. Close on July 30: 2-year at 4.248% (down 3.1bp), 10-year at 4.675% (down 0.4bp), 30-year at 5.216% (up 1.2bp). The 30-year hit 5.226% intraday — first time above that level since 2007. Some analysts called this one of the sharpest bear steepenings since the mid-1990s.

Three structural forces sit behind it:

• Term premium is normalizing upward. The New York Fed’s ACM model put the 10-year term premium near 0.6% in late May 2026, versus near-zero or negative for most of the prior decade.

• Fiscal deficits keep widening, and Treasury issuance keeps growing, pushing investors to demand more compensation for holding duration.

• Stock-bond correlation has flipped. Moderately negative from 2003-2021, it’s run around +0.5 since 2022, averaging closer to +0.6 recently. Standard 60/40 hedging logic stops working under that regime.

  Subscribe now

Financial Conditions: The Core Contradiction

Waller’s claim — the market already did the tightening for the Fed — runs into a direct problem. Has the long-end move actually tightened overall financial conditions? Data says no.

Goldman’s US Financial Conditions Index sits at its loosest level since early 2022, near the least restrictive reading since 2009, about 1.2 standard deviations below its long-run average.

The Chicago Fed’s National Financial Conditions Index fell to -0.515 for the week ending July 3, 2026 — loosest since February. Negative NFCI means conditions run looser than the historical average.

Both indices point to the same contradiction: rising long-end yields have not transmitted into tighter overall conditions. Rallying equities and compressing credit spreads are absorbing the tightening effect from higher long bond yields.

The Taylor Rule Gap

The Taylor Rule offers a reference point. HB Wealth’s calculation: inflation running 1.4 points above target, GDP 1.1 points above potential, implies a policy rate of 6.7%. Actual fed funds rate at the time was about 3.63%. Even running 2027 consensus forecasts through the rule, it still points to 5.5% — well above the current 3.75%.

Caveat: Taylor Rule output is highly sensitive to parameter choice — neutral rate, output gap measure, inflation gauge all shift the number. But across common parameter combinations, the implied rate stays well above the actual rate. By this framework, Fed policy still reads as loose.

A Shift in Communication

The July meeting matters beyond the “no change” headline. Three shifts are stacking simultaneously.

FOMC internal governance: three aligned dissents expose a real fracture over how to handle inflation risk.

Fed-to-market communication: Waller stripping out forward guidance changes how markets form rate expectations — from being told to figuring it out themselves. Robeco’s analysis notes removing forward guidance likely increases short-end volatility, since every meeting now functions as a “live” meeting with no pre-signaled outcome.

Treasury pricing mechanism: long-end rates are increasingly set by market forces rather than by central bank signaling itself.

Waller told the press conference “this Fed will not waver.” But markets aren’t pricing the promise. They’re pricing what happens after the promise. The widely-held read: the sharp rise in long yields directly reflects a credibility gap — Waller’s confident talk that “price stability will be achieved” versus the central bank’s actual refusal to tighten.

If inflation data stays sticky over coming months and financial conditions don’t tighten despite higher long rates, July may not be remembered as a pause. It may be remembered as the start of a framework shift — one where the Fed leans more on market pricing to do part of the tightening job. And judging by the 30-year breaking above 5.2%, the market’s answer may end up forcing the Fed to rethink this communication experiment.

Respect Michael Burry! Shorting during an uptrend. Insightful person.

It is just my perspective; I might be wrong.

What Burry actually shorted was never Nvidia or Micron. It was a premise.

Here’s the mechanism.

On July 2, 2026, Burry disclosed on his Substack “Cassandra Unchained” that he shorted Micron directly at $1051.87 a share. Not puts. He said puts looked overpriced. Micron had run up nearly 700% over the prior year, 241% year to date. He called it “the poster child of cyclicality” — 34 drawdowns over 30% in 42 years. Its deviation from the 200-day moving average was the highest since 1984, higher than at the dot-com peak. Median return on invested capital: 4%. Median ROE: 7%. He called that “frankly poor.” Micron closed down 5.49% that day, at $975.56.

This wasn’t isolated.

On June 30, he disclosed shorts across the entire AI capex chain: Nvidia at $198.09, Applied Materials at $729.40, the SOXX semiconductor ETF at $642.80, Caterpillar at $1060.98 — his first short on Caterpillar ever — and Tesla at $416.22. On Tesla, he wrote: “happy it bounced back to this level.” Tesla had rallied about 8% off recent lows to $411.84 before he shorted it. He waited for the bounce, not the dip.

By July 24, most of these were profitable. Tesla down to $319.69, about 24%. Caterpillar down to $894.54, about 16%. SOXX down to roughly $551. Applied Materials down to roughly $562.80.

On July 27, Nvidia fell 4.99% in a single day, closing at $196.51. About $250 billion in market cap gone in one session. Apple retook the title of world’s most valuable company. Burry posted on X: “The market has voted, and the results are clear.” Attached: a Bloomberg chart showing Alphabet, Amazon, Microsoft, Meta, and Oracle — the biggest AI capex spenders — all underperforming the broader market since June.

This is his largest, widest short campaign ever run. But the trade list isn’t the story. The story is what he thinks he’s seeing underneath it.

The track record that gives this weight

Scion Asset Management manages only about $155 million as of March 2025. Small. What makes Burry matter isn’t size; it’s history: Scion Capital returned 489.34% net from November 2000 to June 2008, against roughly 2% for the S&P 500 over the same stretch.

In November 2025, his 13F showed puts on Nvidia and Palantir — Palantir notional value about $912 million, Nvidia about $186 million, together 80% of his reported portfolio. He clarified on X immediately: actual cost was $9.2 million, not $912 million. Each contract just gives him the right to sell at a $50 strike. “I spent $9.2 million. Not $912 million.” Palantir’s CEO Alex Karp called the short “batshit crazy” on CNBC.

Then on November 10, 2025, Burry deregistered Scion from the SEC entirely. His letter to investors: “It is with a heavy heart that I will be closing the fund and returning capital by year-end. My assessment of securities values is not, and has not been for some time, in sync with the market.”

He moved the entire operation to Substack. $39/month subscription. 300,000 subscribers within weeks. At that scale, subscription revenue alone runs roughly $140 million a year.

The actual argument: language is not reasoning

On July 10, 2026, Burry published a long essay making a sharper claim than “overvalued.” His argument: the AI industry’s original goal was autonomous reasoning. That proved too hard. The field collectively pivoted to large language models — the lower-difficulty path. He calls this the industry’s “congenital defect.”

His line: “We mistook the output of AI (language) for its engine (reasoning).”

The logic underneath: understanding doesn’t require language. Language is an output of intelligence, not its source.

He frames this with what he calls “Ballard’s Test,” named after Melville Ballard, a 19th-century deaf-mute educator who reasoned independently about human origins and the nature of time before he ever acquired language — documented in an 1880 Smithsonian paper. The test: real intelligence must demonstrate reasoning independent of language structure. Does an LLM pass? Obvious answer.

His second concept: the “parameter trap.” The whole industry stacking parameters, brute-forcing scale through compute, in a loop that never touches native cognitive architecture. Every scale-up nudges benchmark scores and demo quality upward — enough to keep investors doubling down, not enough to solve the underlying reasoning problem.

Silicon Valley’s operating logic here: be first or die. In 2016, Musk emailed OpenAI co-founder Greg Brockman: “DeepMind is causing me extreme mental stress.” If DeepMind won, their “one mind controls the world” philosophy would be “quite bad.” What’s driving OpenAI, in Burry’s read, isn’t curiosity. It’s fear. Not scientific logic — an arms race.

Language generation gets better and better. Investors reward that with higher and higher valuations. Somewhere in that loop, the difference between generating language and actually reasoning stopped being checked.

He’s not alone on this

Subscribe now

Noam Chomsky has argued for years that the LLM path to AGI is fundamentally wrong. His distinction: humans use finite means to produce infinite, general thought. You can say “the apple falls” (description) and “if I let go, the apple will fall” (prediction) — AI does both fine. But “if you drop anything, it falls,” “because of gravity,” “if there were no gravity, the apple wouldn’t fall” — that’s hypothesis and counterfactual reasoning. AI doesn’t do that.

Yann LeCun, Turing Award winner, told the New York Times in early 2026 the industry is heading into a “dead end.” Scaling language models alone won’t produce real world-understanding, in his view. He left Meta in late 2025 to found AMI Labs, focused on “world models” instead.

Former DeepMind chief scientist David Silver’s thought experiment: train an LLM purely on text from an era when everyone believed the earth was flat. It becomes a committed flat-earther because it can’t interact with the real world to verify. Real AGI, in this framing, has to learn by exploring the world itself, not by predicting the next token in a corpus.

What this means for pricing

On May 8, 2026, Burry warned: “This feels like the final months of the 1999-2000 bubble.” Driving somewhere, listening to financial radio, he noticed the coverage was “completely, relentlessly AI, all day, nobody talking about anything else.” His line: “Stocks aren’t going up or down because of jobs data or consumer sentiment. They go straight up because they’ve been going straight up. Based on a thesis everyone thinks they understand that’s really just two letters.” He compared the Philadelphia Semiconductor Index’s run — up 65% in 2026 — to its trajectory right before the March 2000 crash.

In early July, he escalated, quoting the Joker from the 1989 Batman film on X: doom incoming. His words: “The AI narrative is nothing more than mass addiction.”

He’s also accused hyperscalers of inflating profits by stretching depreciation schedules on AI servers. His estimate: underreported depreciation of about $176 billion from 2026 to 2028, inflating reported industry profit by roughly 20%.

Subscribe now

Oracle: the size problem

In January 2026, Burry disclosed puts on Oracle, built on aggressive AI infrastructure spending and dangerous customer concentration in OpenAI.

By July 2026, he cut the Oracle short in half — not because he’d turned bullish, but because the position “got too large.” Oracle had already fallen from a 52-week high of $345.72 to $121.50, down about 65%. S&P Global cut Oracle’s credit rating to BBB- on July 9, one notch above junk, citing that roughly half of its $638 billion order backlog comes from OpenAI alone — extreme concentration risk.

Meanwhile, the bigger number: Goldman Sachs projects the four largest hyperscalers — Alphabet, Amazon, Microsoft, Meta — will spend $750 billion on capex in 2026, up about 77% from 2025. Goldman’s cumulative estimate for global AI capex from 2026 to 2031: roughly $7.6 trillion.

A Nikkei Asia investigation in July 2026 found off-balance-sheet debt — long-term data center leases and GPU purchase commitments not counted as liabilities — at five tech giants grew eightfold over four years to $1.65 trillion, dwarfing their $1.35 trillion in formal on-balance-sheet debt. Meta’s off-book exposure alone: roughly $420 billion. Moody’s analyst David Gonzales: “This unrecorded debt burden creates a risk profile far higher than traditional financial statements suggest.”

The real bet

All of this spending, all of this pricing, rests on one premise: that scaling LLMs is the road to AGI. If that premise is wrong, the spending doesn’t need to be wrong for the stocks to be wrong. The price only needs to have assumed more certainty than the premise deserves.

Burry doesn’t need AI to fail. He needs the market’s pricing of AI’s success to be more confident than reality warrants.

In 2008, he questioned, “home prices don’t fall nationally.” In 2025-2026, he’s questioning “AI is worth exactly this price, right now.” Seventeen years apart. The asset class changed. What didn’t change: markets believing they can’t be wrong.

Technology answers “what will happen.” Investing answers “how much has the market already paid for that future.” Those look similar. They aren’t the same question.

LeCun still calls LLMs a dead end while Anthropic’s ARR blew past $60 billion in mid-2026. Chomsky still doubts AI produces real thought while Nvidia GPUs remain sold out. Great technology and correct pricing have never been the same event.

The AI bubble, if it deflates, probably doesn’t do it in one crash. Burry’s own framing — a thousand cuts — fits better: as the gap between actual capability and investor assumption gets harder to paper over with bigger parameter counts, pricier compute, better benchmark scores, belief erodes piece by piece rather than all at once.

Nobody knows where AI actually lands. But Burry’s question stays useful exactly when consensus feels most solid: is the premise everyone’s accepted proven, or just believed?

I found another web apps that provide similar services. I will add this to phase 2. Because it is very important to track them too. So that we can know which factors or drivers trigger the price.

• https://goldsilver.ai/

Gold and silver have been falling since June into July. Gold kept testing 4000, broke below it again on July 29, the first breach since July 21. Silver dropped to the 55-57 range, spot at 56.90 on July 29. Investors split. Some question gold’s long-term case. Others treat the pullback as a reload opportunity. Both camps circle the same question. What is gold actually for?

Greenspan’s answer

Greenspan died June 22, 2026, at his Washington home, from Parkinson’s complications, age 100. Reagan appointed him Fed chair in 1987. He held the seat almost 19 years, through Reagan, Bush, Clinton, and Bush, and retired in 2006.

October 29, 2014, Council on Foreign Relations, Financial Times interview. Question: Is gold a good investment? Answer: Yes. Then the line that gets requoted every cycle: “Gold is a currency. It is still, by all evidence, the premier currency where no fiat currency, including the dollar, can match it”. He called gold and silver intrinsic currencies, accepted without third-party guarantee. Then he flips the question back. Why would a central bank hold an asset with zero yield and a storage-and-insurance bill attached? If gold were just a commodity, central banks would stockpile copper, wheat, oil instead. They don’t. That’s the tell. Not backed by any government. Not anyone’s liability. No default risk. Ten years of buying since that interview keeps confirming the point.

Central banks acting on it

Top five official gold holders now: US 8,133.5 tonnes, Germany 3,350.3 tonnes, Italy 2,451.8 tonnes, France 2,437 tonnes, Russia around 2,326 tonnes. IMF holds about 2,814 tonnes, counted separately. Global official gold sits near 38,700 tonnes, about 18% of all gold ever mined.

2022 through 2024, central banks bought over 1,000 tonnes a year, three straight years. 2025 net buying: 863.3 tonnes. Q1 2026: 243.7 tonnes, up 3% year over year, up 17% quarter over quarter.

Subscribe now

June 16, 2026, World Gold Council published its Central Bank Gold Reserves Survey. 76 responses, a nine-year high. 89% expect global gold reserves to keep rising over the next 12 months. 45% plan to add to their own holdings, a record. 83% expect gold’s reserve share to keep rising over five years. 93% of respondents currently hold gold, up from 81% last year.

Gold has passed Treasuries. The ECB’s June 2026 report, “The International Role of the Euro,” puts gold at 27% of global official reserves at end-2025, ahead of Treasuries at 22% and the euro at 15%. But the ECB is explicit about why: this is a valuation effect, not a buying spree. Gold rallied about 60% in 2025 and 30% in 2024. That price move mechanically inflates gold’s share of the pie even without new purchases. Strip out the price gain, hold gold at end-2023 prices, and gold and the euro both sit at 16%; Treasuries still lead at 26%. So “gold overtook Treasuries” is real in market-value terms, but it’s mostly a price story, not a reserve-manager rotation story. Central bank gold value is near 4 trillion dollars now, versus about 3.9 trillion in Treasuries held. Dollar assets overall, Treasuries plus agency debt, still around 42% of reserves, still the single largest currency bloc.

China. PBOC started this buying run in November 2022, ran 18 straight months to April 2024, paused May through October 2024, restarted November 2024. June 2026: reserves at 75.44 million ounces, about 2,346.446 tonnes, up 480,000 ounces (14.93 tonnes) from May. That’s the 20th straight month of gains since the restart, and June’s add was the largest single month since October 2023. Gold is still only about 8-9% of China’s FX reserves. China’s total FX reserves: 3,416.26 billion dollars at the end of June. Worth noting: China bought its biggest monthly amount in years during a quarter when gold fell 16%, its worst quarterly drop since 2013. That’s not incidental. Buying into the drop, not despite it.

Three forces that set the price

Gold pricing runs on three layers.

• Currency: dollar rate, real interest rates, central bank flow.

• Financial: ETF flows, futures positioning, exchange inventory, market liquidity.

• Physical: jewelry demand, industrial use, mine supply.

Which layer dominates changes by period.

In the first half of 2026, currency factors ran the show. Hawkish Fed rate expectations plus dollar strength dragged gold down almost 30% from January’s record high of 5589, back under 4000 by July.

Silver moved on a different axis, its industrial exposure and speculative positioning routing the move through exchange inventory and futures market structure instead.

Why silver fell: inventory and liquidity mechanics

Once gold’s logic is clear, silver’s crash reads straightforward.

COMEX and London inventories moved in both directions across the year. January 8, 2026, TD Securities strategist Daniel Ghali opened a short in March silver futures at 78 an ounce, target 40. Thesis: once US tariff policy on critical minerals was clarified, silver stockpiled in New York COMEX vaults on tariff fear would flow back to London through the arbitrage window, and market makers, reloaded with supply, would press prices down.

First half of the year, COMEX inventory did drain toward London at points. From late June, COMEX started rebuilding. July 27: COMEX silver inventory at 331,469,686.88 ounces, about 10,309.87 tonnes. LBMA kept growing in parallel. At the end of June, London vaults held 9,464 tonnes of gold and 28,082 tonnes of silver. London gold vault value at end of June: about 1.2 trillion dollars, roughly 757,145 bars.

The mechanism underneath. Rising COMEX inventory isn’t “more silver exists.” It’s expanded deliverable supply. Futures inventory is settlement and delivery capacity, not global silver supply. More deliverable inventory means less fear of a delivery squeeze. That lowers the cost of opening a short and the cost of market makers providing liquidity. It compresses whatever risk premium had built up from tight-supply fear.

Rising LBMA inventory does the same on the spot side. More silver available for lending, swaps, arbitrage. Market makers manage positions and hedge more efficiently between London spot and New York futures. Deeper market, more sell-side liquidity. So the inventory build doesn’t push price down because “there’s more silver.” It pushes price down because liquidity constraints loosen, and price discovery shifts toward funding costs and rate levels and real supply-demand, away from a delivery-risk premium.

About 70% of silver comes as a byproduct of base metal mining. Supply has no elasticity. That’s the structural floor under the shortage narrative, independent of any short-term price move.

One camp says rising London inventory improved market maker liquidity and pricing power directly. Another camp says the whole correction is dollar strength, rising real rates, profit-taking, and inventory is a side factor, not the driver. Whether inventory is the primary cause of the silver drop stays contested.

Where this leaves the price argument

Central bank reserve behavior still points to gold functioning as a monetary asset, not just a commodity. Ten years of buying supports that read. Gold’s price runs on currency, financial, and physical layers. Silver’s drop comes from several forces stacked together, with COMEX and LBMA inventory shifts changing liquidity conditions and adding downward pressure.

Some academics disagree with the “gold is remonetizing” framing. Their counter: sustained central bank buying reflects reserve diversification against a geopolitical backdrop — the 2022 freezing of Russian reserves is the usual reference point — not a structural change in the international monetary system. Whether buying is enough to prove gold is reclaiming a core role in the global reserve system stays disputed. For anyone trading this, the mechanism matters more than the next price print. Leverage cuts both ways: more upside, more exposure.

What macro changes are taking place in this era?

Great power competition is redrawing its own boundaries. Washington and Beijing have converged on similar lessons from recent events. The scale of attrition in Ukraine showed that even indirect involvement against a mid-tier power pushes the sponsoring nation's capacity to its margin. Applying the same military logic to a peer economy of much greater size produces costs that are effectively incalculable. The trade war supplied a second data point: a large share of tariffs on Chinese goods bounced back into the US as inflation and higher supply-chain costs. "Hurt the enemy a thousand, cost yourself eight hundred" stopped being a figure of speech and became a fiscal and industrial constraint.

A deeper limit comes from the collective floor set by most other countries. If the world’s two largest industrial blocs redirect production toward direct confrontation, the resulting supply chain breaks, food shortages, and public health collapse would hit billions of people and bankrupt dozens of states outright. No country wants to be collateral in that kind of systemic showdown. Under these constraints, both sides have concluded that outright defeat of the other is too costly and too risky, and have redirected effort toward the commanding heights of the next industrial revolution instead. The core logic since industrialization hasn’t changed: national power is measured less by territory and population, more by technological output and energy throughput. Whoever builds the next system first — fully intelligent, fully renewable, fully spatial industry — gets the largest say in defining the next international order.

Two different climbing routes are forming. The US leans on capital pooled through the dollar funding and US equities, concentrating resources on a single line: AI. China mobilizes almost all state capital, trying to reshape the entire economy around what it calls new-quality productive forces. Each side has real advantages. The US: abundant capital, a huge market, dense networks of top universities and venture capital. China: stronger unified action as a single state, a large domestic market, and reach into the middle ground of the global economy. This isn’t a race on one technology. It’s a long test of two different models of organizational innovation, capital mobilization, and social distribution.

Subscribe now

Read more