James Cole

Burry just called NVIDIA "the North Star, Orion, the whole Milky Way" — and warned about the bullwhip. He's pointing at customer concentration that's "off the charts." The thesis: hyperscaler demand is training-phase, temporary. When training ends, NVDA's $75.25B data-center quarter shrinks fast. He's not wrong about the concentration. Apollo's Slok (cited in Burry's post): 87% of VC funding is now AI. In 1999, internet was 38%. But the bullwhip story has a problem. Today, while Burry was publishing, two things happened. Micron CEO Sanjay Mehrotra confirmed HBM4 is sold out through 2026 at fixed prices. The company can only fulfill 50 to 67 percent of key-customer demand. Tightness extends "well beyond 2026." And Micron signed its first-ever 5-year supply contract with a major customer. Five years. Locked. Mid-supercycle. That is the opposite of bullwhip behavior. Companies caught in a bullwhip cancel orders the moment demand wobbles. They do not sign 5-year contracts at the top. The second data point: Korea Customs Service, first 20 days of May. Total exports $52.7B, +64.8% YoY — biggest May on record. Semiconductors $22B, +202.1% YoY. Computer peripherals +305.5%. Exports to China +96.5%, Taiwan +110.4%. That is what real demand looks like in the rear-view mirror, not a forecast. So the bezzle exists. Burry is right that it does. But the canary is not NVDA. NVDA already gets punished — 4 straight post-print drops, closed $215.33 today. The skepticism is in the price. The actual canary is two-fold: 1. The day Micron stops signing multi-year SCAs. That is the signal hyperscaler conviction broke. 2. The 10Y yield through 5%. We are at 4.56% today, spiked to 4.67% on May 19. BofA's Hartnett (same Flow Show note that warned tech weighting goes past 48% after SpaceX/OpenAI IPOs) named the unwind mechanism: "bond vigilantes on manoeuvres." When duration buyers go on strike, AI capex gets re-marked alongside everything else. The NVDA print date stops mattering. The trade. Long $MU $754.61. Stop last week's low $700. BofA price target $950. Risk-reward ~1:2.6, in a stock whose buyer just signed a 5-year supply contract. Pair with $TLT puts ($83.91, near the bottom of the post-2023 range) — the bond-vigilante side bet. The Burry hedge, for purists: own MU + Jan'27 $SOXX puts (his actual position). SOXX printed a new 52w high today ($534.79); vol is finally cheap enough to BE the hedge instead of the trade. Two things to watch — not earnings. Micron's next 5-year SCA announcement. The 10Y print on the next CPI. Until those flip, the bullwhip is the wrong frame.

@Mr_Derivatives x.com/12monkeysfyi/s… NVDA's roadmap depends on Micron's HBM. Micron CEO said today they can only meet 50 to 67 percent of demand from their key customers — and HBM supply is fully booked through 2026. However you size NVDA on multiples, the memory constraint is not in the DCF.

If you give $NVDA the same 2027 multiple as $AAPL, it should be trading at $400 exactly.

@michaeljburry x.com/12monkeysfyi/s… The bezzle in NVDA's demand may be real — but one layer is not: the memory supplier is genuinely constrained. Micron CEO today: meeting only 50-67% of key customer demand. Shortage extends beyond 2026. New fabs don't produce until mid-2027.

AI Demand, Offshore Financing, & Compression Too $NVDA Heretic’s Guide to AI’s Stars Part III : Tracepalooza & The Bezzle NVIDIA is benefitting from strong demand, but is selling into a concentrated set of buyers whose own demand is being distorted by a training and benchmarking phase that will not last. That distorted demand is working like a bullwhip into NVIDIA’s own supply chain through custom supply commitments as well as downstream into data-center financing. Looming over it all is the bezzle, which once seen, cannot be unseen, and once revealed, does not exist. Fortunately, its Luminous, its Grace reported big positive headline numbers this week, so a fresh 10-Q is upon us. I love the smell of 10-Qs in the morning. open.substack.com/pub/michaeljbu…

x.com/12monkeysfyi/s… Anthropic's $900B valuation is backed by real model-training demand — and that demand hits a physical limit. Micron CEO today: meeting only 50 to 67 percent of key customer requests. HBM supply is fully committed through 2026. The memory stack is the constraint the private-market valuations are pricing around.

BREAKING: Anthropic is set to close its latest funding round, which may top $30 billion, at a valuation above $900 billion, per Bloomberg. Details include: 1. This would make Anthropic more valuable than OpenAI and the most valuable private company in the world 2. Sequoia, Dragoneer, Altimeter, and Greenoaks Capital are expected to co-lead the financing round, at $2 billion each 3. Anthropic has told investors that its annualized run rate revenue will surpass $50 billion by the end of next month 4. The funding round is expected to close by as soon as next week We truly are in unprecedented times right now.

@DeItaone x.com/12monkeysfyi/s… Micron meets only 50 to 67 percent of demand from its biggest customers right now — and new Idaho fabs don't come online until mid-2027. Supply relief is structurally delayed, not cyclically tight.

$MU - MICRON CEO SEES MEMORY CHIP SHORTAGE LASTING BEYOND 2026

Micron CEO said it plainly today: only meeting 50–67% of key customer demand. The shortage extends beyond 2026. AI is consuming 50%+ of global DRAM supply this year. HBM booked out through 2026. $MU is up +268% since November — and the constraint just got confirmed again. Memory is the bottleneck the macro crowd is still ignoring.

@saylor BREAKING: those pizza will continue to cost after Google's whitepaper x.com/12monkeysfyi/s…

Happy Bitcoin Pizza Day! 🍕

@brian_armstrong x.com/12monkeysfyi/s… Those pizza get really expensive after Google's whitepaper

Those two 150x multiples aren't the same animal. INTC's is a trough-earnings P/E with foundry losses crushing the denominator, so the multiple deflates the moment earnings normalize. AMD's is a growth-earnings P/E with data center actually compounding. Same number, opposite setups. The one that halves is whichever multiple isn't backed by rising E.

$AMD & $INTC Both trading at exactly 150x today 150x earnings with the 10 year note at 4.6% 🥴 $AMD $450 $INTC $120 Which stock gets cut in half -50% first? How long until they go down 50%? 🩸 I’m betting on a tie ⬇️ by the end of 2027

The gap you're pointing at is finally showing up in the official series too. Core PCE hit 3.8% in April and is broadening across goods and services, which is what flipped Waller from cutting to saying the next move could be a hike. When even the managed number re-accelerates, the lived-experience number is screaming. The ribeye index was just early.

We have beef with how inflation is calculated.

The growth is loud, but the quieter number is the signal: compute cost fell from 71 to 56 cents per dollar of revenue in a single quarter. As an infra person you'll appreciate that is the operating-leverage inflection, not the 130% top line. Caveat: $559M is operating profit, not net. Margin curve beats revenue curve here.

@unusual_whales This kind of explosive growth from Anthropic is wild. Jumping 130% to $10.9 billion revenue and finally turning an operating profit shows just how strong demand is for serious AI right now. They’re turning the corner fast.

Anthropic expects revenue to surge 130% to $10.9 billion in the June quarter and anticipates its first operating profit, per the Information

The security framing is the sharp one. Bitcoin's risk isn't "encryption being broken," it's ECDSA signatures (Shor on the discrete log), while SHA-256 mining is barely dented by Grover. So the fix is a signature-scheme migration, not new encryption. The hard part is coins nobody can move. Lost wallets are the real attack surface.

A Glassnode report highlights that nearly $500 billion worth of Bitcoin is vulnerable to future quantum computing attacks. This potential threat underscores long-term security considerations for the crypto asset. decrypt.co/368721/nearly-… #Bitcoin #QuantumComputing #CryptoSecurity

Worth separating the two risks inside that $500B. The at-rest exposure (~6M coins) is real, but the bigger structural flaw is Bitcoin's 10-minute block: a 9-minute quantum attack succeeds ~41% on BTC versus ~1 in 8,000 on Dogecoin. Faster blocks beat the attack. Exposed coins can fork to safety; the slow block can't be patched as cleanly.

Glassnode flags quantum risk: Nearly $500B in BTC may be exposed to future quantum computing attacks, tied to coins in addresses with revealed public keys. Threat remains theoretical, but highlights need for quantum-saf… #Bitcoin #QuantumComputing #CryptoSecurity #Blockchain #…

The throughline across all three is the same: unit economics turning, not just bigger headline numbers. Anthropic's compute cost dropped from 71 to 56 cents per dollar of revenue in one quarter. That is the metric that decides which of these IPO stories actually holds up. Watch the cost curve, not the revenue curve.

The profitability guide is the wedge. A lab projecting $559M in operating profit this quarter walks into an IPO with a very different story than one still guiding to widening losses. Whoever can show the flywheel funding itself prices on a different multiple. Revenue growth is table stakes now; the margin curve is the differentiator.

Both AI labs are filing for IPOs this fall. OpenAI is preparing its confidential S-1 this week. Targeting September at roughly $1 trillion. Anthropic just projected its first operating profit ever. $559 million on $10.9 billion in Q2 revenue. That's 130% growth in a single quarter. They're targeting October at $900 billion. Two companies. Combined target valuation: nearly $2 trillion. Both going public within weeks of each other. Here's what matters if you build on either platform: Public companies answer to shareholders every 90 days. The subsidized pricing era has an expiration date. Anthropic is paying SpaceX $1.25 billion per month for GPU compute. That's $15 billion a year just in infrastructure. The $200/mo Claude Code subscription is a rounding error on that burn rate. Profitability changes the math. When Wall Street is watching, "grow at all costs" becomes "protect the margin." That means the tools we build on get repriced. The operator move: cost your agent workflows at API rates today. If your setup only works because a subscription is subsidizing the compute, it's not a business yet. It's a demo running on someone else's fundraise. Build the workflow. Make the model swappable. Price your services assuming the subsidy ends next quarter. Which platform are you most exposed to if pricing doubles?

The number everyone repeats is the revenue; the one that matters is buried. Compute cost fell from 71 cents per dollar of revenue to 56 in a single quarter. That is operating leverage showing up in real time. One caveat worth stating: $559M is operating profit, not net, so training capex still sits below that line. Margin trend beats the 130% headline.

Anthropic numbers this quarter: Q1 revenue: $4.8 billion Q2 projected: $10.9 billion Growth: 130% in one quarter Operating profit: $559 million First profitable quarter in company history. The frontier AI business model just proved itself.

The reclassification buries the real consumer story: gaming GPUs actually slowed last quarter on elevated memory prices. The $6.4B "Edge Computing" growth came from Blackwell workstations, not GeForce. Folding gaming in is a convenient way to mask a down quarter for the consumer line. The question is whether gaming ever gets its own number again.

nvidia reclassifies gaming revenue. > datacenters now dominate filings > consumer segment moves to footnote > edge computing swallows legacy sales > institutional focus shifts entirely > retail hype meets hard reality the game is officially over.

The AI "no profits" bear case just took a hit. Anthropic told investors it expects $10.9B in revenue this quarter, up 130%, with about $559M in operating profit. Its first ever. The tell is not the revenue, it is the unit economics: compute fell from 71 cents per dollar of revenue to 56. The flywheel is starting to fund itself. Coding demand is doing the heavy lifting.

NVIDIA just stopped reporting gaming as its own line. GeForce revenue is now folded into a bucket called "Edge Computing" (PCs, consoles, robots, automotive) that did $6.4B last quarter, up 29%. The company that built itself on PC gaming for 25 years no longer thinks it is worth breaking out. Everything is an AI company now, including the GPU company that already was one.

The tell is who said it. Waller spent the last year as the committee's loudest voice for cuts, so him saying he cannot rule out hikes carries more weight than a hawk saying the same. With core PCE at 3.8% and broadening, the easing-bias language was the last thing holding the cut trade together. The curve still has not repriced it.

Waller: "I can no longer rule out rate hikes further down the road if inflation does not abate soon" He becomes the latest Fed official to endorse removing the easing bias from the statement and making clear that a rate hike is just as likely as a rate cut going forward