# Leopold Aschenbrenner Just Filed $8.5 Billion of Puts on AI Chips
> We diffed the new Situational Awareness 13F against December: $13.7B disclosed, $8.5B of put notional on AI chips, a 648% CleanSpark add, an Intel flip.
- Author: Barebone Research, Barebone AI
- Published: 2026-05-22
- Canonical: https://barebone.ai/resources/leopold-aschenbrenner-q1-13f-puts-on-ai-chips
- Publisher: Barebone AI (https://barebone.ai)
---
## The Sixth Filing Is a Different Animal
Situational Awareness LP filed its Q1 13F against the May 15 deadline; EDGAR made it public on Monday, May 18. We used Barebone to pull it as it posted, diff all 42 positions against the December book, and reprice both sides at the May 22 close.
Our March breakdown of the fund's first five filings — [zero Nvidia and a $5.5 billion bet that AGI runs on power](/resources/leopold-aschenbrenner-agi-portfolio-power-not-chips) — ended on a specific question: would the Q1 filing show the miner layer surviving the spring?
The miners survived. Almost nothing else stayed still.
The disclosed book went from **$5.52 billion to $13.68 billion** in a single quarter — 29 positions to 42, and 13.5 times the $1.01 billion reported a year earlier. **$8.46 billion** of it — 61.8% — is put options against the AI hardware complex: Nvidia, Broadcom, AMD, Micron, TSMC, ASML, Intel, Corning, the semiconductor index itself, and one name that is not a chipmaker. The CleanSpark stake grew **+648%**. And the Intel calls held through four consecutive filings are gone, replaced by Intel puts — weeks before Intel nearly tripled.
The 24-year-old who spent 2025 refusing to own Nvidia now reports a **$1.57 billion** Nvidia position. It is just pointed down.
The long book barely moved: $3.91 billion of common stock in December, $3.86 billion now. Everything that changed, changed on the short side.
## $8.46 Billion of Puts, Itemized
One piece of mechanics first, because it changes every number below. A 13F reports an option at the market value of the shares it covers, not at what the option cost. A put covering $2 billion of stock shows up as a $2 billion line even if the premium was a fraction of that; strikes and expiries are not disclosed at all. So $8.46 billion is what the puts are written *against*, not what is at risk.
Here is the basket at March 31 marks:
| Underlying | Shares covered | Underlying value | % of book |
|------------|---------------|------------------|-----------|
| SMH (VanEck Semiconductor ETF) | 5,327,900 | **$2.04B** | 14.9% |
| Nvidia | 8,992,300 | $1.57B | 11.5% |
| Oracle | 7,293,000 | $1.07B | 7.8% |
| Broadcom | 3,251,100 | $1.01B | 7.4% |
| AMD | 4,764,100 | $0.97B | 7.1% |
| Micron | 1,727,700 | $0.58B | 4.3% |
| TSMC (ADR) | 1,583,400 | $0.54B | 3.9% |
| ASML | 374,100 | $0.49B | 3.6% |
| Intel | 3,605,400 | $0.16B | 1.2% |
| Corning | 154,600 | $0.02B | 0.2% |
**Oracle is not a chip company.** It is the most aggressively debt-financed data center buildout in the market, and the basket's worst performer while the puts went on — down 24.5% in Q1. A put on Oracle is not a bet against silicon. It is a bet against the *financing* of the buildout, which is a darker idea than shorting a hot chip rally.
**Micron and TSMC appear on both sides.** The fund holds puts on 1.73 million Micron shares and calls on 1.25 million; puts on 1.58 million TSMC ADRs and calls on 1.05 million. Whatever those structures are — collars, strangles, financed spreads — they are not simple bearish bets. The cleaner statement is the names with puts and nothing else: the index, Nvidia, Broadcom, AMD, Oracle, ASML.
**He has run this play before, smaller.** SMH puts were the largest line in the Q2 2025 filing at roughly $570 million; by Q3 the put list had spread to Broadcom, TSMC, Micron, and Nvidia; by December every chip put was gone. This is the same trade at three and a half times the dollar size on the index leg — and more than five times on the Nvidia leg.
A footnote: tiny common slivers sit alongside the puts — including the fund's first disclosed Nvidia stock after six filings of zero: 2,855 shares, about $0.5 million, 0.004% of the book. Plumbing, not conviction. The $6.8 million Infosys put from December survives as the oldest short idea in the book.
## The Long Book Churned Under a Calm Surface
The stock book's total moved 1.5%. Its contents did not.
| Position | Q4 2025 | Q1 2026 | Change |
|----------|---------|---------|--------|
| Intel calls | 20.24M sh underlying | 0 | exited — replaced with puts |
| Lumentum | 1.30M sh | 0 | sold out |
| Coherent | 480K sh | 0 | sold out |
| Tower Semiconductor | 723K sh | 0 | sold out |
| EQT (stock + calls) | 3.18M sh | 0 | sold out |
| Cipher Mining | 10.47M sh | 0 | sold out |
| Hut 8 | 860K sh | 0 | sold out |
| Kilroy Realty | 1.33M sh | 0 | sold out |
| CleanSpark | 1.64M sh | 12.28M sh | **+648%** |
| Bitfarms | 6.90M sh | 19.88M sh | +188% |
| Riot Platforms | 6.17M sh | 11.50M sh | +87% |
| IREN | 8.70M sh | 11.70M sh | +35% |
| CoreWeave stock | 6.10M sh | 7.18M sh | +18% |
| CoreWeave calls | 10.81M sh underlying | 1.81M sh | -83% |
| Bloom Energy | 10.08M sh | 6.49M sh | -36% |
| Solaris Energy | 1.87M sh | 1.11M sh | -41% |
The viral version says he dumped chips and bought miners. The diff is more surgical.
The photonics layer — Lumentum, Coherent, Tower Semiconductor, **$652 million** at December's marks — went to zero in the middle of a quarter where Lumentum nearly doubled. The Intel trade, his most concentrated options bet for a full year, was not trimmed but inverted. EQT is gone, and so is Kilroy Realty, the San Francisco office landlord we called the book's one head-scratcher in March. Resolved: sold.
The miner layer consolidated rather than grew. He exited Cipher Mining ($154 million) and Hut 8, trimmed Core Scientific 10%, and concentrated: CleanSpark's share count grew 648% in a quarter where the stock *fell* 16%, Bitfarms nearly tripled, Riot and Bitdeer roughly doubled, IREN grew a third. Net, at March 31 prices, roughly **$120 million** of new money entered the miner complex — $360 million of buys against $235 million of sells. Miners are now $1.45 billion, the single largest slice of the stock book at 38%.
And the new names are small but on-thesis: T1 Energy ($43.9 million, Texas solar and storage), SharonAI ($18.1 million, AI data centers), HIVE Digital ($6.4 million, another miner).
The popular shorthand for this fund is a barbell: short chips on one side, long the physical layer — power, sites, storage — on the other. The filing mostly supports it, with one refinement: the cleanest conviction long is now SanDisk, $724 million of stock *plus* $389 million of calls and no puts, which makes the barbell long memory too, not just megawatts.
## The Quarter Behind the Snapshot
The put basket looks less like prophecy than weather. These positions were assembled during a drawdown: in Q1 2026 the S&P 500 fell 4.6%, Nvidia 6.5%, Broadcom 10.6%, Oracle 24.5%. Buying puts into that tape was buying insurance while it rained.
The other side of his book was already running: SanDisk rose 168% in the quarter, Bloom Energy 56%, Micron 18% — the memory shortage that began in late 2025 was hardening into what analysts now call a supercycle. Then, after the books closed on March 31, the rain stopped. In the seven weeks that followed, the semiconductor index rose 50%, Micron crossed $700 billion in market value by May 5, and the memory rally kept accelerating into mid-May even on days the broader market fell.
Which sets up the uncomfortable chart.
## The Tape Ran Away From the Puts
Between the snapshot and the May 22 close, every put underlying rallied — three of them by triple digits.
Intel, the position he flipped from calls to puts, is the cruelest line on the board: **+171.6%** since March 31, +224.8% on the year. The photonics names he sold kept running without him — Lumentum +34.7% and Coherent +58.5% since quarter-end. Selling winners early is a survivable sin. Shorting them after is the expensive version.
The same melt-up detonated the long book in his favor:
| Ticker | Name | Q1 2026 | Mar 31 → May 22 | YTD |
|--------|------|---------|-----------------|-----|
| SNDK | SanDisk | +167.7% | **+132.7%** | +522.9% |
| BE | Bloom Energy | +55.9% | +123.3% | **+248.1%** |
| RIOT | Riot Platforms | -2.4% | +98.1% | +93.3% |
| CLSK | CleanSpark | -15.9% | +87.7% | +57.8% |
| CORZ | Core Scientific | +2.7% | +68.9% | +73.5% |
| IREN | IREN | -9.2% | +65.8% | +50.5% |
| CRWV | CoreWeave | +8.2% | +36.2% | +47.3% |
| S&P 500 | benchmark | -4.6% | +14.7% | +9.3% |
Held untouched since March 31, the five largest stock longs would be worth about **$5.7 billion** at the May 22 close, up 94% from $2.95 billion at the snapshot. The barbell did what barbells do: one end paid for the other.
The miner adds deserve a sentence of respect. He bought 10.6 million CleanSpark shares into a falling quarter; the stock is up 87.7% since it ended, and his four largest miner stakes are up 66% to 98% over the same stretch. Within his most contrarian theme, he picked sides — and so far the sides are picking him back.
## Where This Reading Breaks
Everything above is real, and most of it is less conclusive than it looks.
**The $8.46 billion is notional, not risk.** The premium spent is undisclosed — plausibly a tenth of the headline or less. Without strikes or expiries, "he bet the fund on a crash" and "he spent a few points of NAV insuring a violently long book" are indistinguishable, and the second fits a manager whose stock book is wall-to-wall AI infrastructure.
**The snapshot was 48 days stale on arrival.** This fund built and unwound an entire chip-put book once before, inside two quarters. The puts may have been rolled or closed before the filing even posted. 13Fs tell you where he stood in the storm, not where he is sailing now.
**If the puts are still on, they are losing.** Every underlying is up 20% to 172% since March 31. Short-dated crash insurance would have died quietly, costing premium; anything long-dated and directional is underwater unless rolled. The long book's gains very likely dwarf the damage — but "very likely" is doing work no public document can check.
**Do not confuse the $13.68 billion with returns.** $9.8 billion of the headline is option underlying value, which 13F rules inflate by design; the stock book is $3.86 billion. The only public, dated performance number remains the **47%** net gain in the first half of 2025 reported by the Wall Street Journal last August. Nothing newer is public. Disclosed-value growth is inflows plus markets plus notional accounting — not a track record.
And the caveat from March still stands: this is a 24-year-old, six filings into his first fund, who has never managed money through a regime that stayed bad longer than a quarter.
## What the Barbell Is Actually Saying
Read as a whole, the sixth filing is not a bet that AI fails. About 80% of the stock book is miners, power, and data center operators; the call book is memory, storage, and foundry, plus CoreWeave and Bloom. It is a bet that the market pays the wrong layer — the GPU complex and its index carry the crowd, the leverage, and the air, while the constraint keeps migrating to electrons, buildings, and bits of storage.
The watch list writes itself. The Q2 filing, due August 14, shows whether the put basket survived the melt-up or was always a one-quarter umbrella. The Oracle puts deserve separate attention — they are the closest thing in any 13F to a public short on AI's credit risk rather than its valuations. In the miner layer, watch whether his concentration call — CleanSpark, Riot, Bitfarms, IREN in; Cipher, Hut 8 out — maps onto who signs hyperscaler leases this year. And SanDisk, the one name he owns with stock, calls, and no hedge, is the purest tell of what he believes scarcity looks like.
Five quarters ago, this book was six positions and $255 million of power and cooling. The lesson of the first five filings was that he kept moving one step upstream of every consensus before it formed. The sixth shows what he does when the consensus floods the valley he predicted: he keeps the high ground — and sells flood insurance to everyone buying down below.
---
*Data: Barebone | Sources: Situational Awareness LP Form 13F-HR filings (SEC EDGAR, filed February 11 and May 18, 2026), Wall Street Journal (August 2025), CNBC (May 5 and May 11, 2026) | Data as of May 22, 2026*