Aschenbrenner Just Raised $8 Billion and Bought NVDA, AMD, TSM, AVGO, MU and ASML — Same Week Baker Sold All of Them

The Lead

Two AI specialists, both filed three days late on Monday 18 May, both manage concentrated tech books, both publicly known as serious AI thinkers. Their Q1 2026 portfolios are diametrically opposed.

Leopold Aschenbrenner (Situational Awareness LP, AUM Q4 $5.5B → Q1 $13.7B = +148%) raised approximately $8 billion of fresh capital in Q1 and concentrated it entirely on the merchant silicon stack: NEW positions of NVDA $1.57B, AMD $989M, TSM $897M, AVGO $1.0B, MU $1.01B, ASML $500M, ORCL $1.07B. Eight figures of new conviction across the entire AI mega-cap layer he previously avoided. He also trimmed his existing secondary-beneficiary book (Bloom Energy -34%, CoreWeave -47%) and added to crypto-pivot neoclouds (CleanSpark +648%, Riot +86%, Bitfarms +188%).

Gavin Baker (Atreides Management, AUM Q4 $8.18B → Q1 $5.0B = -39%) shrank his book by $3.18B and sold the same names Aschenbrenner was buying. NVDA cut -77%. Microsoft exited entirely. Coherent -70%. Ciena -53%. Micron -47%. QQQ ETF hedge cut -60% (from ~$2B to $808M). Net mega-cap exposure down ~$5B. He recycled the capital into AI connectivity ICs — Astera Labs +109% to $369M (now his largest single stock position), Credo Technology NEW $102M, plus diversified additions in Zoom ($106M NEW), Palo Alto Networks ($90M NEW), Vistra ($79M NEW), HubSpot, Synopsys, AeroVironment.

One AI specialist sees a generational compute build cycle that will keep paying the mega-cap silicon names regardless of share-shifts. The other sees a margin-compression cycle where the picks-and-shovels names (Astera, Credo) will outperform the headline GPU names. NVDA Q1 FY26 earnings on Wednesday 28 May is the verdict moment.

Falsifiable form: if NVDA’s Q1 print shows revenue ≥$50B with Data Center segment ≥40% YoY growth AND guidance maintains $60B+ for Q2 FY26, Aschenbrenner’s bet was right. If NVDA misses or guides flat-to-down, Baker’s de-risk timing was right. Either outcome is inside the next 10 days.

Threads

Two AI specialists filed late, took opposite sides of the same trade

The Q1 2026 13F deadline was Friday 15 May. Both Aschenbrenner and Baker filed Monday 18 May — three days late, no NT extensions visible. Late deadline-day filers in the same week + same direction (late) + opposite directional bets is unusual.

Position-by-position comparison on the disputed names:

Ticker	Aschenbrenner Q1	Baker Q1	Q-over-Q delta
NVDA	NEW $1,569M	-77% to $218M	Aschenbrenner +$1.57B, Baker -~$735M
AMD	NEW $989M	trimmed	Aschenbrenner +$989M, Baker net out
TSM	NEW $897M	not held	Aschenbrenner +$897M solo
AVGO	NEW $1,006M	not held	Aschenbrenner +$1.0B solo
MU	NEW $1,012M	-47% to $257M	Aschenbrenner +$1.01B, Baker -~$230M
ASML	NEW $500M	not held	Aschenbrenner +$500M solo
ORCL	NEW $1,073M	not held	Aschenbrenner +$1.07B solo
MSFT	not held	EXITED $88M	Both bearish — neither owns MSFT
GOOG	not held	held flat	Aschenbrenner sat out, Baker held

Aggregated: Aschenbrenner deployed approximately $7B into mega-cap merchant silicon NEW in Q1 2026, while Baker reduced his mega-cap silicon exposure by approximately $5B in the same quarter.

Both are AI bulls in their public commentary. The disagreement is on how the next 12-18 months play out:

Aschenbrenner’s thesis (inferred from positioning): the AI compute demand curve is so steep that the merchant silicon names get paid regardless of which architecture wins. NVDA loses share to AMD warrants but unit volumes still grow. TPU loses share to AMD but TSM fabricates both. The aggregate AI capex flow (Issue 03 thesis at $400B+ annualised) is bigger than any single share-shift story. Buy the whole stack and let the volume tide lift everyone.

Baker’s thesis (inferred from positioning): the share-shift stories matter more than the volume story over the next 4-6 quarters. NVDA loses dollar share faster than total demand grows (specifically to AMD via the warrants from Issue 04). Mega-cap multiples compress. The alpha is in the picks-and-shovels layer (AI connectivity ICs like Astera and Credo) where competition is less intense and customer concentration is lower. Sell the mega-caps, buy the bottleneck specialists.

Both theses can be partially right. NVDA’s print on 28 May is the cleanest near-term test of which one is more right.

Three Tiger Cubs are de-risking AI mega-cap exposure in unison

Aschenbrenner is the loud outlier going the other direction. What’s been quietly accumulating across the rest of our tracked universe is a coordinated de-risk pattern — three Tiger Cub-style funds making the same directional bet that AI mega-cap exposure should be reduced:

Fund	Mega-cap action	$ magnitude
Coatue (Laffont)	MSFT cut -57%	-$1.58B
Altimeter (Gerstner)	GOOG exited entirely + MSFT -29%	-$162M GOOG + -$179M MSFT
Atreides (Baker)	MSFT exited + NVDA -77% + QQQ -60%	-$88M MSFT + -$735M NVDA + -$1.2B QQQ

That’s three of nine tracked funds all reducing mega-cap AI exposure in Q1 2026, with different specific names but the same directional bet. None of them are exiting AI as a theme — they’re all rotating into more specific sub-themes (Coatue into EQIX + ASML, Gerstner into ARM, Baker into Astera + Credo + Vistra).

The countervailing signal: Berkshire added $10B to GOOG (Issue 06), Aschenbrenner added $7B+ to the full mega-cap stack (this issue), LonePine added $3B to the secondary infrastructure layer (Issue 05). So the de-risk isn’t unanimous — but the concentrated growth-fund consensus is meaningfully off the mega-caps.

This is the cleanest pre-NVDA-earnings positioning signal we have: the funds closest to the AI investment thesis are split, with the most-publicly-AI-bullish names taking opposite sides. That kind of dispersion typically resolves with a big move — direction depends on the print.

Aschenbrenner raised $8B and concentrated it on the merchant silicon stack

The mechanics of Aschenbrenner’s Q1 2026 expansion are worth their own thread. Situational Awareness LP went from a $5.5B portfolio of secondary AI beneficiaries (Bloom Energy, CoreWeave, IREN, APLD, CORZ, plus crypto-pivot neoclouds) to a $13.7B portfolio that added the entire merchant silicon mega-cap stack as NEW positions.

Q1 2026 positions sorted by size:

#	Ticker	$M	%	Action
1	VanEck ETF Trust	2,053	15.0%	NEW (likely SMH semiconductor ETF)
2	NVIDIA	1,569	11.5%	NEW
3	SanDisk	1,113	8.1%	+66% (existing add)
4	Oracle	1,073	7.8%	NEW
5	Micron	1,012	7.4%	NEW
6	Broadcom	1,006	7.4%	NEW
7	AMD	989	7.2%	NEW
8	Bloom Energy	934	6.8%	-34% (trimmed)
9	TSM	898	6.6%	NEW
10	CoreWeave	697	5.1%	-47% (trimmed)
11	ASML	500	3.7%	NEW
12	IREN Limited	401	2.9%	+34%
13	Core Scientific	389	2.8%	-10%
14	Applied Digital	320	2.3%	+19%
15	Intel	168	1.2%	-81% (near-exit)

The new merchant silicon basket (NVDA + AMD + TSM + AVGO + MU + ASML + ORCL + the VanEck ETF) = $8.1B = 59% of the entire book. The pre-existing crypto-pivot / power / data-centre infrastructure book was trimmed from ~$5.5B (100% of Q4 book) to roughly $3.5B (~26% of Q1 book) — same dollar exposure roughly, but a much smaller share of total.

The reweighting tells you Aschenbrenner now believes the upstream silicon stack is the bigger beta to the AI compute build than the downstream secondary beneficiaries he previously concentrated in. He’s not abandoning the secondary beneficiaries — he’s adding the primary layer on top of them.

Two specific observations worth flagging:

• VanEck ETF Trust at $2.05B as his largest position is unusual for a concentrated fund. Likely the VanEck Semiconductor ETF (SMH) used as a basket-deployment vehicle while individual names get sized. May convert to individual names over Q2-Q3.
• Bloom Energy and CoreWeave trims are NOT thesis-rejection — they’re portfolio-balancing as the book scaled 2.5x. Both remain his 8th and 10th-largest positions respectively.

Baker shrank his book 39% and pivoted into AI connectivity ICs

Baker’s mirror-image move was equally deliberate. Book size $8.18B → $5.00B in one quarter is the most aggressive de-risk in our tracked universe this cycle.

Recap of the substantive moves (from earlier analysis):

• NVDA cut -77% (now $218M, 4.4% of book)
• MSFT EXITED entirely ($88M to zero)
• QQQ -60% (~$1.2B removed — likely the directional hedge unwinding)
• Coherent -70%, Ciena -53%, Lumentum -30%, Micron -47% (optical and memory stack trimmed)
• Astera Labs +109% to $369M — now his largest single-stock position (7.4% of book)
• Credo Technology NEW $102M (AI connectivity IC complement to Astera)
• Vistra Corp NEW $79M (power utility — joining Tepper’s VST+NRG positioning)
• Eleven additional new positions in defensive / AI-applied SaaS / cybersecurity / defense names

The Astera + Credo combo at $471M (9.4% of book) is the single most-focused thesis expression in Baker’s filing. Both companies make AI connectivity infrastructure — PCIe retimers, smart cable modules, high-speed interconnects between GPUs and racks. They benefit from rising bandwidth requirements as model sizes grow, regardless of which silicon vendor wins the GPU race. It’s a hedged play on the same AI capex flow that Aschenbrenner is expressing via the full mega-cap stack.

Astera convergence check across the now-11-fund universe: Baker’s $369M is currently a solo position among tracked funds. No other fund holds Astera materially. Same for Credo — solo at $102M. If 1-2 other funds initiate Astera or Credo in Q2, the convergence-promotion threshold fires.

NVDA Q1 FY26 earnings preview — the verdict moment on five issues of thesis

NVIDIA reports Q1 FY26 (calendar Feb-Apr quarter) on Wednesday 28 May 2026 after market close. This is the cleanest near-term test of the entire thesis arc we’ve built across Issues 03-07.

What the print tests:

Issue	Thesis	NVDA print test
Issue 03	Combined hyperscaler capex annualises at $400B+, not slowing	NVDA Data Center revenue should be ≥ $38B Q1 (vs $35B prior) if the capex thesis holds
Issue 04	AMD warrants imply NVDA loses share to OpenAI + Meta over 4-6 quarters	NVDA customer concentration commentary on Q1 call should acknowledge competition; if silent, share-shift is overstated
Issue 05	Lone Pine et al. are pivoting away from mega-cap AI exposure	NVDA performance vs implied positioning tells us whether the funds were early or wrong
Issue 06	Berkshire’s GOOG add is the contrarian Buffett signal	Indirect — NVDA’s beat/miss determines whether mega-cap AI is over-priced or appropriately priced
Issue 07	Aschenbrenner (long NVDA $1.57B NEW) vs Baker (short NVDA -77%) on the print	Direct verdict

Consensus expectations (per Bloomberg / FactSet as of week prior — verify before quoting externally):

• Revenue: $48-52B (vs $44B Q4 FY25)
• Data Center revenue: $40-43B (the line that matters)
• Gross margin: 75-76%
• Q2 FY26 revenue guidance: $55-60B
• Capex / supply commentary: critical for the share-shift narrative

What would change next week’s reading:

• A NVDA beat with Data Center >$43B and Q2 guidance >$60B = Aschenbrenner thesis confirmed, Baker timing wrong
• A NVDA in-line print with cautious Q2 guidance = Baker thesis confirmed, Aschenbrenner deploying into a near-term top
• A NVDA miss with Data Center <$38B or Q2 guidance <$50B = the most-bearish reading; both funds wrong but Baker less wrong; Berkshire’s GOOG add looks visionary

Watchlist Updates

Ticker	Direction	Driver	Lab-calibrated reading
NVDA	conflicted	Aschenbrenner +$1.57B NEW vs Baker -77% — most disputed name in our universe	Holding pattern until 28 May print
ASTERA LABS (ALAB)	bullish	Baker now largest single stock position at $369M (+109%)	Solo conviction in our universe — promotion candidate if Q2 13Fs show 1-2 more initiating
CREDO (CRDO)	bullish	Baker NEW $102M; complement to Astera	Solo conviction — same as Astera
VST	bullish (reinforced)	Baker NEW $79M; Tepper holds $304M; power-utility thesis broadening	Add to watchlist promotion queue
MSFT	bearish (reinforced)	Three funds reduced/exited (Coatue -57%, Gerstner -29%, Baker exit)	Lab caution: bearish setups underperform historically — treat as hypothesis
Crypto-pivot neoclouds (IREN, APLD, CORZ, CleanSpark, Riot, Bitfarms)	bullish	Aschenbrenner added across the board	Convergence is single-fund (Aschenbrenner) but at scale
VanEck SMH ETF	bullish	Aschenbrenner $2.05B NEW (his largest position)	Basket-deployment proxy — not actionable as a single-name signal

The Week Ahead

The single event that matters: NVDA Q1 FY26 earnings, Wednesday 28 May after-close (Thursday 29 May AEST). The reaction in the after-hours session and Thursday US trading will determine whether the Issue 03-07 thesis arc was right, the Aschenbrenner deployment is timely, or the Tiger Cub de-risk was prescient.

Other watchlist events:

• Tuesday 19 May (today/yesterday): No major scheduled events
• Friday 22 May: Watch for any further deferred 13F filings (Form NT extensions still possible)
• Monday 26 May: US Memorial Day, US markets closed
• Wednesday 28 May: NVDA earnings (the big one)
• Thursday 29 May: NVDA earnings reaction trading day; CRM Q1 expected (CRM as proxy on enterprise software AI monetisation)

Issue 08 ships Sunday 24 May with the pre-NVDA-earnings positioning analysis. Issue 09 ships Sunday 31 May with the post-NVDA-earnings reading and the verdict on the Aschenbrenner-vs-Baker bet.

Methodology and disclosures

Filings Intel Digest is a bona fide financial publication. Nothing here is personalised investment advice. The editor may hold positions in companies discussed; current positions and policy are at /about. All claims are sourced to publicly filed documents linked inline. Backtested or historical figures are direction-adjusted and calibrated against the signal lab; calibration is partial — treat any forward statement as a hypothesis, not a forecast.

Frozen as of close-of-trading 15 May 2026 (Friday). Both Atreides and Situational Awareness LP filed 13F-HR on Monday 18 May 2026 (3 days late, no NT extension visible). Position values reflect Q1 2026 quarter-end (31 March 2026) prices, not current.

Aschenbrenner Q1 2026 13F-HR accession: 0002045724-26-000008 Atreides Management Q1 2026 13F-HR accession: 0001777813-26-000006

The Aschenbrenner-vs-Baker thesis dispersion is the cleanest pre-NVDA-earnings positioning signal in our tracked universe. The print on 28 May resolves it.