AI Infrastructure Holdings

Tracked Funds

New Position / Added Reduced Position Unchanged

Fund	AUM	AI Infra Exposure	Top AI Infra Positions	Biggest Adds (Q3)	Biggest Cuts (Q3)	Thesis
Whale Rock Capital	$4.2B	54%	NVDA NVIDIA 18.2% ADD AMD AMD 12.4% ADD MRVL Marvell 9.1% ADD AVGO Broadcom 8.3% HOLD META Meta 6.2% ADD	+ NVDA + AMD + MRVL + META	—	GPU compute moat + custom silicon flywheel. Long NVDA above $900.
Coatue Management	$42B	38%	MSFT Microsoft 14.8% ADD GOOGL Alphabet 10.2% ADD AMZN Amazon 9.4% ADD NVDA NVIDIA 7.1% HOLD ORCL Oracle 4.8% ADD	+ MSFT + GOOGL + AMZN + ORCL	—	Hyperscaler capex cycle + infrastructure enablers. Rotation from app layer.
Tiger Global Management	$35B	29%	META Meta 11.2% ADD AMZN Amazon 8.9% HOLD NVDA NVIDIA 6.4% CUT SNOW Snowflake 5.1% ADD CRM Salesforce 4.2% ADD	+ META + SNOW + CRM	− NVDA	AI application layer king, infra secondary. High-conviction longs with tight stops.
Lone Pine Capital	$22B	34%	MU Micron 13.4% ADD SK Hynix SK Hynix 9.2% ADD NVDA NVIDIA 7.8% HOLD AMAT Applied Materials 5.1% ADD ASML ASML 4.3% HOLD	+ MU + SK Hynix + AMAT	—	Memory bandwidth as next compute bottleneck. SK Hynix + Micron long, NVDA secondary.
Aschenbrenner (Situational Awareness)	N/A	24%	NVDA NVIDIA 9.8% ADD DLR Digital Realty 5.4% ADD EQIX Equinix 4.2% ADD VRT Vertiv 3.8% ADD AMT American Tower 2.6% ADD	+ DLR + EQIX + VRT + AMT	—	AI is a pick-and-shovel play on compute. Invest where the miners (hyperscalers) are forced to buy — GPU racks, power infrastructure, networking. Avoid the application layer hype cycle. Thesis: GPU supply constraint persists through 2025.

Active Theses

Current AI Infra Theses

HIGH

GPU Compute Scarcity

NVDA H100/H200 supply remains constrained through mid-2025. Hyperscalers are paying premiums for guaranteed GPU allocation. Beneficiaries: NVDA, AMD, cloud hyperscalers with preferred access.

Whale Rock Capital Aschenbrenner

HIGH

Power Grid Bottlenecks

AI data center power demand is outpacing grid capacity in Virginia, Texas, and Arizona. Winners: digital infrastructure REITs with interconnection capacity (DLR, EQIX) and power infrastructure plays (VRT, ETSY).

Aschenbrenner

Read the full thesis →

MEDIUM

Hyperscaler Capex Supercycle

Microsoft, Google, Amazon committing $40B+ annually to AI infrastructure. This cycle has 3+ years to run. The order flow feeds NVDA, AMD, and enables the compute layer — MSFT, GOOGL, AMZN are the safe longs.

Coatue Management Tiger Global

MEDIUM

Custom Silicon Inflection

Hyperscalers building proprietary AI chips (Google TPU, Amazon Trainium, Microsoft Maia). This reduces NVDA's monopoly but extends the capex cycle as everyone runs parallel infrastructure. Long-term positive for compute demand.

Whale Rock Capital

HIGH

HBM Memory Bandwidth as Bottleneck

HBM3e supply is the binding constraint on GPU performance, not raw compute. SK Hynix and Micron are positioned to be the bottleneck benefactors. SK Hynix is the highest-conviction memory play.

Lone Pine Capital

Get the thesis update before the next 13F drops.

See the full Q1 2026 AI Infrastructure Research Brief → Read now

Apex Quant agents track fund filings in real-time. When a tracked fund makes a significant move, you'll get the analysis — before it's reflected in headlines.