Nvidia (NVentures + strategic stakes)
Jensen Huang (founder/CEO); NVentures is Nvidia's corporate VC arm, with strategic balance-sheet stakes taken directly by the company.
Nvidia uses its balance sheet to seed the demand side of its own GPU flywheel: it takes equity in the neocloud and infrastructure customers (CoreWeave, Nebius) that buy its chips at scale, plus strategic semiconductor stakes (a ~4% Intel position for x86 co-development) and supply-chain names (Synopsys, Nokia, Coherent). The investing is inseparable from the AI-compute trade because it is engineered to finance and accelerate Nvidia hardware deployment.
- Not open — Not an open-ended vehicle — analysis only, not an available allocation.
Top holdings
~$7.9B (~61% of 13F book) — $5B common-stock purchase at $23.28/sh closed Dec 2025; ~4% stake for x86 CPU co-development. Now Nvidia's single larges
up to $10B — Strategic investment announced Nov 2025 (alongside Microsoft); tied to Anthropic’s ~$30B Azure/Nvidia compute commitment
~$3.66B (~11% of CRWV) — Neocloud GPU renter; Nvidia raised stake to ~47.2M shares in Q1 2026, second-largest 13F holding. Center of the 'circula
EDA / chip-design software — core to the semiconductor design stack.
Strategic stake tied to AI networking / RAN.
~$123M 13F + $2B warrants — AI cloud; 13F shows only ~$123M common, but Nvidia bought $2B of pre-funded warrants in March 2026 (not in the long-equity book)
Recent moves
Q1 2026 13F: increased CoreWeave to ~47.2M shares (~$3.66B); Intel ~$7.9B closed Dec 2025; $2B Nebius pre-funded warrants (March 2026). Exited Arm, Recursion (RXRX), WeRide (WRD), and Applied Digital by Q4 2025/Q1 2026 — note the prompt's Recursion reference is now a sold position. Sources: CNBC, Globe and Mail, 24/7 Wall St.
Loading disclosed positions…
Our take
The cleanest expression of the 'pick-and-shovel finances its own customers' dynamic: Nvidia's equity book exists to make sure GPU buyers have the capital to keep buying. Edge: unmatched visibility into demand. Caveat: the 13F captures only long US-listed equity — it omits warrants (the $2B Nebius position), private/foreign stakes, and the bulk of NVentures' venture portfolio, so the disclosed ~$13B understates true exposure. The 'circular financing' critique (Nvidia funds customers who buy Nvidia chips) is a genuine risk flag, not just commentary.
The disclosed 13F was $18.37B at 3/31/2026, up from $13.10B at 12/31/2025 — but the OpenAI and Anthropic commitments alone are roughly twice that, and neither appears.
Reading the 13F as "Nvidia's portfolio" means looking at the smallest and least strategic third of the book. The sharper gap is off-balance-sheet: Nvidia holds a backstop agreement, initial value ~$6.3B running to April 2032, to purchase CoreWeave capacity CoreWeave cannot sell. That contingent liability triggers in precisely the scenario where the equity stake is also impaired — correlated exposure stacked on one counterparty, disclosed in contract terms rather than in the portfolio. Huang's March 2026 signal that the OpenAI and Anthropic checks are likely the last large ones reads as a genuine strategy change: from equity-financing the labs to simply selling them compute once public markets can fund them.
Thesis
This is not a fund and should not be read as one. Nvidia's equity book is demand-side project finance for its own product cycle: capital placed with the customers, foundries and suppliers whose ability to buy or build GPUs determines Nvidia's forward revenue. The return that matters lands in the semiconductor P&L, not in the portfolio's mark. That makes it structurally unlike Intel Capital-style CVC, which is judged on its own IRR — and it is why the 13F is nearly useless as a read on what Nvidia actually owns.
Three sleeves under one label: (1) balance-sheet strategic stakes — Intel (~$9.5B, ~52% of the 3/31/26 13F), CoreWeave (47.2M shares, ~20%), plus Synopsys, Coherent, Nokia and Nebius; (2) mega private positions — $30B into OpenAI and $10B pledged to Anthropic (announced Nov 2025, alongside Microsoft's $5B); (3) NVentures, a small early-stage arm (~30 deals in 2025) run by a two-person team under Sid Siddeek. Only sleeve 1 is 13F-visible.
Assessment
- Information edge is real and structural: order books, lead times and hyperscaler roadmaps give Nvidia demand visibility no outside allocator can replicate.
- Strategic motive is coherent, not empire-building — capital is aimed at removing the financing bottleneck on its own installed base.
- The Intel stake buys x86 co-development optionality and a hedge against foundry concentration, motives a pure financial investor could not underwrite.
- Deliberate step back from the lab rivalry (Huang, Mar 2026) shows willingness to stop escalating rather than defend a position.
- Circularity is not merely a narrative: Nvidia funds buyers, backstops their unsold capacity, and books the chip revenue — three roles in one transaction chain.
- Counterparty capital intensity is extreme: CoreWeave's Q1 2026 +$2.98B operating cash flow was outspent by ~$7.7B capex, with $536M interest ≈25.8% of $2.08B revenue.
- Disclosure quality is the weakest link: ~$40B of private lab exposure and the NVentures book sit outside any portfolio-level statement.
- No standalone return is reported for the strategic book, so shareholders cannot separate investing skill from chip-cycle beta.
- Concentration at 3/31/26 was extreme — Intel plus CoreWeave ≈72% of disclosed equities.
Record
There is no track record to attribute, and that absence is itself the finding. Nvidia publishes no return, no IRR and no NAV for the strategic book; the only clean data point is the Intel position, marked at $9.48B on 214.78M shares (≈$44.1/share) at 3/31/26 versus the $23.28 September 2025 entry — roughly +90%. Even that is ambiguous: the entry was a lifeline priced in a distressed window, so the gain reflects deal terms and a sector re-rating more than security selection. Everything else is either private and self-marked (OpenAI, Anthropic, NVentures) or so correlated with Nvidia's own end-market that a positive mark and a strong chip cycle are the same event observed twice. The honest read: this book has not yet been tested by a down cycle, and its marks would fall at exactly the moment the operating business does.
Risks & fit
- A neocloud funding stress event impairs the equity stake and triggers the capacity backstop simultaneously.
- Regulatory or accounting scrutiny of vendor-financing arrangements forces restatement or revenue-recognition changes.
- Key-person: the strategy is Huang's; capital allocation of this scale has no visible succession structure.
- An OpenAI IPO re-prices the private marks publicly, removing Nvidia's discretion over carrying value.
- Concentration in Intel ties a large mark to a turnaround Nvidia influences but does not control.
The demand-side-financing read breaks if Nvidia starts reporting the book as a segment with its own return and an independent investment committee, or takes material positions with no procurement or co-development link to its product cycle. The circularity concern is defused if the neocloud counterparties reach self-funding free cash flow (CoreWeave FCF positive on unsubsidised demand) and the capacity backstop expires unexercised. Watch whether deal pace actually falls after Huang's stated pullback — a resumption of $10B+ checks would contradict the shift.
Relevant to readers analysing NVDA as an operating company who need to understand what sits beneath the reported revenue — the extent to which end demand is supported by Nvidia's own capital and contractual backstops. Also useful to anyone using 13F aggregators, as a worked example of how badly that dataset misrepresents a corporate filer. Not a vehicle outside capital can access: NVentures does not take LPs, and the strategic stakes are balance-sheet positions of a public company.
Not a fund: no management fee, carry, LP structure or subscription terms exist. NVentures invests Nvidia balance-sheet capital and does not accept outside money.