Here Are the Companies in Nvidia's $99 Billion Artificial Intelligence (AI) Portfolio. These 3 Are the Best of the Bunch

Nvidia reported a roughly $99 billion investment portfolio at the end of its fiscal Q2, comprising $47.7 billion in public equities, $47.9 billion in non-marketable securities, and $3.3 billion in equity-method investments. Its AI ecosystem investments include nearly $50 billion in frontier labs, led by $30 billion in OpenAI, up to $10 billion in Anthropic, and $5 billion in Safe Superintelligence. The portfolio also targets AI infrastructure through CoreWeave, Coherent, Marvell, Lumentum and neocloud providers, positioning Nvidia as both a chip supplier and capital provider to the AI buildout.

Analysis

The investee pattern matters less as a stock-picking signal than as evidence that NVDA is increasingly underwriting demand for its own ecosystem. This can extend the AI capex cycle by solving financing constraints for GPU renters and model developers, but it also raises the risk that reported accelerator demand is partially vendor-financed rather than fully end-customer funded. The key second-order metric is not NVDA shipment growth; it is whether neocloud utilization, pricing, and operating cash flow rise fast enough to service the infrastructure financing layered beneath those shipments.

COHR, LITE, and MRVL have the cleaner incremental read-through because scale-out networking and optical interconnect content rises with cluster density regardless of which cloud operator owns the GPUs. COHR offers the most direct upside torque if optical supply remains constrained, while MRVL's custom-silicon and connectivity exposure is more dependent on a small number of hyperscaler programs and design-win timing. Over 6-18 months, improved optical availability could become a cap on component margins even while unit volumes remain strong; avoid treating all AI-infrastructure suppliers as equivalent GPU-beta.

CRWV and NBIS are the fragile links in this structure. Their equity upside is highly convex to sustained GPU scarcity and high utilization, but their economics can deteriorate quickly if hyperscalers internalize capacity, inference becomes more efficient, or financing costs rise before contracted revenue converts to cash. Contrarian view: the market may be underpricing NVDA's credit/counterparty exposure while overpricing the signaling value of its minority stakes; strategic investment alone does not validate public-market entry points.

Key Decisions for Investors

• Maintain a core NVDA long only while monitoring disclosure of customer financing, investment gains/losses, and receivable or commitment growth at the next two earnings prints. Reduce exposure if revenue growth remains strong but operating cash conversion materially lags, as that would support the vendor-financing/pull-forward risk case.
• Prefer a 6-12 month long COHR / short CRWV pair rather than outright AI-beta: COHR benefits from broad cluster interconnect intensity, while CRWV bears greater utilization, refinancing, and customer-concentration risk. Size modestly; invalidate the pair if CRWV demonstrates sustained positive free cash flow and contracted backlog converts without incremental leverage.
• Use MRVL as a watch-list long, not an immediate recommendation, pending confirmation of incremental custom-silicon or optical-connectivity revenue in guidance. Enter only after a guidance raise tied to production ramps rather than partnership announcements; risk is program concentration and delayed hyperscaler qualification.
• Avoid chasing INTC or NOK on strategic-investor optics. Require independently visible evidence of foundry customer commitments for INTC or material AI-RAN order conversion for NOK before assigning AI-infrastructure multiples.
• Set an alert around neocloud financing conditions over the next 1-3 months: widening high-yield spreads, equity raises at discounts, falling GPU rental rates, or utilization weakness would be an early signal to trim NVDA-adjacent infrastructure exposure and add to the CRWV short leg.