NVIDIA Investment Portfolio

NVIDIA's AI model investment is essentially a "fund my customers to buy my GPUs" structure. OpenAI, xAI, and Anthropic all channel a significant portion of their capital toward GPU purchases. NVIDIA put $30B into OpenAI (scaled back from an originally discussed $100B) and roughly $10B into Anthropic (as of 2026-05). Jensen Huang has signaled these are likely NVIDIA's last investments in both before their IPOs.

Investment Thesis: As long as this Circular Revenue structure persists, NVIDIA GPU shipments are on a structural uptrend. However, if the SEC flags this as "self-dealing," there is a $30-50B write-down risk.

Beneficiary Logic: GPU demand → power demand → cooling demand → optical interconnect demand. The entire value chain benefits. Greater alpha may come from supply bottleneck areas (power/cooling/optical) rather than NVIDIA itself.

CoreWeave, Nebius, and Lambda are "GPU-native clouds" distinct from AWS/Azure/GCP. The structure where NVIDIA directly channels surplus GPU capacity to CoreWeave ($6.3B) is effectively an NVIDIA consignment cloud.

Investment Thesis: CoreWeave (CRWV) has 400%+ revenue growth post-IPO but is in a loss-expansion phase. Nebius (NBIS) is a Yandex spin-off serving as a GPU hub for Europe and the Middle East. The core value proposition is the price/performance advantage of dedicated GPU clusters versus hyperscalers.

Risk-Reward: CoreWeave is a mirror of NVIDIA's revenue — if NVIDIA GPU demand weakens, it falls in tandem. Customer diversification as an independent cloud is the key determinant.

In September 2025, NVIDIA invested in three quantum technologies — trapped-ion, photonic, and neutral-atom — within a single week. The message: "Quantum is coming; we just don't know which technology wins." In May 2026 it added France's Alice & Bob (cat qubits) as a fourth modality, widening the hedge.

The thesis: if CUDA-Q (the quantum programming platform) integrates across every modality, NVIDIA keeps its software-platform monopoly into the quantum era. Quantinuum (a Honeywell spin-off and the trapped-ion leader) is closest to commercialization.

Timeline: quantum advantage is expected around 2028-2030. Today we're in the quantum-classical hybrid phase — GPUs handle error correction. NVIDIA's hybrid approach is more realistic than a pure-quantum play.

The GB200 NVL72 rack's performance is gated by GPU-to-GPU bandwidth. To break past the physical limits of electrical signaling, NVIDIA has concentrated $6.7B+ into optics and interconnect.

The thesis: Coherent ($2B) and Lumentum ($2B) are the two pillars of optical transceivers. Ayar Labs integrates optical I/O inside the chip — Co-Packaged Optics (CPO). If the next-gen Rubin architecture mandates CPO, Ayar Labs' valuation could explode.

Marvell ($2B) — the bridge between optics and interconnect: In March 2026 NVIDIA invested $2B in Marvell (8-K, 2 million Series A convertible preferred shares). It resembles the optical bets but the texture is different. Marvell is a custom AI silicon and interconnect powerhouse, and this deal bundles an NVLink Fusion partnership with joint silicon-photonics development — clearing the path for customers to slot Marvell custom XPUs right next to NVIDIA networking and CPUs. Just before the investment, in February 2026, Marvell acquired Celestial AI and XConn to lock in optical and switching capability. It signals NVIDIA pulling even the custom-silicon camp into its optical ecosystem.

The swing factor: the pace of the 800G to 1.6T to 3.2T transceiver transition. Coherent and Lumentum already carry a heavy AI premium, yet this is a rare sector where demand growth justifies the valuation.

Jensen Huang has defined robotics as "the next $50 trillion industry." Investments in Figure AI ($39B), Wayve ($8.6B), Skild AI, and World Labs complete NVIDIA's "Physical AI" pipeline from Isaac Sim → Omniverse → the physical world.

Investment Thesis: Humanoid robots are transitioning from "demo → pilot" in 2025-2026. If Figure AI's BMW factory pilot succeeds, manufacturing robot adoption accelerates. Wayve, a UK-based autonomous driving company, takes an end-to-end learning approach distinct from Tesla FSD.

Value Chain: GPU demand per robot is 100x+ that of a smartphone. The entire chain — sensors (LiDAR) + edge compute (Jetson) + simulation (Omniverse) — is an NVIDIA revenue pathway.

As 1GW-class data centers become reality, power is the biggest bottleneck after GPUs. NVIDIA's energy investments are not purely financial — they are a strategy to remove physical constraints on its own GPU demand.

Investment Thesis: CFS (fusion) targets SPARC tokamak commercialization in the 2030s — unlimited clean energy if successful. TerraPower (SMR) targets first reactor operation in 2028 and is more practical. Redwood Materials covers energy storage through battery recycling.

Beneficiaries: Direct beneficiaries of rising data center power demand are nuclear utilities (CEG, VST) and power infrastructure (POWL, ETN). SMR commercialization could trigger a re-rating across the entire nuclear value chain.

The $1B AI Drug Discovery Lab (5-year) with Eli Lilly signals NVIDIA's transition in biotech from simple GPU supplier to joint R&D partner.

Investment Thesis: The BioNeMo platform integrates molecular simulation, protein structure prediction, and clinical data analysis. Recursion (publicly traded) is a leading AI drug discovery player. Lila Sciences ($350M) is a next-gen AI bio startup.

TAM: Global drug discovery market is $70B+ (2030). If AI improves clinical success rates by 10pp, it saves trillions of dollars in pharma R&D costs. DGX Cloud for Healthcare becomes a key driver of NVIDIA's enterprise revenue.

NVIDIA's acquisition of a 4.4% stake ($5B) in Intel is the biggest surprise in the semiconductor industry. On the surface, it's about NVLink-integrated x86 CPU development, but the true significance is:

1. TSMC Dependency Diversification — If Intel Foundry 18A succeeds, some NVIDIA chips could be manufactured at Intel
2. Two-Front Attack on AMD — Integrated Intel CPU + NVIDIA GPU SoC directly counters AMD's CPU+GPU integration strategy
3. U.S. Government Alignment — Securing a domestic manufacturing partner for CHIPS Act benefits

Investment Thesis: Based on the $23.28 entry price, the unrealized gain has grown to roughly 4x the cost basis (as of 2026-05). The success of the 18A process node is the key variable for both Intel's stock price and NVIDIA-Intel synergy.

Becoming Nokia's 2nd largest shareholder ($1B) reflects the vision to transform telecom infrastructure itself into an AI compute platform. Aerial RAN Computer Pro (ARC-Pro) optimizes networks using GPUs at base stations.

Investment Thesis: The compute required per base station increases 10x+ during the 5G → 6G transition. If NVIDIA supplies GPUs to telecom equipment, millions of base stations worldwide become new GPU demand vectors. T-Mobile US field testing (2026) serves as the PoC.

Expansion: A bridgehead for edge AI computing. Autonomous driving, smart cities, and AR/VR all require ultra-low-latency edge compute, with base stations serving as the hub.

Investments in Scale AI (Meta acquired 49%, $14.3B), Databricks, and Hugging Face represent a strategy to capture "data" and "platforms" — the raw materials for AI models.

Investment Thesis: The Enfabrica $900M+ acquihire secures networking chip talent. NVIDIA has invested across the entire chain: AI training data labeling (Scale AI) → model training infrastructure (Databricks) → model deployment (Hugging Face). These are defensive investments strengthening the CUDA ecosystem's software moat.

Enterprise Pivot: Databricks' Data + AI unified platform is the gateway to enterprise AI adoption. The combination of NVIDIA DGX Cloud and Databricks has the potential to become the de facto standard for the enterprise AI market.