Watchlist

~15–19% Anthropic stake (10×/year revenue growth) plus AWS as Anthropic's primary compute partner. Anthropic's $9–10B 2025 run rate is not priced into consensus Amazon models.

Major Anthropic investor and Gemini frontier-lab operator. Dario cites cloud economics (3–4 differentiated players) as the profit model; Google is one of the named players. TPU supply chain adds compute leverage.

The central pillar of AI infrastructure. CUDA ecosystem moat, annual architecture cadence (Blackwell → Vera Rubin → Feynman), and ~$100B+ in supply chain commitments. Industry compute growing 3× per year to an estimated 300 GW by 2029.

Sole leading-edge logic manufacturer; 30-year Nvidia partnership. Every GW of AI compute requires TSMC silicon. Jensen is scaling CoWoS packaging at the same rate as logic nodes — TSMC benefits at every layer.

Nuclear power for AI data centers. Jensen: energy is the one bottleneck that cannot be resolved in 2–3 years. Dario's 300 GW industry build by 2029 requires power generation that only nuclear can supply at the needed density.

Silicon photonics for AI cluster networking. Jensen invested directly in the Coherent supply chain and explicitly named it as having 'reshaped' AI cluster networking. Scales with NeoCloud and hyperscaler buildouts.

NeoCloud leader; Nvidia backstopped ~$6.3B and made a direct $2B investment. Named as OpenAI's third infrastructure track alongside Azure and OCI. Converts Nvidia CapEx supply into OpEx rental demand.

Power management and electrical distribution infrastructure for data centers. Structural beneficiary of the multi-trillion annual CapEx build — every gigawatt of AI compute needs power delivery hardware.

Silicon photonics supply chain partner; Jensen made direct investments in Lumentum alongside Coherent. AI cluster networking at scale requires optical interconnects — demand tracks directly with GPU cluster deployments.

HBM memory is now mainstream, not a specialty product. Jensen called out Micron specifically for being an early HBM believer. HBM demand tracks directly with Nvidia GPU shipments.

Azure committed to hundreds of billions in Hopper/Blackwell deployments — 'several years of work' per Jensen. OpenAI equity stake and Azure as primary hyperscaler for OpenAI Stargate legs one and two.

OCI contracted 5–7 GW for OpenAI Stargate buildout. Jensen separately flagged an imminent GPU-native data processing initiative — Oracle's SQL/database business is the largest CPU workload in enterprise, making it a direct migration target.

Data center power and thermal management. Dario's 3× per year compute build puts industry at ~100 GW by 2028 and ~300 GW by 2029 — at $10–15B/GW, cooling and power delivery infrastructure is structurally undersupplied.

NVLink Fusion partnership with Nvidia opens mobile and edge architectures to the Nvidia AI stack. Jensen framed ARM as a distribution channel into compute markets Nvidia cannot reach alone.

Jensen's contrarian agentic-tool thesis: AI agents will run thousands of Cadence instances in parallel for chip design, removing today's human headcount ceiling. Named explicitly alongside Synopsys as a seat-expansion play.

NVLink Fusion reframes Intel as a Nvidia distribution channel into the enterprise x86 installed base. Jensen called Intel a 'large customer' and said it's a 'great win for both.' Not in any analyst's current AI model.

AI-native drug discovery platform. Dario identified pharma as AI's highest per-token value application — molecule redesign tokens worth 'tens of millions of dollars.' FDA pipeline reform advocacy adds a regulatory tailwind.

Computational chemistry + AI for drug discovery. Dario flagged the FDA pipeline bottleneck as the key constraint; companies combining AI discovery with regulatory-path expertise capture disproportionate value.

Jensen's highest-conviction contrarian call: 'The number of instances of Synopsys Design Compiler is going to skyrocket.' Agents as tool-users will multiply today's human-capped seat count. Market is selling SNPS on AI commoditization fears.

Jensen named Snowflake directly in the CPU→GPU data processing migration. Almost entirely CPU-based today. If Nvidia's 'very big initiative in accelerated data processing' lands, Snowflake becomes a massive GPU customer.