Hyperscaler custom-silicon programs collectively encircle Nvidia at the in-house-deployment layer — what changes when AWS, Google, and Microsoft all ship credible internal alternatives
Nvidia retains dominant position in the customer-facing AI infrastructure layer at $4.5T market cap. But the hyperscaler in-house chip programs (Trainium, TPU, Maia) collectively erode Nvidia's first-party-workload allocation. The H2 2026 picture: Nvidia for customer-facing, hyperscaler-internal for internal workloads. The encirclement is real even if not existential.
The hyperscaler custom-silicon encirclement compounds across AWS Trainium, Google TPU, and Microsoft Maia programs. Through H1 2026 all three hyperscalers reached production-deployment maturity with their in-house chips for first-party workloads. The strategic pattern: hyperscalers use in-house silicon for foundation-model training and infrastructure inference, source Nvidia for customer-facing services (Bedrock, Azure OpenAI, Vertex AI).
What Nvidia loses vs retains
Nvidia loses the first-party allocation share — AWS doesn't buy Nvidia for Bedrock's foundation-model training, Google doesn't for Gemini training, Microsoft doesn't for MAI training. Nvidia retains the customer-facing share — when AWS Bedrock customers want inference, Nvidia powers it; same for Azure OpenAI; same for Vertex AI. The bifurcation reduces Nvidia's growth velocity without reversing the underlying competitive dominance.
The competitive read across vendor positions
AMD's competitive response (Taiwan investments + Arm-PC chip development) addresses the PC-market flank. AMD's record server CPU share creates CPU-GPU coupled procurement leverage. Together with the hyperscaler encirclement, Nvidia faces multi-front competitive pressure that didn't exist 18 months ago. The pressure is real even if Nvidia's $4.5T market cap suggests it's not yet priced in.
The 24-month outlook
Two scenarios are coherent through 2027-2028. (1) Stable bifurcation: Nvidia dominates customer-facing AI infrastructure, hyperscalers dominate first-party workloads, AMD takes meaningful share in both. (2) Encirclement deepens: hyperscalers expand custom-silicon to customer-facing services, AMD compresses the Nvidia valuation premium, Nvidia growth slows. Which scenario materializes depends on whether hyperscaler chip programs can support customer-facing service quality at Nvidia parity — a question that won't have a definitive answer for 12-18 months.
Windows News — AI Chip Wars 2026: Amazon, Google, and Microsoft Surround Nvidia with Custom Silicon → · Al Jazeera — Nvidia unveils new chip to bring AI directly to personal computers →