OpenAI Seeks Alternatives to Nvidia for Faster Inference — Talks with AMD, Cerebras, Groq
OpenAI has begun diversifying away from relying solely on Nvidia for inference workloads after finding Nvidia GPUs too slow for latency-sensitive tasks such as code generation and software-to-software interactions. The company tested chips with on-chip SRAM (offered by startups like Cerebras and previously Groq) to reduce memory-access latency. OpenAI conducted talks with Groq but those ended after Nvidia signed a large licensing deal with Groq; Groq then shifted focus to cloud software and saw some talent acquired by Nvidia. OpenAI has signed a deal with Cerebras and is exploring AMD GPUs to cover around 10% of future inference needs. Nvidia still powers the majority of OpenAI’s fleet and maintains it offers the best performance-per-dollar; a previously announced potential Nvidia investment of up to $100 billion in OpenAI remains stalled. OpenAI CEO Sam Altman emphasized speed is especially critical for coding models. The move signals broader industry competition as Anthropic and Google use alternative silicon (e.g., Google TPUs) to optimise inference latency.
Neutral
Market impact is likely neutral overall. This news signals increased hardware competition and potential supply diversification for AI inference—but does not immediately threaten Nvidia’s dominant revenue stream, since Nvidia still powers most of OpenAI’s fleet and claims superior performance-per-dollar. Short-term: expect volatility in AI/semiconductor equities as investors reprice competitive risks—Nvidia could see modest pressure, while AMD, Cerebras and smaller chipmakers may get speculative interest. Crypto markets (including SOL, ETH-related infra projects) are unlikely to be directly affected, because this is a cloud/AI infrastructure story rather than crypto protocol news. Long-term: reduced vendor concentration could lower operational risk for large AI providers, potentially enabling faster, more efficient AI services that indirectly support demand for compute-intensive blockchain and layer-2 projects. Historical parallels: past reports of major cloud customers diversifying away from a single vendor (e.g., large cloud migrations, ARM vs x86 debates) caused sector rotation among chipmakers but did not destabilize broader markets. Traders should watch company guidance, capex announcements, and any confirmed deployment percentages; earnings and partnerships will be primary catalysts. Key near-term indicators: Nvidia contract updates, OpenAI disclosures on deployment share (the reported ~10% target), and any formal AMD/Cerebras performance benchmarks.