OpenAI's compute deal with Cerebras, alongside deals with AMD and Nvidia, shows that hyperscalers are aggressively diversifying their AI chip supply. This creates a massive opportunity for smaller, specialized silicon teams, heralding a new competitive era reminiscent of the PC wars.
Related Insights
Facing Nvidia's near-total capture of AI data center revenue growth since 2022, AMD CEO Lisa Su made a "bet the farm" move. By granting OpenAI warrants for up to 10% of AMD, she aims to secure a critical design win for their next-gen chip, validating it as a viable competitor to Nvidia.
The AI infrastructure spending boom will continue robustly for at least two more years, creating a window where numerous chip startups can thrive in viable niches. While an eventual bubble pop and consolidation is guaranteed, the immediate future remains bright for even smaller players, challenging the winner-take-all narrative.
GPUs were designed for graphics, not AI. It was a "twist of fate" that their massively parallel architecture suited AI workloads. Chips designed from scratch for AI would be much more efficient, opening the door for new startups to build better, more specialized hardware and challenge incumbents.
Anthropic's choice to purchase Google's TPUs via Broadcom, rather than directly or by designing its own chips, indicates a new phase in the AI hardware market. It highlights the rise of specialized manufacturers as key suppliers, creating a more complex and diversified hardware ecosystem beyond just Nvidia and the major AI labs.
OpenAI is actively diversifying its partners across the supply chain—multiple cloud providers (Microsoft, Oracle), GPU designers (Nvidia, AMD), and foundries. This classic "commoditize your compliments" strategy prevents any single supplier from gaining excessive leverage or capturing all the profit margin.
Beyond capital, Amazon's deal with OpenAI includes a crucial stipulation: OpenAI must use Amazon's proprietary Trainium AI chips. This forces adoption by a leading AI firm, providing a powerful proof point for Trainium as a viable competitor to Nvidia's market-dominant chips and creating a captive customer for Amazon's hardware.
While training has been the focus, user experience and revenue happen at inference. OpenAI's massive deal with chip startup Cerebrus is for faster inference, showing that response time is a critical competitive vector that determines if AI becomes utility infrastructure or remains a novelty.
OpenAI's deal structures highlight the market's perception of chip providers. NVIDIA commanded a direct investment from OpenAI to secure its chips (a premium). In contrast, AMD had to offer equity warrants to OpenAI to win its business (a discount), reflecting their relative negotiating power.
Major AI labs like OpenAI and Anthropic are partnering with competing cloud and chip providers (Amazon, Google, Microsoft). This creates a complex web of alliances where rivals become partners, spreading risk and ensuring access to the best available technology, regardless of primary corporate allegiances.
Beyond the simple training-inference binary, Arm's CEO sees a third category: smaller, specialized models for reinforcement learning. These chips will handle both training and inference, acting like 'student teachers' taught by giant foundational models.
