The AI Chip Market in 2026: Nvidia Dominates, AMD Accelerates, and Startups Change the Game
In the third fiscal quarter of 2026, Nvidia reported revenue of $51.2 billion in data centre, a 66% increase from the previous year, representing 90% of total revenue. In December 2025, Nvidia acquired Groq for $20 billion. In April 2026, OpenAI committed over $20 billion in chips from Cerebras. The AI chip market reached $38.1 billion in 2025 and is expected to grow to $377 billion by 2033.
The AI chip market is operating in a dynamic that few tech markets have experienced: structurally higher demand than supply capacity, with a single dominant supplier and startups capable of raising tens of billions in funding or acquisition based on alternative architectures that are not yet fully validated at scale.
Nvidia's Dominance
In the third fiscal quarter of 2026, Nvidia reported revenue of $51.2 billion solely in data centre, up 66% from the previous period, making up 90% of the company's total revenue. The Blackwell (B100/B200), launched in 2025, delivers approximately double the inference performance compared to the Hopper. The Vera Rubin, expected in the third quarter of 2026, will incorporate HBM4 memory with promises of a new leap in performance.
Nvidia's market capitalisation surpassed $4.3 trillion in 2026, making it the world's most valuable company at several points during the year.
AMD and the Challenge of Gaining Market Share
AMD operates with a market capitalisation of approximately $350 billion, positioned by analysts as an asymmetric risk bet: if its MI350 and MI400 products prove competitive performance in production benchmarks, the gap to Nvidia justifies a significant eventual re-evaluation. The MI350 has been described by AMD as the fastest ramp-up product in the company's history.
Intel, which once held 99% of the server processor market, now controls less than 1% of the dedicated AI accelerator market, retaining approximately 22% share in data centre when including CPUs.
Startups That Changed the Game
In December 2025, Nvidia acquired the intellectual property and talent of Groq for approximately $20 billion. Groq had developed the Language Processing Unit (LPU), a specialised architecture for high-speed inference that demonstrated consistently superior token generation speeds compared to conventional GPUs for specific models.
In April 2026, OpenAI committed over $20 billion in chips from Cerebras, a wafer-scale chip manufacturer. Cerebras produces the world's largest chip physically, occupying an entire silicon wafer, with the aim of eliminating communication bottlenecks between chips that limit performance in large-scale training workloads.
The Rise of Custom ASICs
In addition to established competitors, hyperscalers are developing proprietary accelerators. Google TPU, the trajectories of AWS Trainium and Inferentia, and Meta's AI chips are gaining share in internal workloads. Projections indicate a 44.6% increase in deliveries of custom ASICs from cloud providers in 2026, compared to 16.1% for conventional GPUs.
For organisations relying on AI hardware for their operations, the chip market in 2026 presents both opportunities and risks: greater variety of options than in 2023 and 2024, but also greater complexity in evaluation and potential for accelerated obsolescence of bets made on previous generations.