The three AI chip startups raising billions to challenge Nvidia

Technology

Cerebras develops the Wafer Scale Engine 3 (WSE-3), a single enormous processor the size of a silicon wafer that integrates hundreds of thousands of cores. The company claims the WSE-3 uses a fraction of the power per unit compute while delivering inference and training performance more than 20 times faster than competing solutions . Its approach is built for both training and inference workloads that benefit from massive on-chip compute density.

SambaNova Systems: Inference infrastructure at $11 billion

Fundraising and valuation

SambaNova closed the first tranche of a $1 billion Series F in July 2026 at an $11 billion post-money valuation, roughly 2.3 times its valuation from five months earlier . The round was led by General Atlantic and included participation from Intel Capital, BlackRock, T. Rowe Price, Capital Group, Qatar Investment Authority, and others . CEO Rodrigo Liang told TechCrunch that a second close with additional investors was expected within weeks .

Technology

SambaNova builds custom chips, hardware systems, and cloud services purpose-built for inference — the process of running trained AI models to generate responses . Its flagship SN50 chip, unveiled in February 2026, is designed specifically for large language model inference. The company claims the SN50 can provide more than three times as much throughput as Nvidia's B200 GPU . JPMorgan Chase has selected SambaNova's systems to power secure, on-premises AI inference .

Positron: Energy-efficient inference at up to $5 billion

Fundraising and valuation

Positron raised a $230 million Series B in February 2026 at a valuation exceeding $1 billion, led by Arena Private Wealth, Jump Trading, and Unless, with strategic investment from Arm and the Qatar Investment Authority . In July 2026, the company entered negotiations for an additional approximately $750 million two-phase financing that could value the company at up to $5 billion .

Technology

Positron focuses on energy-efficient AI inference chips designed to dramatically reduce the power consumption of running AI models compared to Nvidia GPUs . The company's Atlas systems are claimed to deliver 3.5 times the performance per dollar and 3.5 times the power efficiency versus comparable Nvidia hardware . Its next-generation Asimov chip is scheduled for tape-out in late 2026 with initial samples expected in early 2027 .

The broader trend: Specialization and 'de-Nvidification'

These three companies are part of a much larger wave. AI chip startups raised over $8.3 billion globally in 2026, with U.S. companies accounting for $4.7 billion and European entrants contributing about $800 million . The median AI chip round is $250 million, an exceptionally high figure by venture capital standards .

Three broad themes stand out:

• Specialization over generalization. Instead of building general-purpose GPUs to match Nvidia head-on, these startups are attacking specific slices of the AI compute stack. Cerebras targets wafer-scale compute for training and inference; SambaNova and Positron both focus on inference, the fastest-growing segment as deployed AI models proliferate .
• Inference is the new battleground. While Nvidia still dominates AI training, competition is rapidly heating up in inference, where purpose-built ASICs can offer better performance per watt and per dollar for running models in production .
• Investors bet on margin disruption. The core thesis is that Nvidia's dominant position and high margins will not remain invulnerable as AI workloads scale massively and hyperscale buyers seek cheaper, more power-efficient alternatives .

The combined $2.55 billion raised by these three companies in a single week in July 2026 signals a market conviction that the GPU monoculture in AI computing is finally being challenged.