Etched, a San Jose-based startup, has reportedly secured approximately $500 million in a new funding round that values the company at $5 billion. Led by the venture firm Stripes and featuring backers like Peter Thiel, this massive raise elevates Etched to "Unicorn" status as it prepares to challenge Nvidia’s dominance. With total funding now nearing $1 billion, the company is accelerating the development of its specialized AI hardware.
Engineering the Sohu AI Accelerator
The centerpiece of the company’s strategy is the Sohu chip, an application-specific integrated circuit (ASIC) designed exclusively to accelerate transformer-based AI models. Unlike traditional GPUs that handle diverse computational tasks, Sohu "hard-wires" the specific matrix multiplication patterns required for large language models like ChatGPT and Claude. Etched is leveraging Taiwan Semiconductor Manufacturing Co.’s (TSMC) advanced 4nm process technology and High Bandwidth Memory (HBM) to produce this specialized silicon. By stripping away the flexibility needed for other neural network types, the company claims it can achieve unprecedented throughput and energy efficiency for the industry's most popular model architecture.
Performance Claims and Market Ambition
Etched has made headlines with bold performance assertions, claiming that its Sohu chip can deliver inference speeds up to 20 times faster than Nvidia’s H100 GPUs. The company suggests that a single server equipped with eight Sohu chips could effectively replace up to 160 H100 units, offering a far more compact and cost-effective solution for data centers. However, these figures remain unverified by independent third-party benchmarks, leaving some industry analysts cautious about how the chip will perform against newer competitors like Nvidia’s Blackwell (B200) architecture. Despite these uncertainties, specialized cloud providers and inference-as-a-service platforms are watching closely, as the potential for a 10-fold increase in efficiency could drastically lower the operational costs of deploying massive AI models.
Strategic Risks and the AI Hardware Race
The path to challenging a market leader that controls an estimated 70% to 95% of the AI accelerator market is fraught with significant execution and technological risks. CEO Gavin Uberti has openly acknowledged that Etched’s specialized approach is a "bet on transformers," meaning a fundamental shift in AI architectures could potentially render the Sohu chip obsolete. Beyond architectural shifts, the company must also navigate multi-year development cycles and the complex logistics of scaling high-end semiconductor manufacturing. Nevertheless, the successful $500 million raise provides the financial runway necessary to recruit veteran talent from established giants like Broadcom and Cypress Semiconductor to help bring their vision to fruition.
This landmark funding round signals that the market is ready for dedicated silicon that prioritizes extreme efficiency for specific, high-demand AI workloads. While the ultimate success of the Sohu chip depends on real-world validation, Etched has successfully positioned itself at the forefront of the next wave of hardware innovation. By narrowing its focus to doing one thing exceptionally well, the company is betting that purpose-built silicon is the key to the future of AI.