Chip Startup Etched Raises $800 Million from Jane Street and TSMC-Linked Venture Tech Alliance

Etched, a startup developing chips for running (inference) of AI models, has exited stealth mode and announced raising 800 million dollars in funding — among the investors are trading firm Jane Street and Venture Tech Alliance, a company linked to TSMC. Etched CEO Gavin Uberti spoke about the company's plans in an interview with Ed Ludlow at Bloomberg Tech.

Why the industry is shifting from training to inference

In the early years of the generative AI boom, demand for computing concentrated around training large models — this is where Nvidia dominates with its GPUs. But as more companies have already trained or adopted ready-made models and are moving to their mass deployment in production, the focus of spending is shifting to inference — the constant, round-the-clock process of generating answers for millions of users. Etched is targeting exactly this market, aiming to challenge Nvidia not through versatility but through specialization.

How much the company raised and from whom

• The funding round totaled 800 million dollars
• Among investors is trading and market-making firm Jane Street
• Venture Tech Alliance, a company connected to Taiwan-based chip manufacturer TSMC, also participated
• Gavin Uberti, CEO of the company, gave an interview to Bloomberg Tech

How Etched's chip differs from Nvidia's GPU

Etched's key bet is chips designed not as general-purpose GPUs but specifically for transformer architecture, which is the foundation of the vast majority of modern language models. According to the company's claims, such specialization should provide significant gains in speed and cost of inference compared to universal graphics cards, which waste some resources on tasks unnecessary for running transformer models. The participation of TSMC-linked investor in the round also indicates that the company expects to obtain manufacturing capacity for its new chip from Taiwan's giant — like most players in the AI chip market, including Nvidia itself.

How strong is a competitor to Nvidia

Nvidia remains the undisputed leader in the AI chip market and holds an overwhelming share in both training and inference of models — largely due to the CUDA ecosystem, to which most developers are locked in. To win market share from such a player, a startup like Etched needs more than just to release a fast chip: it must convince major clients — cloud providers and model developers — to migrate real production workloads to new hardware, and this requires not just performance but a mature software stack, deployment tools, and confidence in supply reliability. This is exactly why a large funding round and participation of investors like Jane Street and TSMC-linked Venture Tech Alliance are important not only as money for development but as a signal to the market about the company's seriousness of intent.

What is unusual about the investor composition

Jane Street's participation in the round stands out compared to the typical circle of investors in AI startups: it is primarily a quantitative trading and market-making firm, not a venture fund or technology giant. Its interest in financing a chip startup reflects a broader trend of recent years — major financial companies are increasingly investing directly in artificial intelligence infrastructure rather than just using ready-made models in their own trading and analytical systems. Participation of TSMC-linked Venture Tech Alliance adds strategic meaning to the round: Taiwan's chip manufacturer this way gains closer ties with yet another potential major customer for its manufacturing capacity.

What this means

The 800 million dollar round and exit from stealth mode show that the market for specialized AI model inference chips continues to attract large capital despite Nvidia's dominance. For the industry, this means growing competition for the fastest-growing segment of AI infrastructure spending — not one-time model training but their constant operation at scale.