Volantis, a San Francisco semiconductor company building AI inference systems that move data with light, has raised an $88 million Series A co-led by Lachy Groom and Abstract Ventures. The company announced the round as it works toward delivering its first integrated inference engines in 2027.

John Doerr, VXI Capital, Triatomic and Susa Ventures also invested, along with angel investors Dwarkesh Patel, Naveen Rao and Sholto Douglas.

What Volantis Is Building

Volantis is going after what chip designers call the memory wall. As AI models grow, the limiting factor in running them is often not how fast a processor can compute but how quickly data can travel between the processor and memory. Electrical connections can only carry so much, and that ceiling restricts both model size and response speed.

The company's answer is photonics, using light in place of electrical signals to connect chips to memory. Its first system, called A-1, combines custom micro-VCSELs, a type of very small laser, with photonic interconnects designed specifically for chip-to-memory links.

Volantis says the architecture will let models exceeding 20 trillion parameters run at up to 10,000 tokens per second per user while lowering the cost of inference. Those are company targets for a system that has not yet shipped, and they would represent a substantial leap over what current hardware delivers.

The Team Behind Volantis

Volantis is led by CEO and co-founder Tapa Ghosh. The team includes veterans of NVIDIA, AMD, Broadcom and Ayar Labs, with experience spanning CoWoS advanced packaging products, tunable VCSELs and silicon photonics systems.

"Today's hardware forces a tradeoff between running the largest models and running them fast. We started Volantis to eliminate that tradeoff," Ghosh said in the announcement.

How Volantis Will Use the $88M

The funding will go toward developing and commercializing A-1, expanding the engineering team and preparing for customer deployments. The company is targeting 2027 for delivery of its first integrated inference engines.

An $88 million Series A is large by most standards but in line with what chip startups require. Designing, fabricating and packaging custom silicon and optics is capital intensive well before any revenue arrives.

Why the Volantis Funding Matters

Most attention in AI hardware has centered on training, the process of building models. Inference, the work of running them for users, is where spending is shifting as AI products reach large audiences. Each query carries a cost, and that cost determines which products are viable.

Hardware that serves larger models faster and more cheaply would change that math. It is why investors continue to fund alternatives to conventional GPU systems, and why photonics, long confined to networking between servers, is now being pushed closer to the chip.

The investor list reflects that conviction. Doerr is one of Silicon Valley's best-known venture investors, Rao has founded AI chip companies of his own, and Patel and Douglas are closely followed voices in AI research circles.

Volantis still has to prove its design in production silicon. If A-1 performs as described, it would give AI developers a new way around one of the hardest constraints in the field.

Source

PR Newswire — "Volantis Raises $88M Series A to Demolish the AI Memory Wall With Photonics"