October 2, 2026 — San Francisco.
Semiconductor startup Volantis has raised $88 million to attack one of AI’s hardest hardware problems: the gap between processors and memory. Its plan is to move data between compute and memory chips with beams of laser light, which could let a single GPU reach 220 memory chips — versus just 8 for Nvidia’s best current offerings.
Why it matters: AI performance is increasingly limited not by raw computing power but by how fast processors can get at the data they need. Every inference request forces a model to shuttle enormous amounts of data between memory and compute. If Volantis’s optical approach works, it could make AI models — and everything from coding assistants to chatbots — meaningfully faster and cheaper to run.
What happened
Volantis announced the $88 million round on October 1, reporting by Reuters. The round was led by Stripe veteran Lachy Groom and Abstract Ventures, with participation from John Doerr (who backed Google and Amazon in their early days), VXI Capital, Triatomic, and Susa Ventures. Angel investors include AI podcaster Dwarkesh Patel, AI chip veteran Naveen Rao, and Anthropic researcher Sholto Douglas.
The technical bet: VCSELs
The core problem: tiny electrical wires can only carry data so far, so Nvidia and AMD ring their computing chips with expensive high-bandwidth memory — capped at about 8 memory chips per GPU. Volantis wants to replace those electrical links with vertical-cavity surface-emitting lasers (VCSELs), which send data as light and eliminate the reach problem.
The clever part is that VCSELs aren’t exotic: they power Face ID-style facial recognition in hundreds of millions of iPhones, and Apple spent a decade building up that supply chain. Volantis is betting that reusing a proven, high-volume component lets it dodge the supply-chain bottlenecks that sink most chip startups. CEO and co-founder Tapa Ghosh told Reuters the company is aiming to deliver a chip next year that would speed up tasks like AI coding, noting that advanced packaging is “never trivial — but it’s not necessarily a new thing to do.”
Why it matters
From an “everything AI, tested” angle, this is a refreshingly grounded bet in an era of moonshot chip pitches: no new physics required, just applying smartphone laser technology to the memory bottleneck that constrains every large model. The investor roster — a Stripe veteran, John Doerr, and working AI researchers — suggests this isn’t just pitch-deck vapor. If Volantis ships on schedule next year and the 220-chip figure holds up in production, it could reset expectations for inference costs across the industry. The watch item is execution: “deliver a chip next year” is an aggressive timeline for a hardware startup, and advanced packaging has humbled plenty of confident teams.
FAQ
What is the “memory wall” in AI chips?
It’s the bottleneck where processors sit idle waiting for data to arrive from memory. Modern AI models are so large that shuttling their weights between memory and compute is often the slowest part of every request.
What are VCSELs?
Vertical-cavity surface-emitting lasers — tiny lasers that transmit data as light. They already power facial recognition in hundreds of millions of smartphones, which gives Volantis an established, high-volume supply chain to draw on.
Who is backing Volantis?
The $88M round was led by Lachy Groom and Abstract Ventures, with John Doerr, VXI Capital, Triatomic, and Susa Ventures, plus angels including AI podcaster Dwarkesh Patel and Anthropic researcher Sholto Douglas.
When will Volantis ship a chip?
CEO Tapa Ghosh told Reuters the company is targeting next year, with early workloads like AI coding in mind. That’s ambitious for a hardware startup — treat the timeline as a goal, not a guarantee.
Sources: Reuters, The Tower Post