A four-year-old chip startup just raised $88 million to bet against copper. Volantis, a Silicon Valley company building light-based connections between AI processors and memory, announced a Series A round on October 1, 2026, co-led by Lachy Groom and Abstract Ventures. The round pushes the company’s total funding past $97 million and places it in a crowded but increasingly well-funded race to solve one of AI’s most expensive problems: getting data to the chip fast enough to matter.
The timing is not an accident. Memory has become the bottleneck of the AI boom, not compute. As Nvidia’s own Rubin Ultra accelerators lose a third of their planned memory capacity to the HBM shortage, and as DRAM now costs more per chip than cutting-edge TSMC silicon, a startup promising to move data over light instead of wire has obvious appeal to investors chasing the next infrastructure chokepoint. Whether Volantis can turn that pitch into shipped silicon is a separate question, and one the company has not yet answered publicly.
What Volantis Announced on October 1
According to a press release distributed through PR Newswire and a company blog post, Volantis closed an $88 million Series A on October 1, 2026. Lachy Groom and Abstract Ventures co-led the round. Named participants include John Doerr, VXI Capital, Triatomic, and Susa Ventures, alongside angel investors Dwarkesh Patel, Naveen Rao, and Sholto Douglas, according to coverage from Unite.AI. The company says the new capital brings its lifetime funding to $97 million, which implies roughly $9 million raised before this round, though Volantis has not published a full breakdown of that earlier financing.
Volantis frames the raise around a specific engineering target: eliminating what it calls the memory wall in AI inference. In its own announcement, the company said, “We’re building a system for AI inference that uses photonics to break the memory wall,” a claim repeated across its public site and marketing material. The pitch is straightforward even if the engineering behind it is not: today’s AI chips can often compute faster than they can be fed data, and Volantis wants to replace the electrical wiring that feeds them with light.
The Memory Wall Problem, Explained
The “memory wall” is not a new phrase, but it has taken on new urgency since large language models started scaling into the trillions of parameters. Running inference on a frontier model means repeatedly shuttling enormous blocks of model weights and intermediate activations between memory and the processor doing the math. If that data pipe is too narrow or too far away, the processor sits idle no matter how fast its math units are.
Today’s answer to that problem is high-bandwidth memory, or HBM, stacked directly next to the GPU die. HBM works, but it has a hard physical limit: the electrical reach of copper traces on a package substrate. Chips can only fit so much memory within the “shoreline” around the processor before signal quality degrades. That limit is part of why Nvidia has locked up more than a third of global HBM supply just to keep its own roadmap on schedule, and why memory suppliers have struggled to keep pace with demand even as they post record results.
Volantis argues that optical links sidestep the shoreline limit entirely. In a technology page describing its architecture, the company said its waveguides “travel 200mm+, breaking the shoreline limit of HBM-based systems, letting us aggregate 100x chiplets into a single, ultra-high-bandwidth memory pool,” a claim it publishes on its technology page. Because light does not suffer the same resistance and crosstalk penalties as copper over distance, Volantis says it can place far more memory within reach of a single processor than a conventional electrical package allows.
How the Photonic Interconnect Actually Works
Volantis’s approach centers on VCSEL-based photonic interconnects. VCSEL stands for vertical-cavity surface-emitting laser, a laser type already common in short-range data center optics and consumer sensors. Rather than designing a new optical transceiver standard from scratch, Volantis is reportedly adapting VCSEL technology to link AI compute directly to pools of memory sitting further away than copper would allow.
The company describes the payoff in blunt terms. On its homepage, Volantis states that “optical reach is ~100x longer than electrical, so you can put ~100x more memory next to a chip.” That is a company claim, not an independently measured benchmark, and Volantis has not published peer-reviewed figures for achieved bandwidth, energy per bit, or latency. Still, the direction is consistent with a broader industry trend: as copper SerDes links push toward their physical limits at higher speeds, data center hardware designers are increasingly looking at optics to carry data the “last few centimeters” between chips, not just between racks.
Separately, one industry report described Volantis’s longer-term targets as including systems that could surround a single GPU with roughly 220 memory chips, support models beyond 20 trillion parameters, and sustain throughput in the range of 10,000 tokens per second per user, with first inference engines targeted for 2027. Those figures have not been confirmed by Volantis itself and should be read as reported goals rather than demonstrated performance.
Who Is Building This: The Founding Team
Volantis was founded in 2022 by chief executive Tapa Ghosh and chief technology officer Roy Meade. Ghosh is described in company materials and by Unite.AI as a Thiel Fellow and a former Y Combinator founder who holds four patents. Meade’s background points more directly at the problem Volantis is trying to solve: he reportedly led Micron’s HBM program before serving as a vice president at Ayar Labs, one of the best-known names in optical chip-to-chip interconnects.
The wider team carries similar pedigree. Inderjit Singh, listed as director of packaging, is credited with helping build the industry’s first CoWoS product, the advanced packaging platform TSMC now uses for nearly every high-end AI accelerator. Daniel Klowden, vice president of engineering, previously worked on an early processor design with direct optical communication, and Chris Chase, director of laser engineering, is credited with taking a new VCSEL laser category into volume production. Unite.AI and Volantis both describe the broader staff as including veterans of Nvidia, AMD, Broadcom, Ayar Labs, and Micron.
Volantis Funding Fact Sheet
| Item | Detail |
|---|---|
| Round | Series A |
| Amount raised | $88 million |
| Announcement date | October 1, 2026 |
| Lead investors | Lachy Groom, Abstract Ventures |
| Named participants | John Doerr, VXI Capital, Triatomic, Susa Ventures |
| Named angels | Dwarkesh Patel, Naveen Rao, Sholto Douglas |
| Total funding to date | $97 million |
| Founded | 2022 |
| Founders | Tapa Ghosh (CEO), Roy Meade (CTO) |
| Core technology | VCSEL-based photonic interconnect for AI inference |
The Competitive Landscape for Optical Interconnects
Volantis is not alone in betting that light will replace copper inside the AI data center. The field already includes several well-funded players chasing overlapping, though not identical, pieces of the problem. Some focus on chip-to-chip optical I/O between processors and switches, others on optical links between compute and memory specifically, and a few are pursuing entirely different physical approaches, such as microLEDs rather than lasers.
| Company | Focus | Notable detail |
|---|---|---|
| Volantis | Photonic compute-to-memory links for AI inference | $88M Series A, Oct. 2026; CTO formerly led Micron’s HBM program |
| Ayar Labs | Optical I/O chiplets replacing electrical chip-to-chip links | Volantis CTO Roy Meade previously held a VP role here |
| Lightmatter | Photonic computing and optical interconnect (Passage platform) | One of the earliest movers in photonic processors |
| Celestial AI | “Photonic Fabric” linking compute and memory | Positions itself directly against HBM bandwidth limits |
| Avicena | MicroLED-based dense short-reach optical links | Uses microLEDs instead of lasers, a distinct physical approach |
| POET Technologies | Optical engines and photonic-integrated packaging | Focused on data center and AI optical communications |
| Eliyan | Advanced die-to-die and chiplet interconnect | Electrical rather than photonic, competes for the same budget |
| Marvell | Optical DSPs, electro-optics, custom silicon | Public incumbent, not a venture-stage comparable |
None of these companies has published a fully audited, apples-to-apples funding comparison, and several operate as divisions of larger public companies or have raised multiple undisclosed rounds, so exact totals are difficult to state with confidence. What is clear is that Volantis’s $88 million round lands in a field where the core technical bet, that optics beat copper over the distances AI chips now need, is shared by at least half a dozen serious competitors, several of which have a multi-year head start.
Historical Context: How We Got to the Memory Wall
Data center interconnects have been quietly moving toward optics for years, just not at the chip-to-memory distance Volantis is targeting. Optical transceivers have long carried data between racks and switches, where electrical signaling would never reach the required distances. The new frontier is pulling that same physical insight inward, from the scale of a data center hall down to the scale of a single package a few centimeters wide.
That shift has been forced by economics as much as physics. HBM shortages have already pushed AI chip prices up sharply across the industry, and memory makers have had to prioritize AI customers over nearly everyone else. Everspin’s CXL-connected magnetic RAM, which claims write speeds up to 100 times faster than comparable memory, is one sign that chipmakers are experimenting with fundamentally different memory architectures rather than waiting for HBM supply to catch up. Volantis’s photonic approach is another branch of the same search: if you cannot get more memory bandwidth from the existing electrical playbook, change the playbook.
Micron’s own record fiscal 2026 results underline how tight the memory market has become even for the companies selling into the shortage. When the company supplying the raw material is setting revenue records off scarcity, the incentive for AI chip designers to find an alternative to buying more HBM, rather than simply buying it faster, becomes much stronger.
In Volantis’s Own Words
Volantis has been consistent in how it frames its ambitions publicly, repeating variations of the same core claim across its press release, blog post, and website. In its funding announcement, the company said, “Volantis is building a new AI inference architecture that eliminates the tradeoff between memory capacity and bandwidth,” a statement published in its PR Newswire release.
Elsewhere, in the blog post detailing the raise, Volantis went further, stating, “We have built a new category of optical interconnect that can boost both memory capacity and bandwidth by over an order of magnitude,” a line published on the company’s news and insights page. These are company claims rather than independently verified benchmarks, and no third-party lab result accompanying the funding announcement has been made public.
Market Impact: What This Means for Nvidia, AMD, and Hyperscalers
An $88 million Series A does not move Nvidia’s or AMD’s roadmaps on its own, but it is a data point in a much larger story: the chip industry’s biggest buyers are actively funding, or at least watching closely, every plausible alternative to the current HBM-centric model. If HBM supply stays this tight through 2028, as some supply-chain analysts have projected, hyperscalers have a direct financial incentive to diversify their bets across memory architectures rather than wait in line.
For Nvidia and AMD specifically, a mature optical compute-to-memory link would be a genuine architectural shift, not an incremental upgrade. Both companies have built years of design and manufacturing expertise around electrical HBM stacks. A credible photonic alternative would not replace that overnight, but it changes the long-term calculus for where the next generation of memory-bandwidth gains comes from, and it gives chip designers a bargaining chip against memory suppliers who currently hold most of the leverage in pricing negotiations.
For cloud providers already paying more for AI infrastructure, the appeal is more direct. Rising chip and memory prices have been squeezing margins across the AI supply chain all year, and any technology that credibly promises more memory per dollar of silicon is going to get a hearing from cloud infrastructure teams, even at the early, unproven stage Volantis is at today.
What Volantis Has Not Proven Yet
It is worth being direct about the gap between Volantis’s funding announcement and a shipped product. The company has not published independently verified figures for achieved optical bandwidth, end-to-end inference latency, energy consumed per bit transferred, production yield, or any commercial customer deployment. The 220-memory-chip, 20-trillion-parameter, 10,000-tokens-per-second targets referenced in trade coverage are reported goals, not demonstrated results, and Volantis itself has not confirmed them as specifications.
That gap is normal for a Series A-stage hardware company. Building a working photonic interconnect that survives manufacturing tolerances, thermal cycling, and years of continuous operation in a data center is a very different challenge than demonstrating a lab prototype. Ayar Labs and Lightmatter, both years ahead of Volantis in funding and development time, are still working toward broad commercial deployment of their own optical I/O products. There is no guarantee Volantis closes that gap faster, or that its specific VCSEL-based approach wins out over rival optical architectures.
Why Investors Are Betting on Photonics Now
The investor list behind this round is itself a signal. John Doerr’s involvement, alongside AI-focused angels like Dwarkesh Patel and Naveen Rao, suggests a bet that memory bandwidth, not raw compute, will be the defining constraint on AI progress over the next several years. That view has gained ground steadily through 2026 as HBM shortages have rippled through pricing across the entire chip industry, from data center GPUs down to consumer devices, where memory scarcity has already pushed component costs sharply higher.
Abstract Ventures and Lachy Groom’s decision to co-lead, rather than simply participate, also signals a willingness to back a technically ambitious hardware bet at a stage where most of the engineering risk is still unresolved. That is a meaningfully different kind of conviction than funding a software layer built on top of existing chips.
Five Predictions for Photonic Interconnects Through 2028
- Expect at least one more optical-interconnect startup to announce a funding round above $50 million within the next six months, as investors continue chasing the memory-wall thesis.
- Volantis will likely need to disclose independently measured bandwidth, latency, or power figures within the next 12 to 18 months to maintain credibility against better-funded rivals like Ayar Labs and Lightmatter.
- If HBM shortages persist as some supply-chain analysts have suggested they might through 2028, expect at least one major cloud provider to announce a pilot program or investment stake in an optical-interconnect startup rather than relying solely on traditional memory suppliers.
- Nvidia or AMD are more likely to acquire or partner with an established optical I/O company than to build the capability entirely in-house, given how far behind a from-scratch internal program would start.
- Expect continued consolidation pressure in the space by 2027 or 2028, as the capital intensity of qualifying photonic interconnects for mass production in AI accelerators favors companies that can either out-fund or out-partner their rivals.
These are editorial projections based on current funding patterns and publicly reported supply constraints, not statements made by Volantis or any named source.
How Volantis Compares to the Broader AI Infrastructure Market
It helps to size Volantis against the scale of the problem it says it is solving. Nvidia alone has reportedly locked up more than a third of available HBM supply to protect its own production forecasts, a figure that dwarfs the entire photonic-interconnect funding landscape combined. Against that backdrop, $88 million is a meaningful vote of confidence in a four-year-old startup, but it is a rounding error next to what the three or four largest AI chip and memory companies spend securing supply every quarter.
That mismatch in scale is exactly why photonic interconnect startups position themselves as potential disruptors rather than incumbents. A technology that meaningfully loosens the memory bottleneck would not need to match Nvidia’s capital spending to matter. It would only need to work reliably enough, at a low enough cost, that chip designers start treating it as a credible second option to conventional HBM stacking.
Frequently Asked Questions
What does Volantis actually make?
Volantis is developing a photonic, or light-based, interconnect system intended to connect AI processors with memory over longer distances than conventional electrical links allow, aiming to reduce the data-movement bottleneck known as the memory wall during AI inference.
How much money has Volantis raised?
Volantis raised $88 million in a Series A round announced October 1, 2026, bringing its total disclosed funding to $97 million since its founding in 2022.
Who are Volantis’s biggest competitors?
The closest competitors in optical and photonic interconnects include Ayar Labs, Lightmatter, Celestial AI, Avicena, POET Technologies, Eliyan, and incumbent supplier Marvell, each pursuing a somewhat different technical approach to the same bandwidth problem.
Is Volantis’s technology an HBM replacement?
Volantis positions its technology as a way to pool more memory around a processor than HBM’s physical shoreline limit allows, but it has not published independently verified figures proving it is a qualified, drop-in replacement for HBM in production AI accelerators.
Why does the memory wall matter for AI right now?
As AI models have grown into the trillions of parameters, moving model weights and activation data between memory and compute fast enough has become as important as raw processing speed, and persistent HBM shortages have made that bottleneck more expensive and more visible throughout 2026.
When will Volantis ship a product?
Volantis has not confirmed an official shipping date. Trade press coverage has referenced a target of 2027 for first inference engines, but that figure has not been verified directly by the company as a committed roadmap.
Who founded Volantis?
Volantis was founded in 2022 by chief executive Tapa Ghosh, a Thiel Fellow and former Y Combinator founder, and chief technology officer Roy Meade, who previously led Micron’s HBM program and held a vice president role at Ayar Labs.
Does this funding round affect Nvidia or AMD directly?
Not directly and not immediately. Volantis is an early-stage, privately funded company with no disclosed commercial deployments, but its funding reflects growing investor interest in alternatives to the HBM-based memory architecture that Nvidia and AMD currently rely on for their AI accelerators.
