Volantis Raises $88M Series A to Build Photonic Memory Infrastructure for AI

Volantis, a San Francisco-based semiconductor startup, has raised $88 million in Series A funding to develop a new photonic architecture designed to address one of the biggest infrastructure challenges facing next-generation AI systems: the growing gap between compute and memory.

The round was co-led by Lachy Groom and Abstract Ventures, with participation from John Doerr, VXI Capital, Triatomic and Susa Ventures. Angel investors including Dwarkesh Patel, Naveen Rao and Sholto Douglas also participated.

Volantis plans to use the capital to develop and commercialise its A-1 inference system, expand its engineering team and move toward customer deployments. The company expects to deliver its first integrated inference engines in 2027.

Building Infrastructure for Trillion-Parameter AI Models

Volantis was founded by Tapa Ghosh and Roy Meade.

Ghosh is a Thiel Fellow and former Y Combinator founder with four patents. Meade previously led Micron's high-bandwidth memory programme and served as a vice president at Ayar Labs.

The founders are focused on a problem that becomes increasingly difficult as AI models grow larger.

Modern AI systems require enormous amounts of memory to store model weights and context. They also need extremely high bandwidth to continuously move that information between memory and compute.

As models become larger and AI agents perform increasingly complex workloads, the amount of data that needs to move through AI infrastructure continues to increase.

Volantis believes existing hardware architectures create a fundamental trade-off between memory capacity, bandwidth, cost and energy consumption.

Its A-1 system is designed to address that problem using photonic technology.

A-1 Targets Models With More Than 20 Trillion Parameters

Volantis says its A-1 inference system is being designed to run models exceeding 20 trillion parameters at speeds of up to 10,000 tokens per second per user.

The company stresses that these numbers are product targets as it moves toward commercialisation rather than performance measurements from an already deployed system.

The goal is to enable AI systems to run larger models without forcing enterprises to sacrifice inference speed.

That could become increasingly important as AI agents take on more tasks.

Faster inference can directly affect how quickly an AI system can respond, complete workflows and execute agentic tasks. As businesses use AI for increasingly complex operations, infrastructure performance becomes an important factor in determining how much work those systems can handle.

Rewiring the Connection Between Compute and Memory

At the heart of Volantis' approach is the connection between computing chips and memory.

Traditional AI hardware architectures rely on different forms of memory with different performance characteristics.

On-chip SRAM provides extremely high bandwidth but has limited capacity.

High-bandwidth memory provides substantially more capacity, but systems remain constrained by bandwidth, cost and energy requirements.

Volantis argues that newer approaches such as 3D DRAM still operate within the same broader trade-off.

The company's solution is a photonic interconnect designed specifically to connect compute chips with memory.

Its optical fabric creates a unified pool of memory and aggregates bandwidth as additional memory is added.

This approach is intended to allow memory capacity and bandwidth to scale together rather than forcing system designers to increase one while compromising the other.

Using Photonics to Move Data More Efficiently

Volantis' architecture relies on optical communication to move data between compute and memory.

The company uses custom micro-VCSELs rather than relying on external lasers.

Volantis says this approach can leverage the established gallium arsenide VCSEL supply chain while avoiding some of the constraints associated with indium phosphide.

The company is targeting end-to-end optical links that consume less than one picojoule per bit.

If the technology achieves its targets, it could help reduce some of the energy and bandwidth challenges associated with moving enormous quantities of data through AI infrastructure.

This is increasingly important as AI workloads place pressure not only on compute capacity but also on the memory systems and interconnects that feed those compute resources.

A Growing Photonics Race in AI Infrastructure

Volantis is entering an increasingly competitive AI infrastructure market focused on photonics.

Lightmatter has raised $400 million at a $4.4 billion valuation to develop photonic computing and interconnect technologies for AI data centres.

Ayar Labs, where Volantis co-founder Roy Meade previously served as a vice president, has raised $500 million in Series E funding at a $3.8 billion valuation for optical interconnect technology designed to address data movement between chips.

Celestial AI has also raised $250 million at a $2.5 billion valuation for its Photonic Fabric platform, which targets bandwidth and data-movement bottlenecks in AI infrastructure.

These companies are approaching different parts of the same underlying problem: conventional electronic connections are increasingly being pushed to their limits as AI systems scale.

Volantis is taking a more memory-focused approach.

Rather than primarily replacing compute with photonic systems, the company is attempting to redesign the relationship between compute and memory so that AI systems can access larger pools of memory with greater bandwidth.

Preparing for Customer Deployments

The new funding will support Volantis as it moves from architecture development toward commercialisation.

The company plans to expand its engineering team and develop its A-1 inference system while working toward customer deployments.

Its first integrated inference engines are targeted for delivery in 2027.

That timeline gives Volantis several important milestones ahead.

The company will need to demonstrate that its photonic architecture can achieve its targeted performance while remaining manufacturable and economically viable for customers.

The challenge is particularly significant in AI infrastructure, where semiconductor systems must deliver improvements not only in theoretical performance but also in reliability, integration and deployment economics.

The Bigger AI Infrastructure Challenge

The rapid expansion of AI is creating pressure across the entire computing stack.

Larger models require more compute. More AI agents generate more inference workloads. And increasingly sophisticated applications require larger amounts of memory and faster movement of data.

That means the bottleneck is no longer simply how many computational operations a chip can perform.

Moving data efficiently between compute and memory is becoming just as important.

Volantis is betting that photonics can provide a different path forward by allowing memory capacity and bandwidth to scale together.

With $88 million in Series A funding and its first integrated inference engines targeted for 2027, the company now has capital to push its architecture toward real-world deployment.

The next major test will be whether A-1 can turn its ambitious targets for trillion-parameter models and high-speed inference into a commercially deployable system.

If successful, Volantis could become part of a new generation of AI infrastructure designed around the realities of increasingly large models and increasingly demanding agentic workloads.