Volantis Raises $88 Million to Develop Laser-Based AI Chip Technology

India, Oct 02 : San Francisco based semiconductor startup Volantis has raised $88 million in venture capital to develop technology aimed at solving one of the major performance challenges facing artificial intelligence processors: moving data efficiently between computing chips and memory.

The company announced the funding on October 1, saying its approach uses beams of laser light to connect AI computing components with memory chips. Volantis believes the technology could allow significantly more memory to be placed around a graphics processing unit, or GPU, than is possible with conventional electrical connections.

The development comes as demand for AI computing continues to put pressure on the semiconductor industry. Modern AI models require large amounts of data to be transferred repeatedly between processing units and memory. The speed of that movement can become a limiting factor even when the processor itself has substantial computing power.

Nvidia and AMD currently address this challenge by using high-bandwidth memory positioned close to their computing chips. The arrangement provides rapid access to data but is constrained by the physical connections between the components.

According to Reuters, Nvidia’s current high-end designs can accommodate eight memory chips around each GPU because of the limited reach of the tiny electrical wires used to connect the components. Volantis says its optical approach could potentially allow as many as 220 memory chips to be placed around a GPU.

The company is using a technology known as vertical cavity surface emitting lasers, or VCSELs, to transmit information through light rather than relying solely on electrical connections.

VCSELs are not a completely new technology. They are already used in consumer electronics, including facial-recognition systems in many Apple devices. Volantis is seeking to adapt the underlying technology for the much more demanding task of connecting AI processors with large quantities of memory.

The approach reflects a broader trend in semiconductor engineering: instead of relying only on faster processors, companies are looking for ways to improve the movement of information between different components.

AI workloads are particularly demanding because large models require processors to repeatedly access huge amounts of data. If communication between processing and memory components cannot keep pace, additional computing capacity does not necessarily translate into proportional gains in performance.

This problem has become more important as companies develop larger AI systems and deploy them for applications ranging from software development and scientific research to business automation.

Volantis chief executive and co-founder Tapa Ghosh said the company’s goal was to use established components in a new configuration rather than depend on an entirely new semiconductor manufacturing process.

That could be important for commercial development because advanced chip manufacturing is complex and expensive. Building on existing technologies may allow companies to focus on packaging and connections rather than creating an entirely new category of semiconductor components.

Volantis said it hopes to deliver a chip next year that could accelerate workloads such as AI coding. The company is effectively targeting the communication layer between the processor and memory rather than attempting to compete directly by designing another general-purpose AI accelerator.

The funding round was led by Lachy Groom, a former Stripe executive, and venture capital firm Abstract Ventures. John Doerr, an early backer of companies including Google and Amazon, also participated. Other investors included VXI Capital, Triatomic and Susa Ventures.

Angel investors in the round included AI podcaster Dwarkesh Patel, AI chip specialist Naveen Rao and Anthropic researcher Sholto Douglas.

The investment illustrates the continuing flow of private capital into semiconductor technologies designed to support the artificial intelligence industry. While much of the attention surrounding AI hardware has focused on leading processor manufacturers, startups are attempting to address specific bottlenecks within the broader computing system.

Memory has become particularly important because AI models depend heavily on rapid access to data. The processor may perform billions of operations, but it still needs an efficient way to retrieve the information required for those calculations.

Traditional electrical connections have physical limitations. As more components are packed into increasingly dense systems, maintaining signal quality and sufficient bandwidth becomes more difficult.

Optical communication offers a potential alternative because information can be transmitted using light. Volantis is betting that VCSEL-based connections can help overcome some of the physical limitations associated with electrical wiring.

The company’s proposal would also alter how memory is arranged around AI processors. Instead of being restricted by the reach of tiny electrical connections, optical links could potentially allow memory components to be positioned more flexibly.

That could have implications for future AI accelerators, particularly as developers seek to build systems capable of handling increasingly large models.

The technology could also contribute to efforts to improve computing efficiency. If processors can access data more rapidly, systems may spend less time waiting for information to arrive from memory. Faster communication between components can therefore improve the utilisation of expensive computing hardware.

However, the commercial success of the approach will depend on whether the technology can be manufactured reliably and integrated into real-world AI systems. Advanced semiconductor packaging involves significant engineering challenges, and optical connections introduce their own design and manufacturing requirements.

Volantis is attempting to reduce those risks by relying on VCSELs that already have large-scale applications. The technology’s existing use in consumer electronics means the underlying components are already part of established manufacturing ecosystems.

The startup’s strategy also comes at a time when semiconductor companies are exploring different ways to address the physical limitations of AI infrastructure. Processing power is increasing rapidly, but improvements in computing capacity have to be matched by advances in memory bandwidth, packaging and interconnects.

The AI industry therefore increasingly depends on technologies that may receive less attention than the processors themselves.

GPUs remain central to AI training and inference, but the overall performance of an AI system depends on an interconnected chain of components. Memory, networking, packaging and data movement can all influence how effectively processors are used.

Volantis is targeting one part of that chain with its optical connection technology. If its planned system reaches commercial production, it could provide another approach for connecting processors with large pools of high-speed memory.

For the semiconductor industry, the significance of such developments extends beyond a single startup. The rapid growth of artificial intelligence is forcing chip designers to reconsider how computing systems are assembled and how data moves through them.

Volantis’ $88 million funding round demonstrates investor interest in technologies that address those infrastructure challenges. Its planned chip for next year will provide an important test of whether laser based connections can move from a promising concept into a practical component of AI computing systems.

As AI models continue to demand greater processing and memory resources, improvements may increasingly come not only from designing more powerful chips but also from finding better ways for those chips to communicate with the memory and other components surrounding them.