San Francisco semiconductor startup Volantis has raised $88 million in venture capital to develop a new approach to one of the most important physical limitations facing artificial-intelligence processors: how quickly computing chips can exchange data with memory. The company says it is developing technology that uses laser light rather than conventional electrical connections to link a processor with a much larger number of memory chips. Reuters reported on October 1, 2026, that Volantis believes its approach could eventually allow as many as 220 memory chips to be connected around a single GPU, compared with eight in Nvidia's current leading offerings. The technology uses vertical-cavity surface-emitting lasers, or VCSELs, components already deployed at enormous scale in consumer electronics. If Volantis can turn that concept into a commercially viable product, it could address a critical bottleneck as AI models demand increasingly large amounts of memory bandwidth and computing capacity.
Why AI Chips Are Running Into A Memory Problem
The performance of an AI processor depends on more than the raw computational power of its processing cores.
AI models are enormous collections of numerical parameters that must be repeatedly moved between memory and computing units while workloads are running. If the processor can calculate faster than the surrounding memory system can deliver data, the computing resources may spend part of their time waiting.
This creates what semiconductor engineers call a memory and interconnect bottleneck.
Nvidia and AMD currently address the problem by placing high-bandwidth memory, or HBM, around their AI computing processors. HBM provides much faster access to data than conventional system memory, making it an essential component of modern AI accelerators.
But there is a physical limit to how many memory packages can be connected around a processor using tiny electrical connections.
According to Volantis, even Nvidia's most advanced current offerings can accommodate eight HBM chips around each GPU because of the limited reach of those electrical connections. Volantis is attempting to remove that constraint by using optical connections instead.
Volantis Wants To Replace Electrical Links With Light
The startup's central idea is to transmit data between computing and memory components using beams of laser light.
Light can travel through optical systems with different physical characteristics from conventional electrical interconnects. Volantis believes this could allow memory components to be positioned farther from the central processor while maintaining very high-speed communication.
The company's proposed architecture could therefore change how memory is physically arranged around an AI processor.
Rather than continuing to pack a relatively limited number of memory chips immediately around the GPU, the system could potentially connect many more memory devices using optical links.
Volantis says its approach could support up to 220 memory chips around a GPU. That is a company claim about technology under development, not an independently demonstrated commercial result.
The Technology Behind The Idea
Volantis is using vertical-cavity surface-emitting lasers, commonly known as VCSELs, to create the optical connections.
VCSELs are not an entirely new semiconductor technology.
They are already used in consumer electronics, including facial-recognition systems found in many Apple devices. Apple has also invested heavily in strengthening the supply chain associated with the technology, according to Reuters.
That existing ecosystem is strategically important to Volantis.
Developing a completely new optical component would create another manufacturing challenge for a young semiconductor company. By building around a technology that already has an established supply chain, Volantis hopes to reduce some of the obstacles involved in commercializing its architecture.
The approach does not eliminate the difficulty of advanced chip packaging. Instead, the company is attempting to combine established components in a new configuration.
The $88 Million Funding Round
Volantis announced that it has raised $88 million in venture capital.
The financing round was led by Lachy Groom, a former Stripe executive, and Abstract Ventures. John Doerr, an early investor in companies including Google and Amazon, also participated. Other investors included VXI Capital, Triatomic and Susa Ventures. Angel investors included AI podcaster Dwarkesh Patel, AI chip veteran Naveen Rao and Anthropic researcher Sholto Douglas.
The investor lineup is notable because it combines traditional venture capital with people who have direct connections to software, artificial intelligence and semiconductor development.
That mix reflects the increasingly important role of specialized hardware startups in the AI ecosystem.
As AI companies spend heavily on computing infrastructure, investors are looking beyond the dominant GPU manufacturers for technologies that could solve the next bottlenecks in the hardware stack.
What Volantis Plans To Build
Volantis aims to bring a chip based on its technology to market next year.
Chief Executive Officer and co-founder Tapa Ghosh told Reuters that the company hopes the chip could improve AI workloads such as coding. The company is attempting to accomplish this without requiring an entirely new manufacturing approach.
The timetable is ambitious.
Semiconductor products must pass through multiple stages before reaching commercial deployment. Designing the architecture is only the beginning. Engineering validation, packaging, manufacturing, testing, software compatibility and customer qualification can all affect the schedule.
For Volantis, the next year will therefore be an important test of whether its optical-memory concept can move from a promising architecture to a manufacturable product.
Why Memory Bandwidth Matters So Much For AI
AI models place unusually demanding requirements on memory systems.
During inference, a processor may repeatedly access model parameters while generating an answer. During training, enormous quantities of data and model information move continuously between processing units and memory.
As models become larger and more sophisticated, the amount of information that must be accessed can increase significantly.
This means that improving the processor alone is not enough.
If memory cannot supply information quickly enough, the additional computing power may not translate directly into proportional real-world performance.
That is why HBM has become such a critical part of the AI semiconductor supply chain.
Volantis is effectively targeting the connection between these two sides of the system rather than attempting to compete directly with Nvidia or AMD by building a conventional GPU.
A Different Way To Think About AI Chip Scaling
The conventional approach to improving AI hardware often focuses on building larger or faster processors.
Volantis is pursuing a different strategy: change the architecture surrounding the processor so that it can access more memory.
If successful, that could create a different path to increasing AI system performance.
Instead of relying only on increasingly sophisticated processing cores, future accelerators could gain performance by expanding the amount and accessibility of memory connected to those cores.
This distinction matters because AI workloads can be constrained by data movement as much as by mathematical computation.
Why Optical Interconnects Are Attractive
Optical communication has already become important in data centers and telecommunications because light can move large amounts of information efficiently over distance.
The challenge has been bringing optical technology closer to the processor and integrating it into highly dense semiconductor packages.
Volantis believes that recent advances in packaging and the existing availability of VCSEL components make this increasingly practical.
The company is not claiming that optical connections will automatically solve every AI hardware problem. Instead, it is targeting a specific physical limitation: the reach of electrical connections between processors and memory.
Removing that constraint could allow engineers to rethink how memory is distributed around AI computing systems.
The Potential 220-Memory-Chip Architecture
Volantis' most ambitious claim is that its optical approach could support up to 220 memory chips around a GPU.
That would represent a dramatic increase compared with the eight-memory-chip configuration Volantis says is possible in Nvidia's current leading products.
However, the number should be understood as a target enabled by the company's proposed architecture, rather than a commercially deployed system already operating at that scale.
Increasing the number of memory devices also introduces other engineering questions.
- Power: More memory and optical components require additional energy.
- Packaging: Dense systems require precise manufacturing and thermal management.
- Latency: Adding physical distance between memory and processors must not undermine performance.
- Reliability: Large numbers of interconnects increase system complexity.
- Manufacturing: The architecture must be producible at high volume and competitive cost.
- Software: Hardware improvements must translate into useful performance for real AI workloads.
The commercial value of the technology will ultimately depend on how well it handles those constraints.
Why Existing VCSEL Supply Chains Could Help
One of Volantis' advantages is that VCSELs already exist in large-scale commercial applications.
That means the company is not starting with an entirely unproven optical component.
VCSELs have been deployed in consumer devices for years, including systems used for facial recognition. This established manufacturing ecosystem could potentially make it easier to source components and scale production than if Volantis were developing a completely new laser technology. 6
There is still a major difference between producing optical components for consumer electronics and integrating them into advanced AI accelerators.
AI chips operate under demanding conditions involving extremely high data rates, thermal loads and manufacturing tolerances.
Volantis must therefore prove that components that work well in consumer electronics can also meet the requirements of advanced AI computing.
Competition Will Be Intense
Volantis is entering a semiconductor market dominated by companies with enormous engineering resources and established customer relationships.
Nvidia and AMD already have sophisticated AI accelerator architectures and close relationships with the memory industry.
Other semiconductor companies and startups are also working on new approaches to memory bandwidth, packaging and interconnect technology.
Volantis therefore does not need to replace the existing AI hardware ecosystem to succeed.
A more realistic commercial opportunity could be to become an enabling technology supplier whose optical interconnects are incorporated into future AI accelerators or specialized computing systems.
The Startup Is Targeting A Bottleneck Rather Than A Market Leader
This strategy is important because competing directly against Nvidia on general-purpose AI processors would require enormous capital and years of software development.
Volantis is instead focusing on a specific technical problem.
If its solution works, it could potentially be valuable to multiple processor manufacturers rather than being tied to a single competing GPU architecture.
That gives the company a different potential path to market.
The company could become part of the broader semiconductor supply chain rather than attempting to build an entire AI computing platform on its own.
AI Hardware Startups Are Attracting More Capital
The funding also illustrates how investors are responding to the second phase of the AI hardware boom.
The first phase centered heavily on GPUs and the computing capacity required to train large AI models.
The next phase is increasingly focused on bottlenecks around those processors: memory, networking, packaging, cooling, power and data movement.
Companies that solve those constraints could benefit even if the dominant processor architectures remain unchanged.
For venture investors, that creates opportunities to finance specialized technologies that improve the economics of the entire AI computing stack.
The Importance Of Advanced Packaging
One of the most difficult parts of modern AI hardware is packaging.
Processors, memory and interconnects must be assembled into extremely dense systems while maintaining reliable electrical, optical and thermal performance.
Volantis is attempting to make its optical approach practical without requiring a completely unfamiliar manufacturing process.
That could be an important advantage if it reduces the number of new manufacturing steps needed for adoption.
But advanced packaging remains technically demanding. Even a design that looks relatively straightforward at the architectural level can become difficult when manufacturers must produce millions of components with extremely tight tolerances.
What Could Make Volantis Technology Valuable
Several factors could determine whether Volantis becomes an important semiconductor company.
- Demonstrated bandwidth: The technology must deliver a meaningful improvement in data movement.
- Energy efficiency: Optical communication must provide benefits without adding excessive power consumption.
- Manufacturing readiness: The architecture must be compatible with scalable semiconductor production.
- Cost: Customers must be able to deploy the technology economically.
- AI workload gains: Real applications must show measurable improvements.
- Customer adoption: Major chipmakers or AI infrastructure companies would need to validate the technology.
Until those conditions are demonstrated, Volantis remains a venture-backed semiconductor startup with an ambitious technical proposal rather than an established supplier.
The 2027 Product Target Is The Next Milestone
Volantis says it aims to deliver a chip next year.
That target gives investors and the semiconductor industry a clear milestone to watch.
A successful demonstration would provide evidence that the optical architecture can operate outside a laboratory environment. Commercial sampling would provide an even stronger signal, particularly if potential customers begin testing the technology in real AI systems.
The most important evidence will therefore come from actual hardware rather than additional fundraising.
Why The Technology Could Matter To AI Coding
Volantis specifically sees AI coding as one potential workload that could benefit from its technology.
Coding models can involve large amounts of contextual information and complex inference workloads. Improving the speed at which processors access model data could potentially improve throughput or responsiveness.
However, Volantis has not yet demonstrated that its proposed architecture will produce a specific performance improvement across commercial coding workloads.
The company's current opportunity should therefore be viewed as a hardware development project whose real-world impact remains to be validated.
What Investors Should Watch Next
The company's progress can be evaluated through several concrete milestones.
- Prototype completion: Whether Volantis produces functioning silicon on schedule.
- Independent performance testing: Whether third-party or customer testing confirms its bandwidth claims.
- Manufacturing partnerships: Whether the startup secures the partners needed for volume production.
- AI accelerator adoption: Whether established chip companies evaluate or adopt the technology.
- Power and thermal results: Whether the optical architecture provides practical system-level advantages.
- Commercial customers: Whether AI infrastructure companies commit to using the technology.
The Bigger Race Is About Moving Data
The AI hardware industry is increasingly discovering that faster computation is only part of the problem.
Modern AI systems require enormous amounts of data to move between processors, memory, networking equipment and storage.
As processors become faster, those data-transfer requirements become more demanding.
That creates a paradox: improving a processor can make the surrounding infrastructure more important, because the faster processor can consume data faster than the existing memory and interconnect systems can provide it.
Volantis is betting that optical technology can help close that gap.
If the approach succeeds, the implications could extend beyond one startup. It could encourage semiconductor designers to rethink how processors and memory are physically connected in future AI systems.
A Small Startup Targeting A Huge AI Bottleneck
Volantis' $88 million funding round is significant because it shows that investors are willing to finance solutions to highly specialized AI hardware problems.
The company is not attempting to build another conventional GPU. Instead, it is targeting the communication layer between computation and memory, using laser-based links to overcome physical limitations associated with electrical connections.
The concept is promising but remains unproven at commercial scale.
The next major test will be whether Volantis can translate its architecture into working silicon and demonstrate measurable benefits on real AI workloads.
If it can, the startup could become part of a new generation of semiconductor suppliers helping AI systems scale beyond the limitations of today's processor-and-memory configurations.
The larger lesson for the AI industry is that the next major performance gains may not come solely from making processors bigger or faster. They may come from finding better ways to move information around those processors.
As AI models continue to expand, that distinction could become increasingly important. The companies that solve the movement of data may ultimately be just as important to the next generation of AI infrastructure as the companies that perform the calculations themselves.
Frequently Asked Questions
What is Volantis?
Volantis is a San Francisco-based semiconductor startup developing technology intended to improve communication between AI processors and memory chips. It announced an $88 million venture-capital funding round on October 1, 2026.
What problem is Volantis trying to solve?
The company is targeting the limited reach of electrical connections between AI processors and high-bandwidth memory. It believes optical links could allow processors to communicate with a much larger number of memory devices.
How does Volantis plan to use lasers?
Volantis is developing an architecture that uses vertical-cavity surface-emitting lasers, or VCSELs, to transmit data between computing and memory components using light rather than conventional electrical connections.
How many memory chips could the technology support?
Volantis says its proposed architecture could allow as many as 220 memory chips to be connected around a GPU. This is a company development target and has not yet been established as a commercial production capability.
Who invested in Volantis?
The $88 million round was led by Lachy Groom and Abstract Ventures. John Doerr, VXI Capital, Triatomic and Susa Ventures also participated, along with several angel investors connected to AI and semiconductor industries.
When does Volantis expect to have a chip?
CEO and co-founder Tapa Ghosh told Reuters that Volantis aims to bring a chip to market next year, meaning 2027. The schedule remains a company target and depends on successful development and manufacturing.
Why is this technology important for AI?
AI processors increasingly require very high memory bandwidth. If computing power grows faster than the ability to supply data to the processor, memory and interconnects can become performance bottlenecks. Volantis is attempting to address that problem by changing how processors communicate with memory.
Comments
Post a Comment