Exclusive: Enlightra emerges from stealth with $15m to speed up chip-to-chip communication using light

As artificial intelligence models grow larger and more complex, the industry’s biggest constraint is no longer how quickly chips can compute — but how quickly they can talk to each other.

Training today’s large language models requires thousands of GPUs to work in parallel, often spread across multiple racks or even entire data centres. While chipmakers have poured enormous resources into making individual processors faster and more powerful, the infrastructure that connects them has struggled to keep up. The result is a widening communication bottleneck that slows performance and drives up energy consumption.

“While everyone was fighting for the faster and bigger GPUs or TPUs [tensor processing units], this part was not completely omitted, but it was kind of overlooked,” Maxim Karpov, co-founder and co-CEO of Enlightra, tells Pathfounders. “And the gap between how fast you can compute, and how fast you can communicate the data between these GPUs has been growing in the past few years.”

Most of today’s GPU-to-GPU communication still relies on copper cables. As data rates increase, those connections become less energy efficient and work only over short distances. At extreme speeds, copper links can connect only a handful of GPUs at once — far short of what modern AI clusters require.

The alternative is light. Optical fibres can transmit far more data, over much longer distances, all while consuming less power. 

Enlightra, a startup that has been working on this problem since 2021, is now emerging from stealth to tackle this issue of data transfer.

The startup, founded by two long-time collaborators and researchers, has quietly raised $15m over the past few years to commercialise a new class of chip-scale photonic lasers designed to remove one of the key remaining bottlenecks in optical interconnects. Investors include Y Combinator, Runa Capital, Pegasus Tech Ventures, Protocol Labs, Halo Labs, Asymmetry Ventures and TRAC VC. 

Both academics by background, Enlightra’s founders went through Y Combinator and Intel Ignite in 2022 and 2023 respectively to learn as much as they could about how to turn their scientific discoveries into a viable startup.

At the core of Enlightra’s technology is a multi-wavelength laser architecture — sometimes referred to as “comb lasers” — that can generate many different colours of light from a single device. These colours appear like the teeth of a hair comb and allow multiple data streams to be transmitted simultaneously over the same optical fibre.

Today, optical systems typically rely on dozens of discrete lasers to achieve this effect, driving up cost, complexity and power usage. Enlightra replaces those components with a single silicon photonic chip that converts light from one laser into multiple wavelengths.

“Essentially, more colors is better,” Karpov says. And it helps with energy efficiency which is, he says, “a big issue for these AI clusters [because] they’re super energy hungry and they want to make sure that they are using their energy as efficiently as possible”.

The market opportunity is substantial. McKinsey estimates that the market for energy-efficient interconnects will reach $24bn by 2030, driven by AI workloads and the rapid expansion of data centre infrastructure. Major technology companies including NVIDIA, Google, Meta and Broadcom are all investing heavily in optical interconnects as data growth accelerates.

“AI is driving an optical revolution,” Dmitry Galperin, general partner at Runa Capital, said in a statement. “Enlightra’s multiwavelength lasers are a foundational technology for the next decade of high-performance computing, bringing the efficiency and scalability the industry urgently needs.”

While Karpov wouldn’t name Enlightra’s current customers, he said the company is already working with leading chipmakers. “I can tell you they are technology leaders developing chips for AI computing,” he said. “If you look at the top five companies developing AI chips, you’ll definitely find our customers.”

Operating in the same space but across the Atlantic, Xscape raised $44m Series A led by IAG Capital Partners last year, while Celestial closed a $175m Series C in March and Ayar Labs completed a $155m funding round that valued the company at over $1bn last December.

Company data: