- VC Breakfast Club
- Posts
- Why photonic chips might be the next deep tech moonshot
Why photonic chips might be the next deep tech moonshot
The race to scale AI won't be won by algorithms alone, it will be powered by physics
Ines
May 04, 2025
Welcome to the VC Breakfast Club Newsletter - the local frontpage of UK venture.
Weāre a group of VCs writing across the UKās venture hubs. We help founders share their funding news and stitch together the UKās venture flywheel.
If youāre announcing a raise, let us know. If youāre fundraising, get in touch.
If youāre an early-stage VC who would like to write with us, reach out to Mike.
š” Why Photonic Chips Could Be Deep Techās Next Moonshot
The future of AI isnāt just about smarter models. Itās about smarter physics.
Traditional compute hardware is nearing its physical limits and energy use is skyrocketing, with trillion parameter AI models chewing through megawatts like it is popcorn. Training GPT-4 reportedly required over 50,000 MWh, while inference now consumes enough power daily to rival small nations. With cooling, interconnects, and bandwidth emerging as key energy bottlenecks, one solution is gaining traction in deep tech circles: photonic computing.
Todayās AI chips burn their energy moving data, not just computing it. As model sizes balloon, shuttling information has become a major power bottleneck.
š¦ Whatās the big idea?
Photonic processors use light instead of electrons to transmit data, while reducing resistance, latency and heat. This makes them far more energy-efficient, especially for AI workloads.
Last year, MIT developed a photonic chip that performed machine learning computations at a fraction of the energy used by traditional electronic hardware.
These chips were āfabricated using commercial foundry processesā - a key statement here, as widespread adoption depends on easy integration with the international labyrinth that makes up todayās compute supply chain.
š¦ From lab to reality
The next big hurdle? Integrating photonics with conventional compute hardware (like CMOS). That is exactly where research is headed, and academics are not alone in the light race!
Lightmatter is a venture pushing the boundaries with hybrid photonic-electronic chips like Envise. It integrates photonics into CMOS-compatible workflows, making photonics a more practical (and fundable) bet for near-term deployment.
Ayar Labs is tackling optical interconnects, not compute logic. Their technology moves data between chips at the speed of light, with 10x lower latency and power.
Optalysys, in the UK, is using light for analog Fourier transforms, tackling edge AI and scientific compute at ultra-low power.
Lightmatter photonic processor chip package.
šŖ Is it venture-backable?
With Mooreās Law plateauing and Dennard scaling long dead, the industry is looking for entirely new substrates. But physics is on its side. Where electrons are hitting scaling limits, photons bring speed, thermal stability, and low-loss transmission to the table.
Photonic computing still faces challenges: bulky components, limited on-chip memory, and complex design tooling. Yet momentum is building. The US Department of Energy has backed photonics startups. The EU is funding integrated photonics via the PhotonHub Europe initiative, part of Horizon 2020. And VCs are now treating photonic hardware as a solution to AIās mounting energy crisis, not just a novel chip play.
Lightmatter raised a $155M Series C and is working with hyperscalers. Ayar Labs has DOE backing and strategic partnerships with Intel and NVIDIA. Investors are backing energy-efficient compute pathways that AI needs to scale.
Make no mistake, photonic computing is not trying to replace silicon overnight. It is augmenting it.
For now, expect more hybrid stacks (photonics + electronics). And if one day someone builds the āNVIDIA of lightā, we may remember 2025 as a tipping point for photonic compute gaining real-world traction.
šļø Key takeaway
Photonic compute isnāt mainstream yet, but it is scuttling out of the lab to augment traditional compute. And if momentum keeps building, we may one day see full-stack computers run on light.
If you are working on photonic chips, hybrid compute, or energy-aware hardware, drop me a line. I would love to hear more about the space and your work.
šāāļø Ines
And that's a wrap! Tune in for Tuesday deep-dives & Sundays breakfast roundups.
Want to say hello? Hit reply - we'd love to hear from you!
š Follow the newsletterās LinkedIn Page
š„ If youāre announcing a raise, let us know.
š„ If youāre fundraising, get in touch.
š Email (Reply to this email)
If you havenāt already, make sure to sign up to get this in your inbox next week.
And if you like this, share it with a few friends!