Bold claim: the race to power AI at scale hinges on the unseen links that stitch data centers together. And this is where Mesh Optical Technologies enters the story. Three former SpaceX engineers—Travis Brashears, Cameron Ramos, and Serena Grown-Haeberli—turned a shared experience building Starlink’s optical links into a startup mission to reimagine data-center interconnects. Now based in Los Angeles, they just announced a $50 million Series A led by Thrive Capital.
Mesh Optical Technologies plans to mass-produce optical transceivers—devices that convert optical signals from fiber or laser into electrical signals for computers. The trio’s aha moment came while designing a next-generation, compute-hungry SpaceX satellite constellation; they assessed the current transceiver market and saw meaningful gaps that they could fill with a more scalable approach.
Why does this matter for data centers? Optical transceivers are crucial for coordinating large-scale GPU clusters used in training and running deep learning models. A familiar metric helps frame the opportunity: a typical multi-GPU cluster contains far more transceivers than GPUs, often multiplying the total by four to five. Mesh aims to produce about a thousand transceivers daily within the next year to qualify for bulk orders in 2027 and 2028.
One established U.S. supplier, AOI, landed a roughly $4 billion contract to supply components for AWS data centers last year, underscoring the scale at which these parts are bought. Mesh believes its approach—co-locating design and manufacturing outside of China and streamlining the supply chain—could offer both resilience and cost advantages, a stance that also aligns with broader national-security considerations about dependence on competing countries for AI infrastructure.
Thrive Capital partner Philip Clark framed the strategic angle at TechCrunch: if AI is as transformative as many believe, then routing critical AI data-center hardware through non-aligned suppliers could be essential for scalable, secure growth. In the near term, Mesh is tackling interconnect challenges to keep AI systems expanding smoothly.
The founders acknowledge the biggest hurdle ahead: achieving lights-out, automated manufacturing in the United States. Much of this expertise is concentrated in China, where even European equipment makers encounter ecosystem hurdles—some vendors’ forms still request a Chinese company registration number. Mesh’s plan is to co-locate design and manufacturing to drive efficiency and lower costs, even claiming a current design that eliminates a power-hungry component and could cut GPU cluster power consumption by 3%–5%—a meaningful saving for hyperscalers chasing every watt.
Beyond data centers, Mesh sees optical wavelength communications as a broader shift in how information travels. Brashears puts it plainly: the world has long prioritized radio frequency; Mesh aspires to lead a transition from RF to photonics and to interconnect not just computers but a wide array of devices—starting with data centers.
About the author: Tim Fernholz writes on technology, finance, and public policy and has chronicled the rise of the private space industry. He is the author of Rocket Billionaires: Elon Musk, Jeff Bezos and the New Space Race. For inquiries, you can reach him at tim.fernholz@techcrunch.com or via Signal at tim_fernholz.21.
