Summary: Ayar Labs closed a $500 million Series E on March 3, 2026 at a $3.75 billion post-money valuation to scale co-packaged optics that replace copper interconnect in AI data centers. The article frames the round as targeting the power and bandwidth bottleneck in large GPU clusters and points to the 8-Tbps-class TeraPHY chiplets and SuperNova light source as the product core.
Ayar Labs, a San Jose company, closed a $500 million Series E on March 3, 2026 — a $3.75 billion post-money valuation that lifts the company’s total fundraising to roughly $870 million. Its product portfolio serves exactly one purpose: pushing copper out of AI data centers in favor of light. In Crunchbase’s weekly ranking of the biggest U.S. rounds (February 28 to March 6), Ayar Labs landed next to Sierra — and that placement carries a message: while every headline chases the next model, investors are putting money on a bottleneck made of physics.
What makes this round worth watching? The plainest answer comes from CEO Mark Wade: “AI infrastructure is hitting a power wall driven by interconnect inefficiency.” The GPUs themselves have not stopped improving; what has stalled is the kilometers of copper needed to tie thousands of them into one cluster, which now cap bandwidth and power at the same time. This round is a wager on how soon that wall starts coming down.
Inside the $500M Series E: The Investor Lineup
Neuberger Berman leads the round and picks up a board observer seat in the process. First-time backers are Alchip Technologies (an ASIC vendor), MediaTek, ARK Invest, Insight Partners, the Qatar Investment Authority, Sequoia Global Equities, and 1789 Capital, while the existing financial investors — Advent Global Opportunities, Boardman Bay Capital Management, IAG Capital Partners, Light Street Capital, and Playground Global — each signed on again.
The strategic column is the one to study: AMD Ventures and NVIDIA both returned. When the two dominant GPU vendors fund a company built to displace copper, they are effectively conceding that interconnect is a gate to clear before they can sell many more accelerators. There is history here as well — Ayar Labs’ $155 million Series D in 2024 already listed AMD, Intel Capital, and NVIDIA among its backers, making this the second time the chip ecosystem has put its name behind the company.
The spending plan is spelled out: scale high-volume production and test capacity for the co-packaged optics (CPO) solution, grow the global footprint (a new office in Hsinchu, Taiwan, among the additions), deepen ecosystem partnerships, and get CPO deployed faster.
The Power Wall: Why Interconnect Is the Story
Cluster scaling runs on arithmetic that offers no discounts. When models grow, scale-up networks must hand every GPU more bandwidth, and every rung up the electrical bandwidth ladder brings steeper power costs and shorter reach. Copper answers to a physical ceiling: signal integrity, reach, and energy per bit all close in at once.
Wade reduces the whole problem to a single line: CPO lets “thousands of GPUs operate as a unified system.” From the capital side, Gabe Cahill, managing director at Neuberger Berman, supplies the counterpart: the AI buildout is “one of the largest capital deployment opportunities of our generation,” and “data center interconnect has quickly emerged as the most critical bottleneck.” No matter the pace of model progress, it is the physical wiring inside the building that decides whether all that compute ever fuses into a single machine.
TeraPHY: Putting the Optical Engine in the Accelerator’s Own Package
At the center of the product line sits the TeraPHY optical engine — described in HPCwire’s coverage as 8 Tbps-class chiplets — which pairs with the SuperNova light source to deliver a complete optical I/O solution. Everything turns on the phrase “co-packaged”: rather than residing in a rack-scale switch, the engine shares a package with the accelerator or switch silicon and performs the electrical-to-optical conversion across the shortest distance available.
The engineering path is deliberately conservative. TeraPHY keeps to the standard form factors, manufacturing, and packaging flows that major accelerator and switch vendors already run, which trims adoption friction — CPO can reach the market without the packaging ecosystem being rebuilt from scratch. The new investors tell part of that story: Alchip and MediaTek both sit deep in the ASIC ecosystem, so their investments double as advance work on integration, and the Hsinchu office stations engineering staff in the middle of the Asian supply chain.
Three Takeaways for Procurement and Architecture Teams
Three consequences follow. First, optical interconnect has left the research agenda for the production calendar: the round’s stated use of funds — scaling volume production and test capacity — sets up 2026 as the year CPO moves from samples to deployments, so teams drawing up next-generation racks should already be feeding optical options into their power and bandwidth models. Second, the strategic roster works as a weathervane: once NVIDIA and AMD are both bankrolling an interconnect startup, how well accelerator roadmaps compose with optical solutions will weigh directly on next-generation cluster purchasing. Third, anyone building cloud or sovereign AI should read interconnect efficiency as electricity and rack density — with power now the binding limit on data center expansion, this architectural constraint arrives earlier than the model choice does.