Ayar Labs adds $150M more as copper interconnects hit the wall
Two ends of the AI buildout moved today: the physical layer and the policy layer. An optical chipmaker pulled in another $150 million to rip copper out of AI clusters, and China's industry ministry put hard 2028 deadlines on rebuilding its software industry around AI.
Ayar Labs, the startup building optical links between AI chips, has raised an additional $150 million — bringing its 2026 funding to $650 million. According to Reuters, the Series E extension adds Taiwanese server manufacturer Wiwynn as a strategic investor and follows a $500 million round in March that was backed by Nvidia and AMD. The money is aimed at one deadline: qualifying co-packaged optics for volume production by the end of 2027, so customers can ramp their own products in the 2028–2029 window.
The problem Ayar is attacking is unfashionable but real: copper. Inside today's AI clusters, GPUs talk to each other over copper cables that lose signal over distance, draw serious power, and overheat when thousands of processors sit side by side — the bigger the cluster, the worse all three problems get. Co-packaged optics moves the fiber connection onto the chip package itself, cutting the energy and complexity limits that copper imposes on rack-scale designs. CEO Mark Wade put it bluntly: copper interconnect is "becoming the limiting factor for AI scale-up."
The take: the AI hardware race is quietly becoming an interconnect race, not just a GPU race. Whoever moves bits between accelerators with the least power gets to build the bigger cluster, and bigger clusters are what the next generation of foundation models runs on. That is why Wiwynn — which assembles the racks hyperscalers buy — is on the cap table rather than just another VC fund. We covered the memory side of the same squeeze this week — Positron bets $875 million that inference doesn't need expensive memory — and China's pricing version of it in China's AI chip prices jump 50% as the memory shortage bites.
China's Ministry of Industry and Information Technology has published an "AI Plus Software" action plan with numbered targets for 2028 and 2030. Per CCTV, the plan calls for intelligent programming tools and AI development platforms to reach 20,000 large software enterprises by 2028, alongside 100 software-firm intelligent upgrade projects and 100 flagship agent-software applications in key industries, plus at least five quality open-source projects incubated. By 2030, key software is expected to be fully AI-upgraded, with agent software and intelligent services as the industry's new growth engines.
Beijing's mechanism here is worth naming: state targets pull procurement, procurement pulls revenue, and revenue feeds the capital markets that are currently paying up for domestic AI silicon. It is the software half of a stack China has been assembling from the chip up, and unlike a US industrial policy fight, it gets executed through procurement rather than legislation.
What to watch: whether Ayar's optics clear volume qualification by end-2027 — every hyperscaler's next-generation cluster design freezes well before that.
The AI buildout is now bottlenecked by copper, memory and power rather than GPU supply. Which constraint breaks first — tell us in the comments.
