Copper Is the New Bottleneck: Ayar Labs Adds $150M to Hit $650M for 2026 as Co-Packaged Optics Go Manufacturing-Ready

For two years the AI industry has talked about compute as the scarce resource — GPUs first, then HBM memory, then power. On September 10, 2026, Ayar Labs put fresh money behind a bet on the next constraint in line: the copper wires that physically connect AI racks to each other are becoming the limiting factor for AI scale-up.

The San Jose-based company, the leading independent supplier of co-packaged optics (CPO) for AI clusters, announced it has secured an additional $150 million, bringing its total primary capital raised in 2026 to $650 million. The new money funds the less glamorous half of the photonics story: not another physics demo, but product development, validation, scaling the manufacturing ecosystem — and a new design center in Bengaluru, India.

What co-packaged optics actually solve

To understand why investors keep writing nine-figure checks to a photonics company, it helps to look at how AI clusters have physically changed.

A modern frontier training or inference run no longer fits in one rack. Systems have scaled from single accelerators to multi-rack architectures, and every link between them has historically been electrical — copper cables and SerDes chiplets pushing signals at ever-higher rates. Copper works, but physics is catching up: as bandwidth per link climbs, electrical I/O burns more power, hits reach limits, and adds complexity in exactly the place where AI data centers can least afford it. In a power-constrained facility, every watt spent moving bits between racks is a watt not spent computing.

Co-packaged optics attacks that problem by moving the optical transceiver inside the accelerator package itself, replacing short-reach copper with fiber at the point where bandwidth density matters most. Ayar Labs’ optical engines — built around its TeraPHY optical chiplet and SuperNova light source — bring high-bandwidth, energy-efficient connectivity closer to compute, letting system architects extend scale-up domains across racks while cutting the power and complexity that copper imposes.

“Copper interconnect is becoming the limiting factor for AI scale-up,” said Mark Wade, CEO and co-founder of Ayar Labs, in the announcement. “This funding allows us to move faster on manufacturing-ready co-packaged optics at scale. We’re deepening our work across the foundry and packaging ecosystem and growing our engineering capacity to support customer programs worldwide.”

That phrase — manufacturing-ready — is the operative one. CPO has spent years as a promising lab technology; the 2026 race is to make it producible in volume, with yield, at hyperscaler quality bars.

Follow the strategics: Wiwynn joins AMD, Intel, NVIDIA

Alongside the capital, Ayar Labs disclosed that Wiwynn — the Taiwan-based cloud IT and data-center infrastructure provider — made a strategic investment earlier this year. That matters more than the dollar amount suggests, because Wiwynn builds the rack-scale systems hyperscalers actually deploy, and its CEO tied the investment directly to the copper problem.

“Co-packaged optics is a foundational technology for the next generation of AI and cloud data centers,” said William Lin, President and CEO of Wiwynn. “Through our collaboration and investment with Ayar Labs, we see a clear path to new architectures that reduce the power and complexity limits of copper interconnects, powering rack-scale deployment for the next wave of hyperscale AI data centers.”

Wiwynn joins a strategic-backer roster that now reads like a supply chain map of the AI hardware industry: Alchip, AMD, Intel, MediaTek, and NVIDIA, alongside financial investors from the $500M Series E announced in March 2026 (led by Neuberger Berman, with ARK Invest, Insight Partners, Qatar Investment Authority, Sequoia Global Equities and others). When the companies that make accelerators, ASICs, foundry silicon, and the racks themselves all hold equity in your interconnect supplier, the technology stops being optional and starts being roadmap.

The company’s ecosystem positioning got another endorsement in June 2026, when Ayar Labs joined the NVIDIA NVLink Fusion ecosystem to bring co-packaged optics to rack-scale AI infrastructure — the clearest signal yet that optical scale-up is being designed into mainstream AI factory architectures rather than remaining a niche research direction.

The 2026 arc: $650M in nine months

The new tranche caps a fast-moving year for the 2015-founded company:

• March 3, 2026 — $500M Series E, led by Neuberger Berman, joined by strategic investors AMD, Alchip, MediaTek and NVIDIA, earmarked for volume production of CPO and global expansion including a new Hsinchu, Taiwan operation.
• March 11, 2026 — Partnership with Wiwynn to bring CPO into deployable rack-scale AI systems for hyperscale workloads.
• March 24, 2026 — Named to Fast Company’s Most Innovative Companies of 2026 list, top five in computing.
• June 2, 2026 — Joined the NVIDIA NVLink Fusion ecosystem.
• September 10, 2026 — Additional $150M disclosed, bringing 2026 primary capital to $650M; Wiwynn’s strategic investment revealed; Bengaluru Design Center announced.

That cadence — fund, partner, integrate, fund again — is the classic pattern of a supplier whose technology has crossed from “interesting” to “procurement-critical.”

Bengaluru: engineering where the talent is

The new Bengaluru Design Center will complement existing teams in San Jose and Hsinchu, and is charged with building end-to-end silicon product-development capability. It will be overseen by VP of silicon engineering Sankara Venkateswaran — the former Intel networking-silicon leader hired in February — with senior director of engineering Arun Balachandar as site lead.

“Our new Bengaluru Design Center supports our transition from technology development to commercialization and production,” Venkateswaran said. “As customer demand and the complexity of our product roadmap increase, expanding our global engineering capacity will help us execute faster, support multiple development programs, and collaborate more closely with partners and customers across the region.”

India’s role in AI infrastructure has mostly been discussed in terms of demand — data centers being built to serve a billion-plus users. Ayar Labs’ move is a reminder that the country is also becoming a design node in the global silicon supply chain, with deep IC-engineering talent pools that hardware companies can no longer ignore.

The bigger picture: interconnect is the new frontier

The funding lands in a week when the AI hardware conversation has been dominated by memory and power — HBM price hikes squeezing Chinese AI chipmakers, memory-first architectures like Positron’s Asimov raising $875M to sidestep HBM entirely, and gigawatt-scale AI factory announcements from Nvidia. Ayar Labs’ raise is the third leg of that stool: even if you solve compute and memory, the network fabric binding thousands of accelerators into one logical machine must itself scale in bandwidth per watt.

Copper’s physical limits are well understood; optics’ manufacturing challenges are the reason CPO took so long to commercialize. What has changed in 2026 is that the industry’s biggest buyers now treat power-per-bit as a first-order design constraint, and the ecosystem — foundries, OSATs, ASIC designers, system integrators — has organized around making optical I/O producible. Ayar Labs claims over $870M in total funding, 200+ employees, and 260+ patents, and its solution is built in partnership with the world’s leading semiconductor manufacturing ecosystem, including TSMC’s COUPE platform.

The bull case is straightforward: every new AI factory built from 2027 onward will need scale-up bandwidth that copper cannot economically deliver, and the company that ships manufacturing-ready CPO first wins a toll position on the entire buildout. The bear case is equally clear — the big accelerator vendors are building their own optical solutions in-house, and CPO yield at hyperscale remains unproven in volume.

Either way, the message from this raise is that the market has priced in the transition. Copper had a good fifty-year run at the heart of computing. In the AI factory, its job is quietly being given to light.