Ayar Labs and the Horizontal Cap Table: Why AMD, Intel, and NVIDIA Are Funding the Same Interconnect Supplier

$650 million of primary capital in one year, a cap table of direct competitors, and a use-of-proceeds statement aimed squarely at the place optical interconnect has always broken down — manufacturing at volume.

What Happened

On September 10, 2026, Ayar Labs announced an additional $150 million in primary capital, bringing its 2026 total to $650 million following the $500 million it had disclosed in March. Unite.AI covered the announcement; trade reporting was picked up across the sector. Wiwynn — a cloud-infrastructure and data-center provider — joins as a newly disclosed strategic backer alongside the existing group: Alchip, AMD, Intel, MediaTek, and NVIDIA. The company disclosed no per-investor amounts, no lead investor, and no revised valuation.

Separately, Reuters reported a roughly $225 million purchase of shares from early employees and investors — a secondary transaction, meaning it pays selling shareholders rather than the company — that valued Ayar Labs above $5 billion. That figure and its sourcing belong to Reuters’ reporting on the secondary; Ayar Labs’ own announcement, as Converge Digest noted explicitly, disclosed no revised valuation. The two transactions are distinct and must not be summed: $650 million went into the business this year, and a separate share sale between investors set a mark above $5 billion.

The stated use of proceeds is specific: transition to high-volume manufacturing — covering product development, validation, and scaling the manufacturing ecosystem — plus a new design center in Bengaluru, India. No revenue figures, customer names, or shipment data were disclosed. These are stated intentions, not reported achievements.

What Co-Packaged Optics Actually Solves

Co-packaged optics places the optical interconnect in the same package as the processor. Data moves between chips, between boards, and between racks over light rather than electrical traces. The constraint it addresses is not how fast a chip computes — it is how fast data can be moved to and between chips at scale. At sufficient accelerator density, electrical interconnects become the bottleneck: bandwidth-limited, power-hungry, and heat-generating. Ayar Labs describes its own goal as AI scale-up beyond the rack, which is that problem stated from the inside.

The Structural Read

The chokepoint in AI infrastructure keeps moving outward, and the capital flow each week is the clearest map of where it has arrived. The sequence is instructive. Memory came first: Reuters reported Chinese accelerator prices rising on a worldwide high-bandwidth-memory squeeze, while Positron raised to build inference silicon that sidesteps HBM entirely by using commodity LPDDR5X. Then power and land: NVIDIA franchised its AI-factory blueprint to eight Australian operators who supply the construction capital and hold the grid positions themselves. What sits past both of those is the problem of moving the data.

Once an operator has enough memory and enough megawatts, the binding constraint becomes getting bytes between packages, between boards, and between racks quickly enough that expensive accelerators are not sitting idle waiting for them. That is precisely where co-packaged optics competes. And Ayar Labs is not the only large optics round this week: Celero announced a $275 million Series C in coherent optics on September 8 — two days before Ayar’s September 10 release. Capital arriving twice in one week at the same infrastructure layer is how you can read which bottleneck has become binding without waiting for anyone to announce it. Several of those comparisons — the HBM squeeze, the Australian franchise, Positron — are sourced reports rather than confirmed closed transactions, and should be read that way.

The use-of-proceeds statement is the third signal worth reading carefully. The money is not for proving the physics of co-packaged optics — that question has been answered. It is for the transition to high-volume manufacturing: validation, scaling the manufacturing ecosystem, building a design center in Bengaluru. That marks a changed question. CPO spent years answering whether it works at all. $650 million of primary capital inside a single year is the cost of answering whether it can be built at volume, reliably, on somebody else’s production schedule. Manufacturing scale-up is precisely where optical interconnect has historically stalled — so the capital is aimed at the right risk. Aimed at, not past it. Nothing in the announcement establishes that co-packaged optics is shipping in volume, qualified by any named customer, or designed into any named product.

Three Implications

The Bottom Line

The most interesting thing about Ayar Labs raising $650 million in primary capital in 2026 is not the number — it is the names beside it. When AMD, Intel, and NVIDIA co-invest in the same interconnect supplier, they are not being collegial; they are disclosing, in the most direct way available, that moving data between chips has become a shared problem that none of them can solve alone and none of them can afford to let a competitor solve for them. The chokepoint has moved outward past memory and past power, and the capital is following it. Whether $650 million is enough to cross the manufacturing threshold where optical interconnect has historically stalled is the only question that matters from here — and this announcement does not answer it.

Sources: Unite.AI — Ayar Labs $150M announcement coverage · Reuters reporting on the ~$225M secondary transaction and above-$5B valuation mark (picked up by SiliconANGLE,

91,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.