Nvidia-backed photonics startup Ayar Labs fills its wallet to mass-produce CPO chiplets

Nvidia-backed photonics startup Ayar Labs fills its wallet to mass-produce CPO chiplets

Summary

Ayar Labs has closed a $500m Series E to accelerate high-volume production of its co-packaged optics (CPO) TeraPHY chiplets. The company says embedding optical I/O directly into GPU and accelerator packages delivers much higher bandwidth and far lower power than pluggable optics or high-speed copper interconnects beyond ~800 Gbps.

Ayar has built prototypes and reference designs with partners including Intel, DARPA, GUC and Alchip; one reference using eight next‑gen chiplets claims more than 200 Tbps aggregate per package. The funding—led by Neuberger Berman and including Nvidia and MediaTek—will scale production, test capacity and global ops, starting with a new office in Hsinchu, Taiwan.

This round arrives as Nvidia and others are investing heavily in photonics manufacturing capacity, and follows broader industry moves such as Lightmatter’s photonic interposer work and Marvell’s acquisition of Celestial AI.

Key Points

• $500m Series E raised to push Ayar Labs into mass production of co-packaged optics (CPO) TeraPHY chiplets.
• Optical chiplets offer much higher bandwidth and lower power than pluggable optics and overcome copper limits above ~800 Gbps.
• Reference designs claim >200 Tbps aggregate bandwidth per package; Ayar aims to scale to very large “scale-up” domains (targeting tens of thousands of GPU dies) while keeping rack power manageable.
• Funding led by Neuberger Berman; investors include Nvidia (returning investor) and MediaTek; expansion includes a Hsinchu, Taiwan office and ramped production/testing capacity.
• Part of a wider industry shift: Nvidia’s substantial investments in photonics suppliers and rival startups (e.g. Lightmatter) point to growing momentum and consolidation in photonic interconnects for AI infrastructure.

Why should I read this?

Because if you care about the next jump in AI datacentre performance or the bill for power and cooling, this matters. Ayar just got the cash it needs to try to turn optical chiplets from demo-stage cleverness into something you can actually buy and deploy — and Nvidia’s fingerprints are all over this story. Short version: it could change how GPU farms are wired and powered.

Author style

Punchy: this isn’t just another funding headline. With heavyweights backing production and bold claims about >200Tbps packages and 10,000‑GPU scale-up targets, this is a signal event for AI infrastructure. Read the detail if you want to understand where datacentre interconnects are heading.

Context and relevance

Datacentres face tightening power and density constraints while AI model training and inference demand ever greater aggregate bandwidth. Co‑packaged optics aim to bridge the gap between short‑reach copper (limited at very high speeds) and longer‑reach pluggable optics (which cost power and add latency). Mass production and supply‑chain scale are the bottlenecks — the new funding and Nvidia’s concurrent investments in photonics suppliers indicate the industry is moving to remove that bottleneck. For vendors, hyperscalers and purchasers, this could affect architecture choices, procurement and long‑term power budgets.

Source

Source: https://go.theregister.com/feed/www.theregister.com/2026/03/03/ayar_labs_500m/