Ayar Labs Raises $500M to Mass-Produce Optical Chips That Connect Thousands of GPUs

Ayar Labs Secures Major Funding Round to Scale Silicon Photonics Production

Nvidia-backed Ayar Labs announced on Tuesday it has raised $500 million in a Series E funding round at a valuation of $3.75 billion, bringing the company's total funding to $870 million2

2

. The funding round was led by investment firm Neuberger Berman and included new investors such as ARK Invest, Qatar Investment Authority, and 1789 Capital, alongside existing backers like Nvidia and Advanced Micro Devices2

2

. The cash infusion comes just one day after Nvidia announced $4 billion in investments into photonic networking providers Coherent and Lumentum, signaling strong momentum in the silicon photonics startup sector1

1

.

Co-Packaged Optics Technology Addresses Critical Data Center Bottlenecks

Founded in 2015, Ayar Labs specializes in developing chips that use light for data transmission through its TeraPHY chiplets, which provide an alternative to copper interconnects for chip-to-chip communications1

1

. Above 800 Gbps, copper interconnects are limited to a couple of meters and often require retimers to maintain acceptable error rates, constraining higher-speed connections like those in Nvidia's NVL72 systems to single racks1

1

. While pluggable optics support longer ranges for rack-to-rack communications, they consume significantly more power and introduce higher latency. By integrating TeraPHY chiplets directly into GPUs or accelerators, Ayar Labs' co-packaged optics designs can support substantially higher bandwidth while using a fraction of the power required by pluggable alternatives1

1

.

Massive Bandwidth Gains Position Technology to Accelerate AI Computing

Ayar Labs has collaborated with Taiwanese semiconductor design services provider Global Unichip Corp to develop reference designs based on its optical I/O chiplets1

1

. One reference design developed with Alchip uses eight of Ayar's next-generation TeraPHY chiplets and claims to support more than 200 Tbps of aggregate bandwidth per package—approximately five times more bandwidth than Nvidia's Rubin GPUs, which top out at 28.8 Tbps of bidirectional bandwidth1

1

. Crucially, since the interconnects are optical rather than electrical, the links aren't limited to a single rack. "We want to be able to scale up to 10,000 GPU dies connected in a scale-up domain, while keeping the rack power and power density to around 100kW," Ayar CTO Vladimir Stojanovic stated in an exclusive interview1

1

.

Strong Investor Appetite Reflects AI Infrastructure Race

Investor appetite in the AI ecosystem has remained strong as venture and private-equity firms bet on the technology to upend traditional workflows, driving steady funding into companies building the infrastructure3

3

. Ayar's technology aims to speed transmission by linking AI computing and memory chips, an increasingly valuable prospect as hyperscalers and governments spend hundreds of billions of dollars for achieving AI infrastructure supremacy2

2

. With data centers increasingly facing power constraints and the performance advantage of scale-up systems becoming clearer, venture capitalists are lining up to push photonics startups like Ayar over the finish line1

1

.

Competitive Landscape and Expansion Plans

Ayar Labs competes with firms such as Celestial AI, which raised $250 million last March before being acquired by Marvell Technology for roughly $3.25 billion in February, as well as Lumentum and Coherent2

2

1

1

. Lightmatter is another startup playing in the co-packaged optics space, having launched a photonic interposer alongside an optical I/O chiplet last year1

1

. Ayar plans to use the new funds to scale production and test capacity, expand global operations including at its new Hsinchu, Taiwan-based office, and accelerate the deployment of its co-packaged optics solution3

3

. The company has already validated its technology through multiple prototypes, including one built in collaboration with Intel and the Defense Advanced Research Projects Agency1

1

. As AI training and inference workloads continue to grow, the ability to connect thousands of GPUs with higher bandwidth and lower power consumption while managing latency and power consumption will be critical for scalability in next-generation AI infrastructure.