Ayar Labs secures $500M to mass-produce photonic chiplets for next-gen AI clusters

Ayar Labs, the Nvidia-backed silicon photonics startup, has raised $500 million in Series E funding to accelerate mass production of its co-packaged optics (CPO) technology, marking a significant milestone in the evolution of AI infrastructure. The funding round, led by Neuberger Berman with participation from MediaTek and Nvidia, comes amid growing recognition that traditional copper interconnects are becoming a bottleneck for next-generation AI systems.

The copper bottleneck in AI infrastructure

As AI models grow exponentially in size and complexity, the limitations of copper interconnects are becoming increasingly apparent. Above 800 Gbps, copper connections are constrained to just a couple of meters and often require expensive retimers to maintain signal integrity. This creates a fundamental problem for AI clusters: while high-speed copper interconnects like those in Nvidia's NVL72 systems can handle intra-rack communications, they fall short when it comes to connecting racks together.

Pluggable optics offer a solution for longer distances but come with their own trade-offs—higher power consumption and increased latency. This is where Ayar Labs' TeraPHY chiplets enter the picture, promising to bridge the gap between copper's limitations and pluggable optics' inefficiencies.

TeraPHY: 5x the bandwidth of next-gen GPUs

The company's technology integrates optical I/O directly into the GPU or accelerator package, enabling significantly higher bandwidth while consuming a fraction of the power required by traditional pluggable solutions. According to Ayar, their next-generation TeraPHY chiplets can support more than 200 Tbps of aggregate bandwidth per package—approximately five times the bandwidth of Nvidia's upcoming Rubin GPUs, which top out at 28.8 Tbps of bidirectional bandwidth.

This leap in performance isn't just about raw speed. By using optical interconnects, Ayar's technology eliminates the distance limitations that constrain copper connections, enabling the possibility of connecting tens of thousands of GPUs across multiple racks while maintaining power efficiency.

From prototypes to production

Ayar Labs has spent years validating its technology through multiple prototypes, including collaborations with Intel and DARPA. More recently, the company partnered with Taiwanese semiconductor design services provider Global Unichip Corp (GUC) to develop reference designs based on its optical I/O chiplets. One notable design, developed with Alchip, combines eight of Ayar's next-gen TeraPHY chiplets to achieve the impressive 200 Tbps bandwidth figure.

The timing of this funding round is particularly significant. Just one day earlier, Nvidia announced it would inject $4 billion into photonic networking providers Coherent and Lumentum ($2 billion each) to scale up their manufacturing capacity. This coordinated investment wave signals the industry's recognition that silicon photonics will be essential for the next generation of AI infrastructure.

The scale-up advantage

Ayar's CTO, Vladimir Stojanovic, outlined an ambitious vision for the technology: connecting up to 10,000 GPU dies in a single scale-up domain while keeping rack power and power density around 100kW. This capability could enable AI clusters with unprecedented computational density, potentially revolutionizing how large-scale AI training and inference workloads are handled.

The power efficiency gains are particularly crucial as data centers face increasing constraints on energy consumption. By reducing the power required for chip-to-chip communications, Ayar's technology could help AI operators build more powerful systems without proportionally increasing their energy footprint.

A crowded but promising field

The photonics startup landscape is becoming increasingly competitive. Lightmatter, another player in the CPO space, launched a photonic interposer alongside an optical I/O chiplet last year that bears similarities to Ayar's TeraPHY. Meanwhile, established semiconductor companies are making strategic acquisitions to enter the market—Marvell Technology's $3.25 billion acquisition of Celestial AI in February demonstrates the high stakes involved.

Ayar Labs' funding round reflects both the promise and the challenges of bringing silicon photonics to market at scale. The $500 million infusion will enable the company to expand its global operations, beginning with a new office in Hsinchu, Taiwan, and to scale up high-volume production and test capacity.

Implications for the AI industry

This investment wave in silicon photonics startups signals a fundamental shift in how the industry approaches AI infrastructure. As traditional scaling approaches—packing more transistors onto chips—face physical limitations, innovations in interconnect technology become increasingly critical.

The ability to efficiently connect thousands of GPUs could enable AI models that are currently impractical due to communication bottlenecks. This has implications not just for training larger models but also for inference workloads that require massive parallel processing.

For Nvidia, which has backed Ayar Labs since 2022, this investment represents a strategic bet on the future of AI infrastructure. As the company continues to dominate the AI chip market, ensuring that its systems can scale efficiently across multiple racks becomes increasingly important for maintaining its competitive advantage.

Looking ahead

The photonics revolution in AI infrastructure is still in its early stages, but the momentum is building. With major players like Nvidia, MediaTek, and a host of venture capital firms betting big on companies like Ayar Labs, the next few years could see a fundamental transformation in how AI systems are built and scaled.

The challenge now lies in translating this technological promise into commercial reality. Mass production of silicon photonics components at the scale required for AI infrastructure represents a significant manufacturing challenge. However, with $500 million in fresh funding and the backing of industry giants, Ayar Labs appears well-positioned to tackle this challenge head-on.

As data centers continue to push the boundaries of what's possible with AI, the innovations being developed by Ayar Labs and its competitors may prove to be the key enabler for the next generation of artificial intelligence systems—systems that can train larger models, process more data, and ultimately deliver more capable AI applications to users around the world.