The End of Compute Is Light: Why NVIDIA Spent $4B to Lock In Two Optical Leaders, and the Next-Gen CPO Light Source Behind It

2 days ago

3 min read

As AI model parameter counts keep climbing, the data center's compute bottleneck has quietly shifted from compute to I/O. We are watching a physical limit approach: the bandwidth and power bottleneck of traditional copper is becoming insurmountable, and silicon photonics is moving from a "nice to have" to a "must have" for sustaining AI's version of Moore's Law.

Just yesterday, NVIDIA dropped a bombshell on the optical communications industry: it announced a $2B investment each in upstream optical giants Lumentum and Coherent, $4B in total. Why would the company that dominates GPUs and AI systems personally reach all the way upstream to suppliers of III-V materials and laser sources?

Most market commentary reads the deal as a simple capacity lock, a defensive move to make sure NVIDIA isn't choked by shortages in 2026, the first year of 1.6T optical module volume production. But stretch the timeline and compare it with the underlying technology roadmap revealed at ISSCC 2026 and in the upcoming OFC 2026 program, and the game turns out to run much deeper.

This is not just about buying a few production lines. It is about NVIDIA controlling the agenda for next-generation co-packaged optics (CPO), and even laying strategic infrastructure for the ultimate scale-up light source: the microresonator comb laser.

In this article we peel back the surface of the $4B deal, from capacity supply and supply chain restructuring down to a hardcore technical breakdown of CPO and microcombs, to show what endgame NVIDIA sees at the end of the road for light.

1. Securing 1.6T and 3.2T Capacity

As AI models' bandwidth needs grow exponentially, copper interconnect is nearing its physical limits. NVIDIA's investment includes "multibillion-dollar purchase commitments," focused on:

1.6T modules: 2026 is seen as the first year of 1.6T optical module volume production, and NVIDIA needs to lock in guaranteed laser component capacity from both suppliers.
Capacity assurance: Lumentum has announced it will use the funds to build a new fab to meet NVIDIA's large future shipment needs.

2. Accelerating the Commercialization of Silicon Photonics and Co-Packaged Optics (CPO)

At last year's Hot Chips, NVIDIA previewed that 2026 would be the turning point for its networking platforms (such as Spectrum-X and Quantum-X) to move to CPO:

CPO in practice: the investment lets NVIDIA work more deeply with Coherent and Lumentum on R&D, integrating laser sources and optical components directly into switch or GPU packages.
Lower power: CPO can significantly cut signal transmission power (an expected 3.5x improvement in energy efficiency), which is critical to easing the growing power crunch in AI data centers.

3. De-risking and U.S. Manufacturing

The partnership specifically emphasizes support for U.S.-based manufacturing. With global supply chains increasingly shaped by geopolitics, NVIDIA is helping both companies expand in the U.S. to reduce over-reliance on any single region for advanced optical components and to secure the strategic safety of AI infrastructure.

4. A Shift in Supply Chain Position: From Customer to Ecosystem Builder

This marks NVIDIA changing the rules of the game in optical communications:

Vertical integration: NVIDIA already owns networking chips and systems through its Mellanox acquisition, but for the most upstream laser sources and optical components it still depends on specialist vendors. This investment effectively "semi-binds" two optical leaders to NVIDIA's chariot.
Non-exclusive agreements: notably, both agreements are non-exclusive, leaving Lumentum and Coherent room to keep serving other cloud service providers (CSPs). But NVIDIA has clearly secured first priority on capacity and a strong voice on technology.

5. Positioning for Next-Gen Light Sources: The Rise of the Comb Laser

In the 1.6T era, and later 3.2T and 6.4T, packing more laser wavelengths into a CPO architecture runs into severe power, size and thermal challenges. Beyond conventional DFB laser arrays, NVIDIA is actively pursuing new laser technologies, and the most critical piece of that puzzle is very likely the comb laser (microresonator frequency comb).

The upcoming OFC 2026 program makes this strategy clear. NVIDIA and Solinide Photonics are jointly presenting a landmark paper (Th2A.13: High-Efficiency Silicon Nitride Microcombs for Co Packaged Optics), showing a high-efficiency silicon nitride dissipative Kerr soliton microcomb designed for CPO. The device achieves a record 69% optical conversion efficiency and generates 24 wavelength channels above 0 dBm with 199.1 GHz spacing, purpose-built for scalable data center interconnect.

As we analyzed earlier in NVIDIA ISSCC 2026 CPO Scale-up, the comb laser is the key to enabling

scale-up network architectures. A single high-power source can generate dozens of equally spaced wavelength carriers, sharply reducing laser count and packaging complexity.

How does this tie directly to the Lumentum and Coherent investments?

Driving this kind of high-efficiency silicon nitride frequency comb requires an extremely high-power, stable continuous-wave (CW) pump laser. Lumentum and Coherent are the dominant giants in manufacturing III-V materials such as indium phosphide (InP). NVIDIA's capital injection is bound to accelerate both companies' technology breakthroughs and capacity build-out for the ultra-high-power pump sources comb lasers need, paving the way for NVIDIA's ultimate multi-wavelength CPO solution.