| |
| Case File № DRT-ED-LIGHT · Special File · The Speed of Light |
| |
| The Next AI Bottleneck Isn’t the Chip. It’s the Wire — and the Money Is Moving to Light. |
| |
| While the world argued about an AI bubble, Nvidia quietly committed billions to companies that move data with light instead of copper. Here’s why the biggest player in AI is betting on photonics, and why most investors haven’t noticed. |
| |
| Threat level 3/5 — Guarded |
| |
|
| |
| |
| |
| Most of the AI debate is about chips: who makes the fastest GPU, who gets the next shipment. But inside the world’s biggest data centers, engineers are fighting a different battle. The problem is no longer how fast a chip can think. It’s how fast thousands of chips can talk to each other. The answer the industry is betting on is light. |
| |
| The bottleneck |
| |
| | Why Copper Is Running Out of Road |
|
| |
|
| |
| Today, most signals inside AI systems travel over copper. As data rates climb, copper loses signal quality and burns more power. It still works inside a single server rack, but it struggles across multiple racks. Nvidia’s planned Vera Rubin Ultra NVL576 links 576 GPUs across eight racks: copper inside the racks, optical links between them. |
| |
|
| |
| Every leap in AI model size means more chips working as one machine. That machine is only as fast as its slowest connection. Once the connections become the limit, whoever controls the fastest, coolest-running links controls the next phase of the buildout. |
| |
| Follow the money |
| |
| | Where the Smart Money Has Already Moved |
|
| |
Coherent $2 billion from Nvidia Announced in March 2026. Optical components. |
|
Lumentum $2 billion from Nvidia Same week. Lasers and optical modules. |
|
Marvell $2 billion from Nvidia Chips that drive optical data links. |
|
Corning Up to $3.2 billion Equity warrants plus multi-year purchase deals, funding new plants in North Carolina and Texas. Target: a tenfold increase in optical connectivity capacity. |
|
Ayar Labs Part of a $500 million round Its co-packaged optics module is expected to begin sampling in late 2026 at 6.4 terabits per second. |
|
| |
|
| |
| Altogether, Nvidia has committed roughly $6.5 billion or more to photonics since March, on top of the optical switching platforms it introduced in 2025 with TSMC, Coherent, Lumentum, Corning and Foxconn. |
| |
| Decoder |
| |
| | What a Chipmaker Buying Light Really Means |
|
| |
| | “Copper within racks, optical between them.” — How the industry describes the design of Nvidia’s next-generation systems |
|
| |
| What it says |
| |
| AI clusters are getting too big for copper. |
| |
| What it means |
| |
| Our theory: the world’s most valuable chip company doesn’t write multibillion-dollar checks for a side project. It writes them for a bottleneck it can’t afford to have. When the leader starts funding its suppliers’ factories, it’s telling you where the next shortage is. |
| |
| Winners & losers |
| |
| | Who Gains if Light Replaces Copper |
|
| |
Optical component makers Lasers, modulators and optical engines move from niche parts to core infrastructure. |
|
| |
Fiber and glass producers More links between racks means far more fiber inside every data center. |
|
| |
Copper-based connectivity Still dominant inside racks, but losing the longer links as clusters grow. |
|
| |
Power-hungry data centers Optical links use less energy per bit, and energy is now AI’s other bottleneck. |
|
| |
| |
|
| |
| This is a shift in AI’s architecture. The first wave rewarded the companies making the brains. The next one may reward the companies making the nervous system. The big money has moved, and most investors are still watching the chips. |
| |
The riddle · answer at the end of the file Nvidia makes the most valuable chips in the world. Why would it spend billions helping other companies build parts that aren’t chips at all? |
|
| |
| |
|
| |
| Riddle answer |
| |
| Because a chip can only go as fast as its connections. If the links between GPUs can’t keep up, Nvidia’s customers can’t use all the computing power they paid for, and they stop buying more. Our theory: by funding the optical supply chain, Nvidia is protecting its own growth. That makes the companies building those links one of the most direct ways into AI’s next wave. |
| |
