Every time a hyperscaler spends a billion dollars on Nvidia GPUs, the surrounding infrastructure — cables, switches, transceivers, optical components — has to be upgraded to match. The smarter the GPU gets, the more the interconnect matters.
That’s the thesis driving Nvidia’s reported $6.5 billion photonics spending spree in recent months. The bottleneck in AI data centres is no longer the GPU. It’s the pipes connecting them.
🔍 THE BOTTOM LINE: Copper wiring is hitting physical limits inside AI data centres. As clusters scale to hundreds of thousands of GPUs, the interconnect — the roads between the buildings — becomes the single biggest constraint on performance. Nvidia sees this clearly and is buying its way into the companies building the optical replacement.
The Copper Ceiling
Think of an AI data centre as a city. The GPUs are the buildings where all the work happens. For those buildings to function, you need roads — fast roads that can carry enormous traffic without congestion.
Those roads are now clogged. Traditional copper wiring can only push electrical signals so far before the signal degrades, the heat spikes, and power consumption explodes. As clusters scale to hundreds of thousands of GPUs — the scale required for frontier model training — copper hits a wall.
The solution is light. Fibre optic connections that move data using photons instead of electrons, which is faster, cooler, and far more energy efficient.
Jensen Huang made this explicit at Computex 2026: copper works as long as physically possible, but at greater distances and larger scale, optics takes over.
The $6.5 Billion Shopping Spree
CNBC reported in March that Nvidia committed $4 billion to two photonics companies alone — Coherent and Lumentum. Since then, the spending has accelerated. The full picture, as outlined by AI semiconductor analyst @MelvinInvests on X:
- Coherent (COHR): $2 billion Nvidia investment. Makes the lasers, transceivers, and optical components at the foundation of all fibre optic communications. Customer order books extending to 2028.
- Lumentum (LITE): $2 billion Nvidia investment. Laser and photonics components.
- Marvell (MRVL): $2 billion Nvidia investment. Jensen Huang called Marvell “the next trillion dollar company” at Computex 2026. Marvell also acquired Celestial AI for $3.25 billion, gaining photonic fabric technology delivering 16 terabits per second of bandwidth.
- Corning (GLW): $500 million Nvidia investment. Known for phone glass, now building optical connectivity for data centres. Up over 100% year to date.
That’s at least $6.5 billion committed to photonics in recent months. The message is clear: Nvidia believes the companies building the roads between the GPUs may end up being just as valuable as the companies building the GPUs themselves.
The Pure Plays
Two companies are the most direct bets on the interconnect thesis:
Credo Semiconductor (CRDO) makes high-speed cables and optical chips that connect GPUs inside data centre racks. Revenue tripled in fiscal 2026 to $1.3 billion, growing 272% year over year at its peak. Four of the world’s largest hyperscalers each individually account for more than 10% of Credo’s revenue.
Astera Labs (ALAB) solves the connection problem between different chip types — the PCIe and connectivity chips that manage data flow between GPUs, CPUs, and memory without errors or slowdowns. Revenue grew 93% year over year to $308 million in Q1 2026 alone.
Both are growing at rates that suggest the interconnect market is not a future opportunity — it’s a present crisis.
Why This Matters Now
The interconnect bottleneck is a direct consequence of scale. When you had 8 GPUs in a rack, copper was fine. When you have 100,000 GPUs in a cluster — the scale required for frontier model training — the distance data has to travel between GPUs, and the bandwidth required to keep them all synchronised, breaks copper.
We’ve covered the power bottleneck and the grid interconnection queue in previous articles. The interconnect bottleneck is the third leg of the AI infrastructure constraint: you can generate the power, you can build the data centre, but if the pipes between the chips can’t move data fast enough, the GPUs sit idle.
This is also why Nvidia is investing rather than just partnering. Owning stakes in the photonics supply chain gives Nvidia visibility into the one part of the AI stack it doesn’t fully control today. Mellanox gave Nvidia InfiniBand. The photonics investments give Nvidia the next layer: the optical connections that replace copper when copper runs out.
❓ FAQ
Is this investment advice? No. This is an analysis of Nvidia’s strategic positioning in the AI infrastructure stack. Stock performance depends on many factors beyond the technology thesis.
Are these numbers verified? The CNBC report on the $4 billion Coherent+Lumentum investment was confirmed via Hacker News linking to CNBC’s March 2026 article. The Marvell and Corning investment figures were cited by @MelvinInvests on X but not independently verified by a second source. The Computex 2026 claims (Jensen calling Marvell “next trillion dollar company”) have not been independently confirmed.
What about Chinese photonics? China is investing heavily in optical interconnect for its domestic AI chip programme. Huawei’s Ascend chips have historically lagged specifically on interconnect performance, which is the capability gap this technology aims to close.
When does copper actually run out? It’s a gradient, not a cliff. At current scaling rates, copper interconnects are viable for intra-rack connections but increasingly problematic for inter-rack and inter-row connections in large clusters. The transition to optical is already underway.
🔍 THE BOTTOM LINE: The AI industry’s obsession with GPU counts has obscured the real bottleneck: moving data between those GPUs. Nvidia’s $6.5 billion photonics spending spree says the next phase of AI infrastructure won’t be about building faster chips — it’ll be about building faster roads. Light is the future of AI interconnect, and Nvidia is buying its way to the front of that queue.
