By James Aspinwall, co-written by Alfred Pennyworth (my trusted AI) — March 3, 2026, 16:44
On March 2, 2026, NVIDIA announced a $2 billion strategic investment in Lumentum Holdings — a move that signals a fundamental shift in how AI infrastructure will be built. This isn’t about faster GPUs. It’s about the wires between them, and why those wires need to become beams of light.
The Problem: Copper Hit a Wall
NVIDIA’s GPUs are extraordinary. But a GPU is only as fast as its ability to talk to the GPU next to it. At scale — hundreds of thousands of accelerators in a single AI factory — the communication layer becomes the bottleneck, not the compute layer.
As one analyst put it: “AI scaling is no longer primarily a chip story. It is a communication story.”
Traditional copper interconnects hit a hard physical limit around 1.6-Terabit speeds. At that threshold, copper cables become too hot and too slow. The infamous “2-meter limit” means copper can’t effectively bridge between GPU racks without generating excessive heat and signal degradation. When your AI cluster needs thousands of simultaneous high-speed links per rack, copper simply can’t keep up.
The Solution: Silicon Photonics
Light doesn’t have copper’s problems. Optical interconnects transmit data using photons through glass fiber — faster, cooler, and with dramatically less power loss over distance.
The numbers are compelling:
- 40% reduction in AI networking power consumption
- 1.6-Terabit speeds today, with 3.2-Terabit projected by 2027-2028
- No distance-dependent signal degradation within data center scales
The endgame is Co-Packaged Optics (CPO) — optical transceivers integrated directly onto the chip package, enabling optical-to-the-chip architectures where GPUs communicate at the speed of light across centimeters or across a warehouse floor.
Jensen Huang framed the vision directly: “Together with Lumentum, NVIDIA is advancing the world’s most sophisticated silicon photonics to build the next generation of gigawatt-scale AI factories.”
What NVIDIA Gets From Lumentum
Lumentum is a leader in advanced laser components and optical subsystems — exactly what’s needed for silicon photonics at scale. The deal includes:
- $2 billion investment to fund R&D, manufacturing expansion, and operations
- Multi-billion dollar purchase commitment for advanced laser components
- Future capacity access rights — NVIDIA locks in supply before competitors can
- A new U.S.-based fabrication facility dedicated to scaling production
The agreement is nonexclusive, meaning Lumentum can still sell to others. But the capacity rights and purchase commitments give NVIDIA a significant first-mover advantage on supply.
Lumentum CEO Michael Hurlston confirmed the scope: “This multiyear strategic agreement reflects our shared commitment to advancing the optics technologies that will power the next generation of AI infrastructure.”
The Strategic Play: A Supply Chain Moat
This is the real story. NVIDIA isn’t just buying components — it’s vertically integrating the networking fabric of AI infrastructure.
NVIDIA made a parallel $2 billion investment in Coherent, another photonics company, on the same day. That’s $4 billion total committed to locking down the optical supply chain.
The pattern is familiar. NVIDIA built a moat around GPU compute with CUDA. Now it’s building a moat around AI networking with photonics. By securing capacity from both major optical component suppliers, NVIDIA makes it harder for competitors — including Amazon and Google with their custom silicon — to access the same critical technology at scale.
Why This Matters Beyond NVIDIA
The copper-to-photonics transition is inevitable. Every hyperscaler building million-GPU clusters will need optical interconnects. NVIDIA is positioning itself to control not just the chips, but the connections between them — which, at AI-factory scale, may matter just as much.
For the broader industry, this partnership validates that the next wave of AI infrastructure investment isn’t about making individual chips faster. It’s about making the entire system — hundreds of thousands of chips working as one — communicate efficiently enough to justify the power bill.
The era of gigawatt-scale AI factories runs on light, and NVIDIA just made sure it controls the supply.
Sources: NVIDIA Newsroom, CNBC, Network World, GlobeNewsWire