NVIDIA Announces Spectrum-X Photonics, Co-Packaged Optics Networking Switches to Scale AI Factories to Millions of GPUs
Second vendor CPO power claim on comparable framing: 3.5x power efficiency, 4x fewer lasers, 63x signal integrity, 10x resiliency, 1.3x faster deployment vs pluggables; 100 Tb/s and 400 Tb/s Ethernet configs, 144x800G InfiniBand. ROADMAP, not deployment — dated 2025-03-18, Ethernet stated as coming in 2026. Lumentum named in NVIDIA's own CPO partner list (bears on LITE).
NVIDIA Spectrum-X Photonics and Quantum-X Photonics
What this document is — read this first
A GTC 2025 announcement, dated 18 March 2025. It names Quantum-X Photonics (InfiniBand) as available "later in 2025" and Spectrum-X Photonics (Ethernet) as coming in 2026. Every multiplier below is therefore a claimed specification on an unshipped product, not a measurement of anything deployed. That distinction is the whole reason this source ranked second in its pair, and it must be carried into anything written from it.
The claims
Versus pluggable optics:
| Claim | Multiplier |
|---|---|
| Power efficiency | 3.5× |
| Lasers required | 4× fewer |
| Signal integrity | 63× greater |
| Network resiliency at scale | 10× better |
| Deployment speed | 1.3× faster |
Port configurations and aggregate bandwidth:
- Spectrum-X Ethernet: 128 × 800 Gb/s or 512 × 200 Gb/s → 100 Tb/s; and 512 × 800 Gb/s or 2,048 × 200 Gb/s → 400 Tb/s.
- Quantum-X InfiniBand: 144 × 800 Gb/s (on 200 Gb/s SerDes), claimed 2× faster and 5× higher scalability than the prior generation.
Availability as stated: Quantum-X Photonics InfiniBand later in 2025; Spectrum-X Photonics Ethernet in 2026.
Partner ecosystem (as listed)
TSMC, Browave, Coherent, Corning, Fabrinet, Foxconn, Lumentum, SENKO, SPIL, Sumitomo Electric, TFC Communication — plus transceiver suppliers Coherent, Eoptolink, Fabrinet and Innolight.
Lumentum appearing in NVIDIA's own CPO supply list is the position-relevant fact in this document, more so than any multiplier: it is the vendor naming the supplier, which is a different order of evidence from a sell-side note asserting the relationship.
Quotes
- Jensen Huang, CEO, NVIDIA — "AI factories are a new class of data centers… NVIDIA is shattering old limitations of hyperscale networks."
- C.C. Wei, Chairman & CEO, TSMC — TSMC's solution "helps NVIDIA unlock ability to scale to a million GPUs and beyond."
Limitations
- Roadmap, not deployment. Announced March 2025 for delivery across 2025–2026; this file records what was claimed, and no part of it is evidence that the products shipped, shipped on time, or met these numbers in the field.
- Vendor marketing multipliers with no stated baseline part. "63× greater signal integrity" in particular is not a standard metric with a defined unit — treat it as a marketing composite, not a measurement.
- No absolute power figure (watts per port, pJ/bit) anywhere in the release, so the 3.5× cannot be independently recomputed.
- The comparison baseline ("pluggable optics") is unnamed, as it is in the Broadcom release.
Bears on
LITE (Lumentum) and the photonic-interconnect thesis, same as N2.
Read as a pair with N2 (Broadcom TH6-Davisson), this is what the rung actually gains. Two independent vendors, on comparable framing, both land on roughly 3.5× for optical interconnect power versus pluggables. Neither number is trustworthy alone; the convergence is. The usable statement is: the CPO power claim across the two largest switch vendors is ~3.5× against pluggables, both vendor-stated, one sampling and one roadmap as of the source dates — which is a far weaker and far more honest claim than either release makes.
Source: NVIDIA Announces Spectrum-X Photonics, Co-Packaged Optics Networking Switches to Scale AI Factories to Millions of GPUs — NVIDIA, 18 March 2025 (GTC 2025).