Nvidia's Next $1 Trillion Opportunity May Have Nothing to Do With GPUs

Nvidia is highlighted as expanding beyond GPUs into AI networking, with its networking stack spanning InfiniBand, Spectrum-X Ethernet switches, and BlueField DPUs. The article cites $2 billion investments each in Coherent and Lumentum, plus another $2 billion in Marvell Technology, to strengthen optical interconnects, silicon photonics, and custom AI accelerator capabilities. It argues this broader AI infrastructure strategy could support additional earnings power, while noting Nvidia trades at about 22x forward P/E.

Analysis

The market is likely still pricing NVDA as a GPU monopoly with an accessory networking story, but the more important shift is that the attach rate on networking is becoming a gating item for AI spend. That changes the earnings mix: compute is still the headline, yet the incremental dollars are moving toward higher-margin system integration, switching, optics, and control-plane products where NVDA can monetize the full rack, not just the chip. The second-order winner is whoever can reduce deployment friction for hyperscalers under tight power and rack-density constraints, because the constraint is shifting from silicon availability to usable throughput per watt and per dollar of capex.

The clearest competitive risk is to standalone networking suppliers and custom silicon vendors whose value proposition weakens if hyperscalers prefer an integrated vendor stack. That said, the article understates the possibility that NVDA’s networking push is as much about supply-chain assurance as product breadth: by investing into optics and custom accelerator ecosystems, it is trying to de-risk bottlenecks that could otherwise cap GPU shipments. If those bottlenecks ease, the net effect is not just more networking revenue; it is a higher ceiling on GPU utilization and a faster digestion of hyperscaler capex, which can extend the AI build cycle by multiple quarters.

Contrarian view: the market may be too focused on the durability of AI capex rather than the mix. If hyperscalers prioritize networking and power infrastructure, some of the near-term upside may accrue to the picks-and-shovels layer faster than to the most obvious compute leaders, especially if GPU pricing normalizes or competitive pressure rises. The biggest reversal risk is a pause in infrastructure orders driven by internal efficiency gains or a digestion phase after the current buildout wave; that would hit the ecosystem in 3-6 months, not immediately, because backlog and deployment schedules are still working through the system.

Key Decisions for Investors

• Maintain long NVDA, but size for a multi-quarter view; use any post-earnings weakness as an entry point because the next leg is likely driven by networking attach and system-level gross margin expansion, not just GPU unit growth.
• Pair long NVDA / short a basket of pure-play networking legacy names on a 3-6 month horizon; thesis is that integrated AI-stack vendors capture the value migration while undifferentiated transport vendors face pricing pressure.
• Accumulate COHR and LITE on pullbacks as a leveraged expression of optical interconnect capex; best risk/reward if AI buildout remains supply-constrained, with upside over the next 6-12 months if hyperscaler spend broadens beyond GPUs.
• Use MRVL as a tactical long only if you want custom silicon exposure, but keep stops tight: it benefits from the AI platform expansion, yet the stock is more vulnerable to a slowdown in custom accelerator design wins over the next 1-2 quarters.
• For a lower-risk expression, consider an NVDA call spread 6-12 months out to capture continued AI infrastructure re-rating while capping downside if the market starts to discount a capex digestion phase.