NVIDIA Corporation (NVDA) Presents at TD Cowen's 54th Annual Technology, Media & Telecom Conference Transcript

NVIDIA said its networking revenue reached $14.9 billion, up 199% year-over-year, driven largely by its AI rack architecture and integrated computing systems. Management framed the offering as a "single unit of computing" and an "AI factory," underscoring strong demand tied to AI infrastructure buildout. The remarks were made at TD Cowen's TMT conference following last week's earnings release.

Analysis

The key takeaway is not just that networking is booming; it is that NVDA is increasingly monetizing the entire rack as the unit of sale, which raises the switching costs for customers and compresses the decision window for competitors. Once the network fabric becomes co-designed with the compute plane, the addressable share shifts from GPU-only budgets to full system budgets, making attach rates and platform lock-in more important than raw chip share. That dynamic is structurally favorable for NVDA, but it also means the next leg of upside depends on execution at scale rather than product novelty.

Second-order winners are the suppliers and ecosystem names tied to rack-level deployment, optics, interconnect, power, and cooling. The bottleneck is no longer just chip supply; it is increasingly installation velocity, data-center power availability, and integration complexity, which should extend demand for high-speed networking components and create persistent scarcity premiums. The losers are point-solution vendors that still sell AI infrastructure as discrete components rather than a bundled system, because procurement teams will optimize for time-to-train and power efficiency, not standalone performance specs.

The main risk is that the market extrapolates a multi-quarter growth curve into a multi-year straight line. If customer capex digestion, power constraints, or pricing pressure emerge in the next 3-6 months, the networking narrative can decelerate faster than the core compute story because it is more tied to system rollouts and less to broad model-training demand. A secondary risk is internalization: hyperscalers will keep pushing to own more of the stack, so NVDA’s long-run margin mix may depend on how much of the system remains proprietary versus commoditized.

The contrarian view is that this is less about a one-quarter blowout and more about NVDA redefining TAM through architecture, which the sell-side may still underappreciate. If the market is only underwriting GPU unit growth, it is missing the option value in network, software, and rack orchestration attach. But if customer concentration stays high, the upside is real yet fragile; the best risk/reward is likely in owning the platform leader while selectively shorting the weakest incremental beneficiaries of AI infra spend.

Key Decisions for Investors

• Stay long NVDA over the next 3-6 months on any post-event consolidation; the setup favors continuation if rack-level demand remains tight, but trim if the stock starts pricing in flawless 2027-2028 adoption.
• Pair long NVDA / short a basket of legacy networking vendors over 1-2 quarters; the thesis is that system-level bundling will steal wallet share from standalone hardware names with weaker AI attach.
• Add exposure to AI infrastructure enablers tied to power, cooling, and optical interconnect on 2-6 month horizons; these are the most likely second-order beneficiaries if rack deployments keep scaling.
• Use call spreads rather than outright longs in NVDA if implied vol remains elevated; risk/reward is best for upside participation while limiting downside if capex digestion appears by the next earnings cycle.
• Watch for any signs of hyperscaler component in-sourcing; if procurement shifts meaningfully inward, reduce exposure to the broader AI hardware chain and rotate back to the platform leader.