NVIDIA doubles down on “AI factories” buildout, but buyers’ ROI questions are starting to surface

NVIDIA CEO Jensen Huang used the company’s latest earnings communication to make a broad bet that customers will keep scaling what he called “AI factories,” data centers built specifically to run large-scale AI workloads. In prepared remarks tied to first-quarter results, Huang said the buildout is accelerating at “extraordinary speed,” and argued that agentic AI is starting to do productive work across industries, not just generate demos.

That confidence is backed by results. NVIDIA reported record first-quarter revenue of $81.6 billion, with Data Center revenue at $75.2 billion, up 92% year over year. The company also said the number of partner data centers exceeding 10 megawatts has nearly doubled in a year, now surpassing 80 sites, a sign that customers are not merely purchasing chips, but standing up full production environments.

In the earnings call transcript, NVIDIA tied the acceleration to two drivers. First, hyperscale workloads are continuing to transition from CPU-based computing to GPU-accelerated computing, spanning applications from search and advertising to recommender systems and content understanding. Second, NVIDIA said the adoption of AI-native products and services is inflecting, with mainstream AI moving from one-shot inference toward reasoning and then agentic workflows. NVIDIA also projected that hyperscale capital spending could exceed $1 trillion in 2027, and that overall AI infrastructure spending could reach $3 trillion to $4 trillion annually by the end of the decade.

NVIDIA also acknowledged why the ROI conversation is now moving closer to the center of the market. The company said “today’s data centers are revenue-generating AI factories” but that they are constrained by power and capital. Rather than framing the economics around GPU purchase price, NVIDIA told investors that operators should look at lifetime factory cost to produce “intelligence,” including metrics like token per watt, tokens per dollar, uptime, utilization, time to production, software durability, and asset life.

In that context, a recent market report highlighted what it described as a “tiny problem” behind the upbeat messaging: enterprise buyers are reportedly starting to ask harder questions about what they are getting back for their spend. That report pointed to indicates of tighter discipline at several large technology users, describing challenges linking token spending to measurable product gains and changes to internal AI usage or licensing plans. The report did not, however, provide specific NVIDIA customer-by-customer return thresholds.

NVIDIA’s own messaging leaned in the opposite direction. In the call remarks, the company said the value of its AI infrastructure is rising, pointing to increases in cloud pricing for certain accelerators, and it argued that customers are generating profitable revenue beyond the depreciable life of their GPUs. NVIDIA also noted it has been preparing for capacity demand on the supply side, increasing total supply-related commitments (including inventory, purchase commitments, and prepaids) to $145 billion in the quarter, underscoring that it sees ongoing factory buildout as the base case.

What NVIDIA did not spell out is the exact shape of any ROI slowdowns, if they occur. The company did not disclose how many enterprise pilots are being delayed, what utilization or payback targets are failing, or whether specific customer procurement plans have changed due to “token spend” scrutiny. For now, the gap is informational: NVIDIA is arguing for factory economics and pricing strength, while market commentary suggests buyers may still be waiting for clearer, company-specific financial attribution.

Heading into the next earnings cycles, investors will likely look for confirmation that the buildout remains financed at the pace Huang describes. Key signposts include continued commentary on hyperscaler capital spending, data center utilization and pricing, and whether NVIDIA’s “token-per-watt” value proposition translates into more consistent, measurable outcomes for enterprise operators who are balancing aggressive infrastructure expansion with tighter budgeting discipline.