The real leadership issue is not whether AI infrastructure demand remains strong; it is whether enterprise buyers let supplier momentum dictate architecture decisions. Nvidia’s results reinforce that compute capacity will stay strategically important, but CIOs should separate market growth from workload fit. A booming supply chain does not automatically justify standardising every AI use case on the same stack.
That makes portfolio segmentation more important than headline GPU strategy. Training, batch inference, latency-sensitive inference, internal copilots and embedded product features each create different cost curves, utilisation patterns and switching constraints. The management question is less “How much GPU capacity do we need?” than “Which workloads deserve premium infrastructure, and which should be designed for portability or lower-cost alternatives?”
A practical implication is that AI business cases now need an infrastructure sensitivity model, not just an accuracy or productivity target. For each major workload, IT leaders should test at least three variables: expected token or query growth, unit economics under different processor options, and the operational cost of being tied to a single cloud or software ecosystem. That exposes where the business case depends on today’s pricing assumptions holding up.
The other overlooked issue is negotiating power. As hyperscalers build rival silicon while still buying Nvidia at scale, enterprises may benefit from a more competitive market over time—but only if they preserve optionality in tooling, model deployment patterns and procurement commitments. The strategic advantage is not owning the latest chip. It is keeping application architecture aligned with business value while the infrastructure market reorganises beneath it.
Nvidia’s latest earnings, announced Wednesday, offer another data point that the enormous investment in AI infrastructure has further to run. But beneath the headline numbers, the results also highlighted the complicated infrastructure decisions facing technology leaders.
First, the numbers: Nvidia reported $96.2 billion in quarterly revenue, a 106% increase from a year earlier, while data center revenue reached $89 billion, up 117%. The company forecast $108 billion in revenue for its current quarter and said it expected revenue to grow by about 70% in fiscal 2028.
That last figure is particularly notable. Nvidia is forecasting extraordinary growth several years into the future, even though it is already operating at unprecedented scale; the forecast is also significantly above Wall Street’s expectations of roughly 45% growth, according to The Wall Street Journal. CFO Colette Kress said the projection was supply-constrained, with customer demand expected to be roughly double what Nvidia can currently supply.
Related:The AI infrastructure boom is coming for enterprise budgets
For CIOs, the interesting question is what to make of that outlook when the infrastructure market is changing as quickly as the workloads driving it.
Nvidia is betting the infrastructure cycle has years to run
Nvidia’s forecast provides enterprise technology leaders with a useful indication of how the industry’s largest infrastructure supplier views demand.
The company isn’t simply expecting another strong quarter, it’s building a supply chain and product roadmap around continued expansion well into fiscal 2028. Its supply commitments have grown to $279 billion, according to the Financial Times, while AWS announced that it would deploy another 2 million Nvidia GPUs across its infrastructure in 2027 and 2028.
The significance for CIOs is how these numbers provide a window into the assumptions being made by the companies with the greatest visibility into the infrastructure market. And those assumptions matter because enterprise infrastructure decisions have to account for much longer time horizons than an earnings quarter.
A CIO deciding how to provision computing capacity, structure cloud commitments or build out a data center is now a CIO making decisions against a backdrop in which the leading supplier expects demand to remain exceptionally high — even while the technology and economics underneath that demand continue to shift.
Compute economics becoming part of the application equation
One of the more revealing comments from Nvidia CEO Jensen Huang was that AI had reached an "inflection point" now that its tokens were becoming productive and profitable, adding: "Now, compute is revenue."
Related:Rethinking the IT portfolio and budget in the AI era
Everyone knows AI can use enormous amounts of computing power. But as AI workloads move into production, the cost of compute increasingly becomes part of the economics of the application itself. An organization running an AI service at scale, therefore, has to consider how infrastructure costs scale with usage, rather than treating compute as a relatively fixed technology expense.
That helps explain why Nvidia’s expansion beyond GPUs matters. The company expects CPU revenue to more than double in fiscal 2028 and is increasingly positioning its platform around inference as well as training.
The hardware choice can therefore become a workload-level economic decision. A processor that makes sense for training a large model may not be the most efficient option for serving that model at scale. A custom accelerator may make economic sense for a hyperscaler running a workload millions of times, while an enterprise with a more varied portfolio may value general-purpose infrastructure and portability more highly.
For CIOs, infrastructure planning increasingly has to connect architecture decisions with application economics. It’s about determining not just which hardware offers the most performance, but also which architecture produces an acceptable cost profile for the workload the organization expects to run.
Related:Latest AI research puts CIOs and CTOs in the driver’s seat
Nvidia’s customers are also its competitors
The other unusual feature of the market is that Nvidia’s biggest customers are increasingly developing alternatives to Nvidia’s technology, even as they sign new deals.
AWS is committing to another 2 million Nvidia GPUs while continuing to develop its own accelerators. Google, Microsoft and Meta have similarly invested in custom silicon, while OpenAI and Anthropic are pursuing their own chip strategies.
Of course, it’s not a simple swap: Those companies still have powerful incentives to buy Nvidia hardware, as developing an alternative does not immediately provide the scale, software ecosystem or availability that Nvidia offers today. But owning more of the underlying technology can give hyperscalers greater control over cost, performance and supply over time.
Huang addressed the competitive question on the earnings call by emphasizing Nvidia’s broader platform, arguing that its advantage extends across the AI computing lifecycle and across different cloud environments. He also said he expected companies such as OpenAI and Anthropic to remain Nvidia customers even as they develop their own processors.
That creates a more complicated infrastructure landscape for CIOs, but potentially one with more opportunities. As hyperscalers expand their competitive offerings, CIOs will have more choices available to them.
The decision is increasingly about ecosystems rather than individual chips. An organization choosing a computing platform is also making decisions about software compatibility, cloud relationships, available tooling, workload portability and the alternatives that will remain viable if the market changes.
The fact that Nvidia’s largest customers may simultaneously be their competitors is a useful reminder that today’s market positions are not permanent. Enterprises don’t need to replicate the infrastructure strategies of hyperscalers, but they do have reason to distinguish between workloads where Nvidia’s integrated platform provides a meaningful advantage and those where maintaining alternatives is strategically valuable.
The infrastructure decision is becoming a business decision
Nvidia’s results provide plenty of evidence that the infrastructure buildout is continuing at extraordinary scale. They also show why simply extrapolating that growth into larger enterprise infrastructure budgets misses the more difficult question.
CIOs are making technology decisions in a market where demand is expanding, but so are the number of architectures capable of meeting that demand. Nvidia is broadening its own platform while hyperscalers are developing alternatives. Inference is changing the economics of workloads. And the costs of the infrastructure itself are shifting as demand for components such as memory increases.
That makes flexibility increasingly valuable, but flexibility doesn’t necessarily mean avoiding commitment. It means understanding where a technology decision creates a durable business advantage and where it simply reflects the current state of the infrastructure market.
Nvidia’s earnings suggest that CIOs will have plenty of computing capacity to plan for over the next several years. The harder part will be deciding which infrastructure bets are worth making when the economics of the workloads — and the companies supplying the technology — continue to evolve.
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.