The AI trade could shift back in Nvidia’s favor. Here’s why
Infrastructure has been a huge part of the artificial intelligence rollout, with companies building data centers, networking hubs and power plants to increase capacity for power-hungry algorithms. But physical infrastructure is likely to take up a smaller share of capital expenditures in coming years relative to the core component of all that hardware – chips. The reason is that data centers can burn through chips in just a few years, while plants and bigger pieces of equipment can last decades before they wear out. That’s good news for Nvidia as the premier maker of graphics processing units, or GPUs. “While the curve for data center financing will likely stabilize around 2028, chip financings will likely continue to grow into 2030, particularly as replacement demand becomes a theme,” JPMorgan strategist Tarek Hamid wrote in a Tuesday note to clients. Hamid thinks that spending on GPUs and AI-specific chips could grow to 60% of total annual spending by 2030 from roughly 50%. That’s relative to what he calls “data center box capex” — the more general infrastructure component. The bank also expects more than $3 trillion in financing for AI chips and essential hardware components over the next five years, with silicon spending increasing to about $800 billion in four year’s time from $340 billion in 2026. NVDA YTD mountain NVDA year to date Nvidia is already reaping the benefits of the AI boom. For the fiscal first quarter alone , it reported $81.6 billion in revenue, up 85% from the year-earlier period. The company sees itself as “uniquely positioned at the center” of the AI transition, according to CEO Jensen Huang , and the shift toward GPU in the capex mix spend will only center it further. Nvidia chief financial officer Colette Kress said in May that AI spending is on track to reach $3 trillion to $4 trillion annually by the end of the decade. The forecast has been mentioned many times in notes from Wall Street analysts. Still, the stock lags other AI chipmakers for the year — in part due to growing demand for CPUs, or central processing units. Nvidia is up more than 12% in 2026, while CPU maker AMD has more than doubled. Chip lifespan, capex and required ROI The shifting capex mix is all about lifespan and depreciation. Some data center components can last up to 30 years, according to UK computing services firm Infiniti , while the lifespan of a GPU can be a tenth of that time. This shorter lifespan coupled with strong AI demand could lead to more purchases of Nvidia’s graphics chips. Overall AI capex forecasts continue to grow as well. JPMorgan sees total spending of $5.5 trillion through 2030, up from a $5.1 trillion prediction made in November. But while hyperscalers like Amazon and Microsoft are reporting substantially increased revenues from their clouds, sector-wide returns on investment, which hinges on increased productivity for the companies buying AI services from the hyperscalers, is far from proven. The U.S. saw just 0.3% labor productivity growth in the first quarter, according to the Labor Department . “Unless we get some blockbuster growth numbers for output in June, we will have another quarter of weak productivity growth,” economist Dean Baker at the Center for Economic Policy at Research wrote earlier this month. “Either the story of job-killing AI is yet another economic myth, or the AI is much smarter than we think, and is hiding from the statistical agencies.” Economists have modest expectations for long-term productivity gains coming from AI. Daron Acemoglu at the Massachusetts Institute of Technology put the number at “no more than a 0.71% increase” in total factor productivity over the next decade. It’s not clear whether those gains will translate to the boosted profits needed to justify the increased spending either. This is a concern that could grow more worrisome if AI enterprise utilization doesn’t justify the spending, which some industry insiders are taking note of. “I think it’s going to be about half the capacity that they’re going to be actually building [compared to] what was planned. That seems to be the way it’s going,” David Linthicum, former chief cloud strategy officer at consultancy Deloitte told CNBC on Wednesday, adding that he’s looking at construction starts data as a key indicator. Nvidia GPU shipments Near term, however, the stars are aligning for Nvidia to reign supreme in the AI trade. As the AI buildout continues and more money pours into the sector for boosted investment, Nvidia has its fingers in more pies than just about any other company. JPMorgan expects Nvidia to ship 8.9 million GPUs this year, versus 4.5 million comparable TPUs from Google and just 1.9 million Inferenta and Trainium chips from Amazon. They see Google TPU shipments making quick headway into Nvidia’s territory, with 8 million TPUs to be shipped next year, versus 9.9 million GPUs from Nvidia. The company also has a wide array of hardware deals. Partnerships singled out by JP Morgan in a June 17 note were with OpenAI, Anthropic, Amazon and Microsoft, among others.
