PwC estimates that information and communications technology equipment, including chips, currently accounts for about 70% of AI infrastructure spending. That proportion could reach 93% by 2050 as operators replace equipment every few years. Annual spending on AI data-center infrastructure is projected to rise from roughly $800 billion in 2026 to $1.8 trillion in 2050.
The forecast treats infrastructure investment as a continuing process, with newer chips needed both for more complex AI models and for inference applications with greater computing requirements. More powerful equipment can change the economics of running workloads, encouraging operators to upgrade rather than rely indefinitely on existing hardware.
Nvidia offered a separate near-term spending outlook during its August earnings call. It said the five largest U.S. hyperscalers were on course for $800 billion in capital expenditure this year and projected $1.3 trillion in 2027. Those figures exclude spending by neocloud providers and AI laboratories such as OpenAI and Anthropic, so they do not represent the entire market.
The company says its latest Vera Rubin systems will use twice the power of the preceding Blackwell chips but deliver a tenfold improvement in performance per watt, according to CNBC. These are Nvidia's performance claims. The commercial argument is that higher computing output relative to power use could lower the cost of workloads despite greater total system power consumption.
Nvidia's $164 billion in first-half data-center revenue corresponds to an annualized pace of just under $330 billion. That calculation describes the pace implied by the first six months, rather than a reported full-year result. The much larger industry spending forecast includes opportunities across AI infrastructure and is not a forecast of revenue belonging solely to Nvidia.
Semiconductor industry tracker Silicon Analysts estimates that Nvidia holds about 80% of the AI accelerator market. The company is seeking to maintain its position while expanding into server processors, where it anticipates a substantial increase in revenue.
Nvidia is also working with MediaTek and Marvell Technology to connect custom AI processors to its rack-scale server systems. Those partnerships form part of its effort to broaden the ecosystem around its infrastructure rather than rely only on sales of standalone accelerators.
The analysis also identifies physical AI as an emerging market Nvidia is pursuing. Long-term earnings estimates shown by YCharts point to annual growth above 50%, while Yahoo Finance data put the company's price/earnings-to-growth ratio at 0.45 using projected earnings growth over the next five years.
Those valuation measures depend on growth estimates being realized. PwC's projections likewise describe a possible long-term investment path, while Nvidia's eventual share will depend on its products and market position as infrastructure and chip replacement cycles develop.