Nvidia CEO says AI infrastructure spending remains sustainable

Nvidia CEO Jensen Huang said that the current level of AI-related capital expenditure is appropriate and sustainable, with plans for ongoing infrastructure build-out over the next seven to eight years.

Huang emphasized that demand for AI remains high and that the technology has become both useful and widely adopted.

Despite concerns about excessive spending by major tech firms like Amazon, Google, Meta, and Microsoft, Huang argued that the industry is not experiencing idle capacity, citing profitable revenue streams from companies such as OpenAI and Anthropic.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Behind the CEO’s confidence lies massive spending and loss-making customers

Huang calls today’s AI capital expenditure (CapEx) sustainable. Amazon, Google, Meta, and Microsoft still put nearly $200 billion into CapEx in 2024, within total data center infrastructure spending of $290 billion 1.
Goldman Sachs Research uses a Wall Street consensus estimate that puts 2026 capital spending at $527 billion for large public “hyperscaler AI companies” 2.
Huang uses OpenAI and Anthropic as proof of demand. Both reportedly burn cash. OpenAI logged a $5 billion loss in 2024 on about $3.4 billion in revenue. Anthropic lost $5.3 billion in 2024 on $918 million in revenue 3.
Strain also hits some AI application-layer companies, which sit on top of foundational AI models and sell end-user products. Perplexity reportedly spent 164% of its 2024 revenue on computing costs from Amazon Web Services, Anthropic, and OpenAI 3.

The AI build-out is reshaping the physical world and its bottlenecks

Hardware spend goes well beyond chips. It lifts demand for data center gear from firms like Schneider Electric, plus backup generators from Caterpillar 1.
Electricity supply has become a choke point. Deloitte estimates US power demand from AI data centers could rise more than thirtyfold by 2035 to 123 gigawatts, up from 4 gigawatts in 2024 4.
Survey results from data center and power company executives put grid limits near the top of the list. About 72% call power and grid capacity very or extremely challenging 4.
Long-run value may sit less in AI models, and more in physical control of power, land, plus grid connections 5.