Jensen Huang showed up on CNBC this week with a message for everyone worried about AI spending: relax, everything’s fine. The Nvidia CEO defended the absolutely massive capital expenditures his biggest customers are planning, calling it sustainable and appropriate. Easy for him to say when companies are literally lining up to buy his chips.
The numbers are frankly ridiculous. Meta, Amazon, Google, and Microsoft could collectively drop $660 billion on capital expenditures this year. Most of that cash is flowing straight into AI infrastructure, which means straight into Nvidia’s pockets. Wall Street had mixed feelings about this spending spree, rewarding Meta and Alphabet while punishing Amazon and Microsoft for their aggressive investment plans.
The Biggest Infrastructure Buildout Ever
Huang described what’s happening as “the largest infrastructure buildout in human history.” That’s quite a claim, but when you look at the scale of investment, it’s hard to argue. He says demand for computing power is “sky high” and that Technology companies can use that power to make serious money.
His pitch basically boils down to this: buy more chips, make more money. He pointed to Meta shifting from CPU-based recommendation systems to generative AI systems powered by his hardware. Amazon Web Services will supposedly transform product recommendations through AI. Microsoft’s enterprise software gets smarter. Everyone wins, especially Nvidia.
The interesting part is Huang’s claim about AI labs like OpenAI and Anthropic. He said both companies are “making great money” and that if they had twice the compute power, their revenues would jump four times. That’s a bold multiplier effect to throw around. Nvidia invested $10 billion in Anthropic last year and plans to pour more money into OpenAI’s next funding round.
The Question Nobody’s Asking
Here’s what caught my attention: Huang claims even six-year-old chips like the A100 are still being rented out. Every GPU Nvidia has ever sold is apparently generating revenue for someone right now. That speaks to sustained demand, sure, but it also raises questions about whether the newest, most expensive chips are actually necessary for everything being labeled “AI.”
The business model he’s describing depends entirely on AI companies continuing to generate profits from their services. If people keep paying for AI and the AI companies keep making money, then they’ll “keep on doubling, doubling, doubling, doubling” according to Huang. That’s a lot of doubling that needs to happen for a $660 billion bet to pay off.
Wall Street’s mixed reaction tells you everything you need to know about investor confidence. Some companies got rewarded for their AI spending plans while others got hammered. The market isn’t convinced everyone placing these massive infrastructure bets will see returns justify the expense.
What happens when the doubling stops, or when it turns out that six-year-old chips work just fine for most AI applications that actually make money?
