The new commodity

Last month I sat in on a call with UBS technology analysts because I wanted to ask them their view on US Big Tech valuations. I asked them if the massive, capital-intensive bets these companies were making on generative AI would ever turn profitable.

I pay for an LLM service, so there’s obviously value there, and we all know businesses are racing to use these tools (mostly to cut costs). But the capex demands in Silicon Valley are bottomless.

Randy Abrams, head of Taiwan research, said the team thinks about this. It looks at cloud-service growth as one metric; another is to look at how well companies like Amazon, Alphabet or Meta are monetizing e-commerce and other facets of the consumer internet. The more hyperscaling growth, the more easily tech companies can underwrite their AI-related investment.

Abrams also noted that the capex spend, relative to cashflow, by telecom companies on the early internet thirty years ago was far higher. More than half of hyperscalers’ operating cash flow is free cash flow (after capital expenses), he said.

The risks to Big Tech are a slowdown in scaling and development of AI applications used by enterprises, but UBS doesn’t see such a trend.

But I wouldn’t rest easy. DeepSeek exploded on the scene shortly after this call, trashing Big Tech stock valuations, particularly Nvidia’s. The industry is now debating the validity or nuances of DeepSeek’s claims to have developed its LLMs at a fraction of the cost. Or whether this is about open-source AI versus proprietary AI.

Meanwhile I’m also seeing more AI startups getting funded and more financial institutions trying different things.

I don’t doubt the impact of generative AI and, building in the background, the race to artificial general intelligence. But it seems to me that new business models as well as DeepSeek-style engineering triumphs are bringing the costs of AI down.

It’s a commodity. Maybe it only stops being a commodity when some super intelligence emerges from some massive data center, twirling its mustache, and wipes out the competition. At which point we’ll all have something else to worry about.

But the analyst’s view of why Big Tech’s exorbitant spend on AI, from data centers to chip production to nuclear power plants, seems to be looking at the wrong things.

Cloud usage up? Tick. Consumer internet monetization? Tick. ?Enterprise adoption of genAI tools? Tick tick.

But what if the costs of AI are declining? If the expansion of open-source models are obliterating any differentiation of models? If there’s no moat?

The Nasdaq has recovered as questions about DeepSeek reassured investors that Big Tech’s narrative remains intact. My 401(k) hopes they’re right!

My way to gauge the truth? How Big Tech operate in DC. They’ve become the mightiest lobby in the world under the new Trump administration at a time when protectionism has come into vogue. These companies will do anything to protect their rents. The more extreme the political extraction they claim – the louder the China bashing – the more likely the financials wouldn’t stand up in a free, competitive market.