Screaming about bubbles

Don't be fooled, don't be a turkey this thanksgiving

The screaming is obnoxiously loud . In all kinds of media, the AI naysayers keep whining about investing circularity and impossible revenue goals for AI companies. The circular argument goes as follows:

1) Nvidia and AMD have invested huge amounts in Open AI only so it can buy GPU computing capacity.

2) Oracle takes on billions in debt ( god forbid!) to build data centers for Open AI

3) Microsoft has a contract to supply OpenAI with $250 billion computer capacity over the next five years.

4) Thus, if Open AI can’t meet its revenue goals and has no clear path to profitability , capital investment will plummet .

5) In turn all the “picks and shovels” companies such as Nvidia, Broadcom , Arista, Micron, and others will tank.

It’s not a bad argument. But it’s fallacious. The biggest investors in data centers are Google, MSFT and Amazon who cannot meet demand from customers seeking computing capacity for AI projects. MSFT’s AI demand grew at 40% last quarter and Google and Amazon also saw large growth rates.

This revenue comes with healthy margins, so the massive capital investment by the hyperscalers is already yielding fruit (cash). The customers are not just Open AI and Anthropic, but thousands of startups fueled by venture capital, and more importantly, large enterprises who clearly see the value in AI. Sure, some start-ups may fail, just like so many dotcoms when the bubble burst in 2000. But those that succeed in this new AI fueled frenzy could become the next Google or Facebook; both rose out of the dotcom ashes to dominate search and social media, respectively.

There is no AI bubble yet. The operative word is yet. For CY 2026, chip and equipment orders are already in at Nvidia, Broadcom, Arista Networks, and Lam Research. TSMC, the semiconductor foundry of choice is sold out. And if there is a bubble, it will deflate slowly not rapidly.

Any first slowing signals will come from hyperscalers. Only if all of MSFT, GOOG and AMZN announce drastically reduced capital expenditures for 2027 will the market tank. For 2026, these three have already announced vastly more Datacenter Capex spending . Thus, any indications of 2027 spending will not appear until after Q2 2026 results come in.

Recall, that AI compute demand rests on demand for tokens. Any AI query generates tokens which are computed to provide an answer. More complex queries means more tokens generated. Simple — GPU use ( compute) is related to token volume .

Token use currently is limited to text prompts. When multimodal inputs and outputs (audio, video, and text) are added to the mix, token demand will explode. Software agents that automagically finish work even with complex workflows are another monstrous demand driver. Predictions are that by 2030’s token volume will be 100 times that of 2025. That, mes amis, requires a lot of computing capacity (GPUs) . I’m also confident of AI’s transformational capabilities and so this is not a bubble, yet.

In short, Nvidia, Broadcom, Arista , Micron and other “pick and shovel “companies are safe through Q2 2026. It’s not a time to sell now. I will reconsider my view in mid-2026 for signs of any cracks, but not before.

Gobble Gobble to you and for any men out there drink some Wild Turkey this Thanksgiving. It grows hair on chests. Invest soundly mes amis.