@Manlobbi, We agree that huge capex can be good if the returns on capital being employed are good. How can you or anyone else for that matter tell what the returns on at least $500 Billion (likely significantly more) of AI data center capex will be with any degree of confidence, taking into depreciation schedules.

Wr don’t need blind faith in Warren to work out the CapEx as the returns are already showing up in the numbers.

First, “demand ahead of supply" is not just earnings call rhetoric but backed by an insane ($560B+) Cloud backlog, with over half of that converting to revenue in the next two years.

They aren't building on spec; they are fulfilling signed contracts.

Their cloud revenue doubled to $30B+ and Cloud operating income tripled to $11B, now margins are 36%. Adding $30B in annualized operating profit from Cloud alone shows strong incremental returns even with the heavy depreciation hit.

* (Briefly flipping to subject of last earnings call that wasn’t liked: There is a lot of fuss about their search ads having slowing rate of growth (low teens instead of high teens) but this is deliberate and temporary - and ad growth will resume higher over the medium-term. It was partly currently related but for the larger part, they have deliberately removed ads in the AI responses (which took volume from other search queries) and they reported that they're adapting ads into those AI responses carefully and it is going well, and they will scale up. So they sensibly reduced ads in the short-term (which Wall Street is too impatient about) to make sure aggregate volume of queries would hold up with normal search plus non AI search, prior to ramping up ads on the AI said which they are just starting - has to be done carefully/tasteful to retain search value so they do it smaller scale in various ways and then ramp up. I expect ten years away search revenue will combine with afentic fees to reach products and services, and the combo will be astronomically higher than today, however Google will have a lower total market share. But the combo will grow so much that Google will keep growing what I’ll call its “product finding ads and fee” business, but the market will go through panic and euphoria towards Google about how this combo changes over time.)

Back to capex.. From my own observations, the productivity gains from AI are very real, but we are still in the early stages and the tools are completely underutilized. There is a massive runway ahead for increasing token usage as adoption actually matures. Token costs will keep falling but token use will scale up at a far faster rate than the falling token costs. In short there is a huge future market for companies patting extraordinary fees for the best AI response in competition to each other (similarly as firms presently compete for the best staff) and the expenditure is fairly small now. The “bubble will pop” likely when AI corporate expenditure is a fairly higher percent of total expenditure (when it starts to catch up to salary costs) and it is nowhere near that now for most large firms.

Finally, the IBM comparison misses the mark. IBM in 2011 was a stagnating legacy business getting disrupted by the cloud. Alphabet is growing revenue 25% on trend. That is not from arbitrary hopeful buildouts as with Cisco in the early 2000s but contracted demand converting at 36% margins.

Starting from quotes, Google will make an excellent long-term investment, but the IV10/price ratio is even a little higher today for Meta so if starting today I'd purchase Meta over Google. Meta have more to gain (new AI-oriented apps launched within their impenetrable network moat) and less to lose (unlike Google at less risk of long-term weakening prompt distribution), and shockingly valuable capacity to invest incremental capital in their own moat without reducing it, and priced cheaper than Google accounting for normalized long term margins. Google’s main moat is distribution (Android, browser default deals, etc) plus habit, search quality, etc. Meta’s moat is pure network effect and harder to erode or replace, especially as any competing ecosystem they can hack up a copy and distribute in their own network moat (thus outcompeting X with Threads going from zero market share to over 50% of the twitter-like-market in just 3 - yes 3 - years). That was after they killed the Vine dominant related product by hacking up a clone called Instagram. Reels neutralizing TikTok, and so on. And they can keep doing that without anyone able to stop them, antimonopoly regulation able to stop acquisitions but not clone hacks that have an even higher ROIIC and moat expansion. And Meta will be reprised in 2029 when they are registering signifiant revenue beyond the current ad sources - including new agentic fees to consumers, plus ads becoming exceedingly more targeted than they are now. Consumers will shift towards laziness and ask agentic prompts: "I’m after something that does <x and y> and is trusted. Just buy it for me now at lowest cost, usual card, thanks", which for Meta will combine revenue share fees, targeted ad fees, on top of the consumer high token use fees. If not revenue sharing then it can be placement fees (or combo of both) but in any case the targeting specificity is insanely higher than browsing a Facebook page and advertising revenue scales with volume times specificity, both certain to increase more than I expect the market understands.

- Manlobbi