Outskirts of Shrewd'm / Help With This Silly Computer
No. of Recommendations: 10
Is anyone else concerned about how much of Big Tech's recent "profit growth" is coming from AI investment revaluations rather than core operations?
For example, when companies like Alphabet own stakes in AI startups such as Anthropic, new funding rounds at higher valuations can generate billions in accounting gains that flow through earnings even though no products were sold and no cash was received.
At the same time, many AI companies appear deeply interconnected through investments, cloud spending agreements, and strategic partnerships. How much of the reported growth across the ecosystem is truly organic demand versus capital circulating within the same group of players?
For BRK.B investors who focus on durable earnings and cash flows: should these mark-to-market gains be largely ignored when evaluating Big Tech profitability? And if AI valuations eventually normalize, how much reported earnings could disappear?
Curious how others are adjusting their analysis of Alphabet, Microsoft, and the broader AI trade.
No. of Recommendations: 10
> Is anyone else concerned about how much of Big Tech's recent "profit growth" is coming from AI investment revaluations rather than core operations?
> For example, when companies like Alphabet own stakes in AI startups such as Anthropic, new funding rounds at higher valuations can generate billions in accounting gains that flow through earnings even though no products were sold and no cash was received.
I will go a step further. This reevaluation is creating non-trivial money out of thin air, just like the reevaluation of in residential housing loans did 20 years ago. Mistakes in that reevaluation and leverage crashed the global economy, and it will again, and soon. SpaceX, Anthropic, and OpenAI aren't rushing to go public because because they are fools.
When you look at things like O'Leary's proposed UT datacenter, you can see speculative, low-information leveraged investors cashing in on this current valuation bubble. This is a symptom of a late stage bubble, which we've all been through twice in the last 30 years. What it has particularly in common with the dot-com bubble is the circular financing arrangements within tech. Space-X's IPO is only viable because of the rent from Anthropic, which is scrambling for compute because of an industry wide mandate to leverage LLMs, and funded out of corporate IT budgets that will demand ROI or funded out of VC backed (hundreds of billions of dollars) in firms building AI platforms or "ai-native" firms build on those platforms. Just like the internet, this has huge promise. Just like the internet, the investment in that promise has overshot in the short term.
Where it gets complicated is the global standing of the US, and prior routes of resolving this bubble when it collapses. The last two times the US leveraged its standing as the global reserve currency and good relations with allies to monetize the bad debt created in the bubbles. That might not be available this time, for broadly understood reasons that require no elaboration.
Where it comes back to BRK, for me at least, is 1) the cash balance and 2) areas of overlap in the portfolio and operating businesses. I think the cash balance is likely to be used well. I'm sure if I understand all this, as a random person on the internet, then Berkshire probably has a very strong view into this problem and is planning accordingly. I still worry about it, but it has not stopped me from accumulating under $480. The operating companies seem well positioned. The rubber hits the roads at the datacenters and the industry can turn off the capex / stop the buildout (impacting nvidia, prim/iesc/abb et al in the electric infra, micron/seagate/arista hardware vendors in the hyperscaler stacks, and so on). But the built out datacenters need power. If the industry hits the point of turning off the power and idling datacenters, it will be very bad and the cash pile will be valuable. If they leave them on, BHE will be valuable. And BHE's renewable generation capacity is something I expect to be valuable if the USD drops or at least drops relative to global hydrocarbon markets (if it becomes part of a basket of currencies ng and crude are priced in, for example).
To your specific question: Alphabet, 25 years ago, was just a superior search engine on the basis of Pagerank (insight into links based on linear algebra properties of link graphs expressed as matrices). For 10-15 years they have *always* been the dominant AI company.
Microsoft is screwing the pooch. They are dogfooding Copilot in the infrastructure products, and poorly. Azure is having reliability issues, and Copilot is poorly dependent upon Azure. They have created a dependency mess in their global software and hardware infrastructure. Their brand, at an enterprise level, is reliability and stability, and is moving their cheese to the *wrong place* in their ai strategy. I think they will course correct, but I see them executing a Boeing vis a vis the 737 Max currently.
I like Anthropic more than OpenAI. Deepseek, globally, is extremely strong but I cannot use it. Mistral is up an coming. Google, to my mind, is taking a wait-and-see position.
Apple is another company that has much greater capabilities than they tend to advertise, and I think they are also taking a smart and cautious stance.
I think BRK has chosen well long term if they're going to invest in this space by investing in good candidates to be long-term winners and having operating companies that will benefit from the deeper secular trends.
No. of Recommendations: 15
Curious how others are adjusting their analysis of Alphabet, Microsoft, and the broader AI trade.
Ten foot pole?
Even assuming things go largely to plan, there are some troubling back of the envelope issues about the future profits. One cause of furrowed brows for me is that the useful lifetime of the equipment being purchased will be *very* short given how fast the system designs are evolving, meaning any realistic view of owner earnings will have to treat them with a very short amortization. Realistic amortization charges might exceed revenues for some players. Who wants a data centre built around systems that are a generation or two out of date, or aren't optimized for the right mix of functions?
The firms in question may not use realistic schedules when they file their statements, but that just delays the reckoning, it doesn't make it go away.
Jim
No. of Recommendations: 7
Where it comes back to BRK...
As a side note on Berkshire and stocks reacting to current exuberance, anybody look at a chart for Snowflake this week? If I'm not mistaken, the price is now in the vicinity of twice what Berkshire got when the position was closed. Oh well, there's always another streetcar coming along.
Jim
No. of Recommendations: 9
Jim,
A big difference with Alphabet, Amazon, and Apple as compared with others in the field is the degree of investment in their own datacenter and silicon design competency, relationships with the fabs, and proprietary training and inference chips. I think they are most directly aware of and prepared for demand destruction (and redirection of those resources internally) and the operational characteristics and depreciation.
Nvidia to me, is the big 10 foot pole, even if they are trying hard to diversify. Micron seems like a disaster waiting to happen due to Chinese state-supported competition and commodity like characteristics of the DRAM / HBM market.
Google, at least, is still just an advertising company with elite R&D capabilities and the best of an overextended lot. If I had to buy and then not touch the position for 5 years, I'd feel much safer buying GOOGL than NVDA or MU.
No. of Recommendations: 0
yes.. SNO I got in shortly after BRK... held for a while and gave up for pastures green...oh well...
No. of Recommendations: 4
Microsoft is screwing the pooch. They are dogfooding Copilot in the infrastructure products, and poorly. Azure is having reliability issues, and Copilot is poorly dependent upon Azure.
Github (MS-owned) apparently just adjusted their Copilot subscription pricing from flat fee to usage-based, and there is a good deal of panic among Copilot addicts who are seeing their projected fees go up by literally orders of magnitude (I saw one screenshot showing a change from $29 to $30K+/month - although they could save $40 or so by moving to a higher tier subscription!). I don't think they'll be alone in this for long, and a shakeout may come quickly if the companies into "AI"-centrism and "tokenmaxing" contests find they're paying vastly more for their "AI"s than for the people prompting them. As we're used to hearing, if something can't go on, it won't.
No. of Recommendations: 1
who are seeing their projected fees go up by literally orders of magnitude
Recently I read an article about how large companies have seen their entire 2026 AI budget be consumed in just the first few months of the year. I can't seem to find the article again though.
No. of Recommendations: 3
Hey Mark, Just use AI, takes 2 seconds :)
-----------------------
You are likely thinking of the recent reports highlighting how companies are completely blowing through their annual artificial intelligence allocations due to skyrocketing usage-based "token" fees.
The specific example making waves across financial and tech circles involves Uber, whose CTO announced that the company exhausted its entire 2026 AI budget in just four months—primarily driven by engineers heavily adopting developer tools like Claude Code and Cursor.
A quick look through your recent browser history shows you actually visited a resource tracking these exact corporate shifts:
Ramp AI Index (Visited May 27, 2026)
Key Context from the Reports
The Uber Deficit: Uber's rapid adoption of AI coding assistants resulted in per-engineer monthly API consumption hitting $500 to $2,000, burning through their yearly budget by April.
The "$500 Million" Blunder: A massive story broke detailing an AI startup that accidentally racked up a $500 million monthly bill on Anthropic's Claude platform simply because they forgot to implement employee usage caps.
The Shift to "Tokenmaxxing": Industry coverage from The Wall Street Journal and Axios points out that while early AI models were heavily subsidized, the pivot to usage-based token pricing has caught corporations off-guard. Employees trying to look tech-forward have been consuming massive compute power on casual tasks, leading tech giants like Meta and Microsoft to suddenly restrict or audit internal access.
Macro Warning: Google CEO Sundar Pichai even addressed this macro trend at Google I/O, noting that top enterprises are processing trillions of tokens a day and blowing past annual AI budgets before mid-year.
Would you like help digging up any of the specific articles from Briefs Finance or Axios covering the corporate fallout of these runaway token costs?
