Building a hyperscaler isn’t cheap
And: The current era of search is closing
Welcome to Cautious Optimism, a newsletter on tech, business, and power.
Happy Thursday, and hello from rainy Providence! So much for taking the girls to the park this afternoon. Starting today, Alibaba’s Q1 earnings included notable information concerning AI demand. We often consider US-predicated AI demand as the bellwether for global AI demand, mostly because the majority of large, public AI-serving companies are based here. Your Microsofts and Googles and Amazons and Metas and Nvidias, in other words.
But Alibaba’s earnings notes imply that AI demand is hot the world ‘round, helping bolster its Cloud Intelligence Group’s growth rate in the first quarter to 18%, far above the 11% it managed on a trailing-year basis. Even more, Alibaba told investors that “AI-related product revenue maintained triple-digit year-over-year growth for the seventh consecutive quarter” thanks to its “AI products are seeing broader adoption across a wide range of industry verticals including Internet, retail, manufacturing, and media, with a growing focus on value-added application.”
Nice. Strong global demand for AI compute implies that the TAM for modern genAI services is as big as we might have hoped. After all, if AI is going to be big in the States, and the EU, and China, well that’s about 60% of the global economy. — Alex
📈 Trending Up: AI safety data … testing limits … bad faith … Chinese GDP growth … data privacy … cybersecurity dealmaking … prices … baby podcasts? … spaghetti-at-wall-throwing … cans, kicked … SALTy negotiations … US control of Gaza? …
AI model prowess? Check out this release from Google, and this release from Meta, and ask yourself what’s coming next.
📉 Trending Down: AI independence? … not making whoopsies … cybersecurity … ethics? … births … EU comity … DOGE teeth … peace talks …
Building a hyperscaler isn’t cheap
CoreWeave reported its first quarterly results yesterday, blowing expectations out of the water. In Q1 2025, the GPU cloud reported top line of $981.6 million, far in excess of the $853 million expected by analysts. However, shares of CoreWeave are off 5% following the results.
Why? CNBC spoke with the company’s CEO, who said that analysts and accountants are going to take a little while to digest how the company counts future and committed revenues. Let’s dig into that a little bit.
CoreWeave expects Q2 2025 revenue of between $1.06 billion and $1.1 billion, adjusted operating income of $140 million to $170 million, and capex between $3 billion and $3.5 billion. For the full year, the neocloud expects revenues of $4.9 billion to $5.1 billion, adjusted operating income of $800 million to $830 million, and capex of $20 billion to $23 billion.
CoreWeave’s year-over-year growth was 420% in the first quarter. That’s insane.
How can CoreWeave support that level of spend? The company signs deals then spends against them. As its CFO said during its earnings call (transcript source);
The vast majority of our revenue comes from long-term committed contracts providing us with strong revenue visibility and attractive unit economics. This visibility in our growth enables us to take a success-based approach to capital investments that are matched to our customer contracts.
And how is that demand looking? Great, with a few caveats:
CoreWeave reports that its “revenue backlog,” made up of its “remaining performance obligation[s],” and “revenue in future periods under committed customer contracts” closed the quarter $25.9 billion. That figure grew 63% on a year-over-year basis.
And, CoreWeave also said that its new deals are driving growth:
In March, we signed an OpenAI strategic deal, the contract value for which is up to $11.9 billion […] We also added new enterprise customers and a new hyperscaler and signed a recent $4 billion expansion with a large AI enterprise. This $4 billion expansion was signed in Q2 and will be reflected in our revenue backlog numbers beginning in Q2. It is not included in our Q1 reported numbers. To be clear, this is additive to the Q1 ending revenue backlog of $25.9 billion.
Got that? Investors were a little bit puzzled. It turns out that the new $4 billion expansion and the OpenAI contract are not represented in the $25.9 billion, instead of merely the former. Hence why the company’s CEO noted that the analyst squad are taking a bit of time to sort out how to think about the company. Traditional SaaS this is not.
But apart from a growing backlog and new deals, the vibes were also good. Here’s how the company described overall AI demand:
We are just getting started and demand for our platform is robust and accelerating. In particular, we are excited to see the broad-based increase in demand for inference as well as the accelerating adoption of AI by our enterprise customers. AI growth and adoption remains severely limited by capacity constraints. CoreWeave is scaling as fast as it can to meet the demands of our customers.
All told the company remains an interesting, if binary bet that investors can make. Should AI demand grow as some expect, CoreWeave does seem to have its head screwed on straight when it comes to landing, financing, and meeting new contracted demand. But if AI demand growth slows, the company could find itself sitting atop a pile of debt that would feel increasingly burdensome.
Still, by pairing capex with contracts, it’s hard to imagine that CoreWeave will get too far over its skis in supply terms, unless a major customer dramatically cuts back on their spend. Which, given the way technology is evolving today, doesn’t seem too likely in the near-term.
Apple, India and China
You deserve a moment of levity, so please enjoy this quote from POTUS, as reported by TechCrunch:
“I said to him, ‘Tim, you’re my friend, I’ve treated you very good. You’re coming up with $500 billion, but now I hear you are building all over India. I don’t want you building in India. You can build in India, if you want to take care of India because India is one of the highest tariff nations in the world’,” Trump said.
Trump, who has been on a tariff offensive against most of the world, said Apple would now be “upping” production in the U.S., though he didn’t give a timeline or any specifics.
A good question to ask at this juncture is whether or not Apple makes iPhones in the United States. Nope, and signals that the company will move production of its flagship revenue driver to the States are currently nil.
That said, Apple is spending domestically. It announced plans to build some servers in the United States, and invest more in helping upskill domestic manufacturing. Good, and kind of the company to do less efficient spend to help allocate more of its opex to Americans. But, moving mass-production as Trump wants? Not going to happen.
Here’s an interview with the FT reporter whose new book on Apple in China has annoyed the company.
The current era of search is closing
Today Google’s search work is a firehose of profit. With $50.7 billion in Q1 2025 revenue, “Google Search & other” brought in more than half of the company’s $90.2 billion worth of total top line in the quarter. And it contributed the bulk of the company’s operating income, despite Google Cloud putting up more than 100% year-over-year profit expansion to $2.2 billion in the most recent quarter.
Despite its yet-impressive success, Google’s search hegemony is under threat. Not only is it facing court-forced changes to its business practices — forced data sharing, possible divestments, the proposed end of paid-preference deals with companies like Apple — the core search experience that Google has long held in a vice-grip is being challenged by AI-native products that are aces at answering questions instead of turning up a page of links buried under a barrage of advertisements.
Google, Microsoft, Perplexity, OpenAI, and other companies are betting that the future of search is more AI-predicated, and more definitive — answers instead of links. To that end, if you were to build a search product today, would you attempt to ape Google’s traditional offering more closely, or build something closer to the ever-present chatbot interface that we are all now intimately familiar in the AI era?
The latter, right? Agreed.
That’s why a recent move by Microsoft to deprecate “Bing Search APIs [on] August 11, 2025” feels like a big deal. A major tech company turning off a single API is not big news. But seeing Microsoft move away from offering up trad search as a paid service underscore how the search market is changing.
If demand for Bing’s API was hot, Microsoft would continue to support it. So long as the income was worth the annoyance. By sending it to the technology graveyard, Microsoft is telling us a little bit about its internal focus regarding search, and where market demand is flowing. Namely to AI-powered search tools (see CoreWeave’s demand notes above, and recall that Microsoft has been a customer of the GPU cloud).
So, the current era of search is coming to an end. It was great. Then it was fine. And by the end, we found ourselves in a world where Google was content to stuff results so full of paid placements and ML-cruft that at times you had to scroll past the fold to get a single native results.
Good riddance.
It will be a disappointment if AI search winds up similarly broken — a CO maxim: Any product dependent on advertising incomes will eventually compromise user experience for incremental revenue — but that may not happen as paying for AI access is the norm today, unlike paying for search. Perhaps that will keep AI search un-enshittified for a while yet.