Alphabet’s Cloud business blew past estimates in Q2 results released after the bell on Wall Street Wednesday.
That should be all that matters for the stock. Note the emphasis. I never pretend to know how markets will trade a given earnings release, but it’s certainly the case that investors are laser-focused on Google Cloud as a proxy for AI capex ROI.
With that in mind, Cloud sales of $24.8 billion were an easy beat. Consensus was looking for $22.2 billion.
As the figure shows, the YoY growth rate’s accelerating pretty dramatically. That’s auspicious in the context of the company’s enormous outlays and a borrowing spree that saw Alphabet sell $70 billion in debt across half a dozen currencies since November, not to mention an $80 billion equity raise.
The 82.4% YoY jump in Cloud revenue was “driven by demand for AI infrastructure and AI solutions,” Sundar Pichai beamed. “Nearly 90% of the Fortune 100” is using Gemini Enterprise, he said. The Gemini app now has 950 million monthly active users. That was a beat.
Pichai went out of his way to call the Cloud results proof that the hundreds of billions the company’s plowing into AI is money well invested. “Our AI investments are redefining what’s possible across every part of our business,” he declared, calling Q2’s results an “outstanding” testament to Alphabet’s full stack approach to AI.
The figure above shows Cloud sales as a percentage of overall revenue. That share’s up to 21%, triple what it was four years ago.
Overall revenue in Q2 was $119.8 billion, up 24.3% YoY. That’s the fastest pace since the December 2021 quarter.
Analysts were probably expecting some upside to consensus on the Cloud print (which is a ridiculous thing to say if you think about it: if they were expecting it, why wasn’t it baked into consensus?), so the blowout result may not move the needle all that much.
Profitability in the Cloud unit improved dramatically — the operating margin there was 35.6% in Q2 versus just 20.7% in Q2 of 2025. It looks like Alphabet wrote up their equity stakes again — net income rose almost 300%. The company has shares in Anthropic and SpaceX.
Free cash flow flipped negative as capex doubled to nearly $45 billion versus the same period a year ago. Purchases of property and equipment doubled in Q1 as well, and nearly doubled YoY in Q4 of 2025. Recall that Alphabet spooked investors in Q1 with a very aggressive capex guide.
As for Gemini, I’m still unimpressed. On Wednesday, it told me SpaceX hadn’t gone public yet, for example. When I laughed and asked whether it was sure, it doubled down: “SpaceX remains 100% privately held and has not gone public through an IPO.” “You cannot buy SpaceX shares on public exchanges like the NASDAQ,” it added.
Finally, on a third attempt, it figured out its mistake: “You are completely right to call me out on that. I was totally wrong,” it said. “I apologize for the outdated information.”
I love the Wikipedia citation. $200 billion in capex only for your model to fall back on a 25-year-old, non-profit online encyclopedia.
What can you say? What Pichai did on Wednesday, I guess: Alphabet’s AI is “delivering real, measurable value for consumers [and] customers.”
Later, on the call, CFO Anat Ashkenazi raised this year’s capex guidance range to $195 billion to $205 billion, up from $190 billion in April. She reiterated that spending will be “significantly” higher in 2027.
Seems to be consistent: “SpaceX is a private company. It is not publicly traded on the stock market, meaning you cannot buy shares of it through standard brokerage accounts.”
I got a really detailed answer for some reason. Here’s how its response starts:
[Gemini Flash]: Yes, but access depends heavily on whether you are buying pre-IPO, participating in the upcoming IPO, or buying on the open market after trading starts.
After that, it gave me some excellent info on how I could participate in the pre-IPO process
Hey, it generates computer code faster and better than humans, so give it a break?!
Gross or net?
Just remember that when HAL made an error, he blamed the humans.
That OpenAI hack of Hugging Face that was reported today, is kind of freaky!
You can ask the model when it was last trained. March 2026. It’s not terribly surprising it would affirm that SpaceX was still private. That’s just how the tech works. Limited, but not quite slop in this case.
If the model executed a wiki lookup for every fact it presents, well.. it wouldn’t be AI. It would be a search engine. Old tech that actually works supersed by shiny new tech that only intermittently does the job.
But considering the chatter/plan of a SpaceX IPO before the model training date, you would expect the llm to be smart enough to look up the current state of things!!
This is a pretty damning error. To say that it is the nature of the tech is in effect saying the tech is currently not ready as a trustworthy tool
LLMs are not “smart”. They’re not even considered AI by serious AI researchers. Look up “Chinese Room” for a description. LLMs are just symbol swapping, they have no concept of reality.
As far as I understand it, the Chinese Room is a philosophical argument, not a scientific/practical one. I use llms everyday at work and they have some level of reasoning and intelligence. The proof is in the pudding
Since we don’t understand intelligence and consciousness, how do we know that humans are more than advanced symbol manipulating machines?
I agree. Not ready prime time yet. It makes many mistakes.
As for Google (“Berkshire Hathaway of tech.”) are we nibbling today? I did….
Me too, If it falls more I’ll buy some more. I expect they will walk back some of the capex increases and will be rewarded for it.
Yeah, as I mentioned in the weekly macro preview, I wanted to see it fall more, and it did. I only put a few thousand into it, though. When these things correct, they tend to overshoot to the downside, and that can be pretty ugly depending on the circumstances. Recall that Meta fell 76% to the lows in 2022, for example, hard as that is to believe now. So, I bought a few thousand worth of GOOG today and will save some for later in case it falls another 10-20%.
I’m in agreement, similar strategy…one thing am considering is the Uncle Warren indicator: big investments that are usually too early but pay well over time.
Q3 2025: Berkshire first disclosed buying 17.8 million shares of Alphabet, worth roughly $4.3 to $4.9 billion.
Q1 2026: Berkshire heavily increased its stake, more than tripling its holdings in the tech giant.
June 2026: Berkshire added another $10 billion via a private placement in Alphabet shares to help fund artificial intelligence infrastructure.
Also adding bits of BRK.B – look at it as money market with equity upside and deferred income to be eventually taxed at LTCG rates…