Meta, Microsoft, and Google blew blinding amounts of AI smoke during earnings—here’s what we actually learned

The three giants of the AI valley have spoken. What have we learned?

Not much—other than the fact that all three ( Alphabet , Microsoft , and Meta) see no limit to the amount of wealth they’re itching to pump into their AI projects .

Over the course of three hour-long earnings calls, the only hard numbers about AI that we heard this week related to capital expenditures. The Big Tech trio spent a combined $40.5 billion on the infrastructure, land, and chips that power their AI services during the second quarter. And each company indicated that those numbers will only get bigger next year.

The rest of the AI info that was served up—particularly anything related to the revenue side of the ledger—amounted to mostly empty rhetoric.

Let’s start with Meta. On Wednesday CEO Mark Zuckerberg said the company was on track to have “the world’s most used AI assistant by the end of the year.”

Zuck didn’t provide much context to help listeners understand what this feat might mean. Is an AI assistant in the same category as a “copilot,” the generative AI tool that rival Microsoft is fervently pushing to its customers ? Unclear.

Let’s assume an “AI assistant” is a traditional chatbot, in the same class as OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini. How does Meta AI stack up? Search around, and you’ll discover that even trying to make an apples-to-oranges comparison is not easy. Google said in May that “over 1 million people have signed up to try Gemini Advanced in just three months,” and on its earnings call last week CEO Sundar Pichai said that Gemini is now used in “all six of our products with more than 2 billion monthly users”—nice large numbers that don’t really tell us anything. For its part, OpenAI said in May that “every week, more than a hundred million people use ChatGPT.”

All of this suggests that Meta, which now has 3.3 billion daily users across its family of apps , could theoretically declare victory if at year’s end there are 101 million people per week using the AI assistant that it has very conspicuously bolted on to Instagram, Facebook and WhatsApp. The bar doesn’t appear to be very high.

As for when these Meta AI users will lift the company’s top line, Zuckerberg takes a long-term view: Generative AI will not be a meaningful driver of the business for “years.” Don’t expect generative AI to move the needle next year, or even in 2026, Zuckerberg said—AI revenue in 2025 and 2026 will come from the traditional, behind-the-scenes AI tools that the company uses to improve feed recommendations and advertising, but not from Meta AI.

Satya Nadella had lots of impressive sounding but similarly hazy data points to share on the Microsoft earnings call , and he unleashed them in a rapid-fire barrage. Azure AI customers increased by more than 60% year over year, with roughly 60,000 customers in total. Microsoft 365 customers increased 60% from the previous quarter, with the majority of enterprise customers “coming back to purchase more seats.” And the number of customers with more than 10,000 Microsoft seats more than doubled quarter over quarter, we learned, but it was left unspecified if those are all paid seats. (Microsoft confirmed after publication that they were paid seats).

The closest Nadella came to dropping a hard number involved GitHub, and the AI “copilot” that helps computer programmers write code.

GitHub’s annual revenue “run rate” is now $2 billion, Nadella said. Run rates are a slippery thing, of course—what is the $2 billion run rate extrapolated from: revenue rung up in the most recent quarter? Month? Day? Hour? Nadella didn’t say. But he noted that copilot accounted for over 40% of GitHub’s revenue growth this year (note that this revenue growth is not the same thing as the just-mentioned revenue run rate, so that $2 billion number is not relevant). He then said that GitHub copilot is already a larger business than the entirety of GitHub when Microsoft acquired the company back in 2018.

There’s no public number for GitHub’s revenue at the time of its acquisition, since it was a private company. But a New York Times article at the time cites analyst estimates that GitHub’s revenue was “running at $200 million a year.”

So there you have it, one somewhat solid data point about AI from the Big Tech earnings reports: GitHub copilot’s revenue is somewhere just north of $200 million a year, and very likely below $400 million (since Nadella would presumably have described the business as “more than twice the size” if that were the case, right?).

Wall Street analyst notes recapping the earnings are full of commentary about “strong demand signals” and promising AI uptake. And Meta’s stock is riding high today, following Wednesday’s earnings report, thanks to the enduring strength of its advertising business. But when you look closely at what we actually learned about generative AI, there’s not much there there.

Correction: The number of Microsoft 365 copilot customers increased 60% in Q2 , rather than doubled. And the number of customers with at least 10,000 enterprise seats more than doubled, as opposed to the total number of enterprise seats being 10,000.

This story was originally featured on Fortune.com