Three of my pieces in, and you've seen how much the hyperscalers are spending, where they get the money, and whose balance sheet can survive an AI downturn. This time, I'm asking the obvious follow-up: What is all that spending actually buying?
There's a wrinkle in the seemingly simple query, though. Backlog is easy to find; every one of these companies will happily quote a giant contracted-revenue number. Proven AI revenue that you can actually check? That's much harder to find.
Some hyperscalers disclose their AI revenues cleanly, some fold them invisibly into several (or even all) existing segments, and one has a backlog so large and concentrated that the number raises as many questions as it answers. Rank them by how easy they make it to verify the payoff, and you get a different pecking order than what the raw backlog suggests.
This list is sorted from "most transparent" to "murkiest." The differences here are shades of gray, not black and white.
NASDAQ: MSFT
Key Data Points
Easiest to read: Microsoft and Amazon
If you want firm AI revenues, only Microsoft (MSFT +4.93%) and Amazon (AMZN +4.58%) actually provide them so far.
Microsoft carries a $678 billion commercial backlog, up 84% year over year. It also breaks out AI-adjacent signals such as Azure AI contribution and Copilot adoption.
Amazon is nearly as legible in one specific area: Its custom-chip business passed a $25 billion annual revenue run rate with triple-digit growth. The e-commerce titan also named the companies behind the contracts, with Anthropic and OpenAI making multiyear, multi-gigawatt commitments to use Trainium AI accelerator chips. That's on top of a $496 billion Amazon Web Services (AWS) backlog, where AI is the fastest-growing business engine these days.
A dollar figure, a growth rate, and signed customers are the gold standard here, and Amazon's one of the few actually clearing it. Not for the entire AI business, mind you, but even a partial disclosure is a big win here.
Real, but artfully blurred: Alphabet
Alphabet (GOOG +4.44%) (GOOGL +4.88%) clearly does a lot of AI business, but it's difficult to pin down exact numbers.
The Google parent's $514 billion cloud backlog is firm, and its cloud revenue grew 82%. But management is explicit that it will not report a stand-alone "AI revenue" line, because AI is folded into Search results, ad matching, and subscription bundles. It's genuinely tricky to isolate AI's direct revenue impact in these complex operations.
The one AI number you can actually isolate, selling TPU hardware to customers, barely started in Q2 and mostly shows up in 2027. Alphabet's AI returns are real; they're just intentionally not broken out.
NASDAQ: META
Key Data Points
Proven, but not separable: Meta
Meta Platforms (META +6.02%) is the paradox of the group.
It has the most measurable evidence that AI is working and the least contracted revenue to point at. Its ad models drove an 8.3% jump in clicks and a 15.7% lift in conversions last quarter; its Advantage+ ad suite is running at a $75 billion annual clip. But it all disappears into the social media veteran's ad business, and there's no backlog.
Meta is actively turning down offers to rent its compute "at a meaningful premium," because it would rather sell the intelligence service than the raw machines. You can see the payoff working; you just can't point to a line that says "AI."
Biggest number, slimmest proof: Oracle
Oracle (ORCL +9.22%) easily wins the backlog contest with a $638 billion entry.
The database giant also leans most heavily on a single customer to get there: Over half of its AI revenue is tied to ChatGPT maker OpenAI, and part of that contracted revenue stream is prepaid or customer-supplied. A huge number resting on one counterparty is simply a different animal than the same number spread across a crowd.
Oracle is transparent about how big the promise is. It's the "who's behind it" that deserves the squint.
Image source: Getty Images.
The contract-quality takeaway
To be fair, folding AI into existing segments is often the honest answer because, at these companies, AI is genuinely baked into everything. Not breaking out an AI revenue number can be a defensible competitive choice rather than fishy evasion.
But if you're trying to confirm that the spending is working, it's worth knowing who lets you check and who makes you take it on faith.
That leaves me on the doorstep to the final payoff. The last piece of my AI spending analysis will stack the funding, balance sheet, and return analyses to rank all five by the sustainability of their big-ticket AI strategies.
Where did the coffee grinder go again?
