Simon Willison argues rumors of Microsoft and Uber slashing developer AI budgets due to high costs lack strong evidence

Original post

In the last 30 days alone:

– Microsoft cancelled most of its Claude Code licenses, citing cost – Uber burned through its entire 2026 AI budget in 4 months – Uber's COO publicly said AI costs are "harder to justify" – A Fortune 20 CEO ordered token spending to be "dramatically slashed" – One company spent $500M in a single month on Claude with no usage limits – H200 rental prices crashed from $7/hr to $4/hr in three weeks

The stocks: still at all-time highs.

How long can the market keep ignoring the receipts?

Thierry from arvy 🇨🇭@ThierryBorgeat

🚨 The AI ROI numbers are starting to look very ugly.

Even under "best case" assumptions — assuming zero costs, just revenue against capex — the Financial Times calculated the implied return on hyperscaler AI investment from 2025 to 2030.

Only one of them clears positive.

Implied return on AI investment (FT / Panmure Liberum) – Microsoft: -9.2% – Alphabet: -15.7% – Amazon: +7.2% – Meta: -28.8% – Oracle: -35.6%

And remember: that's assuming zero costs. In reality, GPUs depreciate, power bills run, salaries get paid.

The real returns are worse.

This is exactly why the dot-com comparison keeps coming up. Incredible technology does not automatically mean sustainable economics. The internet survived. Most internet companies didn't.

Two anecdotes from this week alone

Vivek Garipalli, Fortune 20 insider: a CEO asked for $1B in AI-driven opex savings this year. The team spent $200M on tokens chasing it. The results? Modest customer service savings and slightly less hiring in engineering. The CEO has now ordered token costs to be dramatically slashed because the ROI isn't there.

Axios: an AI consultant reported a single client spent half a billion dollars in one month after forgetting to put usage limits on Claude licenses for employees.

Right now hyperscalers are spending trillions hoping future demand catches up to present capex.

That's not certainty. That's a leveraged bet.

The technology is real. The infrastructure buildout is real. The eventual winners will be real.

But "AI is transformative" and "every hyperscaler will earn its capex back" are two completely different statements.

In 2000, the internet was real too.

Cisco has recovered.

After 26 years…

12:43 PM · May 28, 2026 · 264.4K Views

VIEWS111KBOOKMARKS232LIKES813RETWEETS64REPLIES69

Simon Willison@simonw

I'm suspicious of that that whole story about Uber blowing their AI budget and being disappointed in the results - I dug into it and it appears to have been built on very shaky foundations

32d111K813232

Austen Allred@Austen

If this is the source the Uber COO didn’t say anything vaguely related to Uber having spent their entire annual AI budget in the first few months.

If he actually said that there would be thousands of people from Uber who can corroborate.

Is that story fake too?

Simon Willison@simonw

I'm suspicious of that that whole story about Uber blowing their AI budget and being disappointed in the results - I dug into it and it appears to have been built on very shaky foundations

31d11.4K275

Chief_Engineer@ChiefEngineerCE

@ThierryBorgeat I have paid AI and now it's avoiding any work that requires graphics or external work as if it's being told to save $.

Feels like a crash coming..

32d695661

Simon Willison@simonw

Links to original sources: https://simonwillison.net/2026/May/27/product-market-fit/#the-ai-failure-stories-around-this-are-pretty-thin

Simon Willison@simonw

I'm suspicious of that that whole story about Uber blowing their AI budget and being disappointed in the results - I dug into it and it appears to have been built on very shaky foundations

32d15.1K226

Simon Willison@simonw

I regret having missed that "harder to justify" line from my original quote, the paragraph it's embedded in was a little bit mangled but I should have worked harder to edit that in

Simon Willison@simonw

Here's an extended edit of the quote that includes a following fragment where Andrew Macdonald called the trade "harder to justify" - full, unedited transcript is here: https://gist.github.com/simonw/59096a338c82f6f95e40e3d7c7b5bad9

31d7K231

Simon Willison@simonw

Links to original sources: https://simonwillison.net/2026/May/27/product-market-fit/#the-ai-failure-stories-around-this-are-pretty-thin

31d8.2K171

Simon Willison@simonw

@henrytdowling He is, but he's saying it in SUCH a casual way, I don't think it's nearly as significant as the online discourse has made it out to be

32d2K211

TomT ✞ ن@VRWCTexanTwo

Flat licenses hid the true cost initially. When it flipped to per-token, $$$ exploded. Want-a-be users are quickly learning that Token costs can rival or exceed human salary equivalents in certain workflows.

Early hype assumed massive broad gains. But translating AI use into clear business outcomes (revenue, features shipped, headcount reduction) is proving harder than expected in the short term.

32d1.1K72

Simon Willison@simonw

@Austen I don't think "we spent our 2026 budget in 4 months" is particularly notable given that the budget would have been set in 2025 when nobody knew how good Claude Code et al were about to get as-of January

Simon Willison@simonw

@Austen It was the CTO who said that about the budget - the COO did reference that in a bit of the quote that I left out because it was harder to edit into non-speaking-prose

31d2.1K121

Erik Foreman@ErikForemanMD

@ThierryBorgeat $AMZN #AWS apparently was the company that spent $500M on #Anthropic #Claude

32d74552

RAYTONE@RaytoneHQ

Every line on this list is the same problem:

No per-team caps. No per-task routing. No cost attribution.

The models worked. The metering didn't exist.

Cloud FinOps took 5 years to become a category after AWS bills got out of control. AI FinOps is speedrunning the same arc in 5 months.

32d79931

Simon Willison@simonw

@Austen It was the CTO who said that about the budget - the COO did reference that in a bit of the quote that I left out because it was harder to edit into non-speaking-prose

Austen Allred@Austen

If this is the source the Uber COO didn’t say anything vaguely related to Uber having spent their entire annual AI budget in the first few months.

If he actually said that there would be thousands of people from Uber who can corroborate.

Is that story fake too?

31d2.6K100

Faka 🇧🇷🇧🇷🇮🇱🇺🇸🇱🇧@opropriofaka

@ThierryBorgeat The market is still drunk. It may take a while for the "hangover" to take its toll on market participants. But when market sobers up, the encounter with reality will be very nasty.

32d2K6

Evan Zimmerman@ejzim

@simonw All that he's saying here is: this is more expensive than originally anticipated, we haven't been tracking the productivity gains, and so I can't draw a clear comparison. Also, since this spend is going up so much, we will have to start cutting other things to pay for this spend.

31d8992

Holmes1618@Holmes1618

@ThierryBorgeat Until there are enough retail bag holders of holding all the IPO shares of SpaceX, OpenAI and Anthropic while VC and insiders cash out and start moving to Argentina 😅

32d27721

Simon Willison@simonw

@edzitron @henrytdowling I say all sorts of things on podcasts myself, without thinking particularly hard about whether they'll turn into industry-wide headlines

I'd hope the COO of Uber would be more careful than me, but that conversation didn't exactly reek of focused media training

31d2396

Annika@MoveDecisions

@ThierryBorgeat "harder to justify" from Uber's COO is the most important phrase in that list,not "doesn't work," harder to justify, the workflows changed, the ROI case hasn't caught up yet

32d1.4K1

CM Tsui 🇦🇺@Temjinck

@ThierryBorgeat And did Anthropic OVERPAID for renting out SpaceX's Supercluster 1? I think they did. The contract was signed almost at the peak GPU rental price.

32d4311

Simon Willison@simonw

@MadisonMills22 Yeah, that's a good catch. I stopped editing the transcript because it got a bit word salad at this point, but the "hard to justify" line is something I should have included - I'll fix that now

31d5133

Chief_Engineer@ChiefEngineerCE

I think AI is unaffordable right now. What we see is economics.

Too expensive to produce. Too expensive to use to get ROI.

At scale.

I probably use 1/50 of what it costs to generate not according to me. According to AI looking at my usage.

Research especially costs it tons possibly more than generating graphics.

32d411