Microsoft bans Claude to cut AI costs

**SPEAKER_1** (0:01)
Ray-Ban Meta lets you explore the world without a screen getting in the way, so you can stay present in the moment.

**SPEAKER_2** (0:06)
Hey Meta, tell me what kind of dessert this is.

**SPEAKER_3** (0:09)
That's a Stroopwafel, a Dutch waffle with spiced syrup in the middle.

**SPEAKER_2** (0:13)
Is it sweet?

**SPEAKER_3** (0:14)
Yes, perfect for a snack or dessert.

**SPEAKER_2** (0:18)
Mm, delicious.

**SPEAKER_1** (0:20)
Get answers on the go without interrupting your flow. Ray-Ban Meta, iconic style meets Meta AI, available at Walmart and other authorized retailers.

**SPEAKER_4** (0:30)
Morning Decisions.

**SPEAKER_5** (0:32)
How about a creamy mocha frappuccino drink or a sweet vanilla?

**SPEAKER_3** (0:36)
Smooth caramel maybe or a white chocolate mocha.

**SPEAKER_5** (0:39)
Whichever you choose, delicious coffee awaits.

**SPEAKER_4** (0:41)
Find Starbucks Frappuccino drinks wherever you buy your groceries.

**SPEAKER_5** (0:45)
When you finally find your thing, you want the whole world to know about that thing. So you use a thing called Canva to make it an even bigger and better thing.
Whether you want to create flyers for that thing, make presentations for that thing or design merch for that thing, you can do anything so people can see your thing, feel your thing, love your thing. The next thing you know, it's a thing. Canva, the thing that makes anything a thing.

**SPEAKER_6** (1:15)
Microsoft is canceling internal licenses for an AI coding tool called Claude Code, literally forcing the engineers working on Windows, Teams and Surface to use Microsoft's own homegrown tool, GitHub Copilot.

**SPEAKER_4** (1:29)
I mean, the irony there is just hard to miss. A company that is pushing AI into every conceivable product, had to cut off access to a competing AI tool simply because its own engineers liked it too much and were using it constantly.

**SPEAKER_6** (1:42)
Yeah. The primary driver for the mandate is, well, it's entirely financial. This was a massive, completely unexpected budget drain hitting right as their corporate fiscal year was coming to a close.

**SPEAKER_4** (1:53)
So they look at the runaway costs of their own developers using a third-party tool and they just pull the plug to stop the bleeding.

**SPEAKER_6** (1:58)
Exactly.

**SPEAKER_4** (1:59)
But if the wealthiest tech giants are suddenly hitting a wall with their own AI usage costs, what happens when the underlying math of the entire AI boom starts to break?

**SPEAKER_6** (2:08)
Well, to figure that out, we have to look at the fundamental difference in how AI agents are built compared to traditional software because instead of a flat fee subscription, AI relies on token-based pricing.

**SPEAKER_4** (2:18)
Right, which is a huge shift.

**SPEAKER_6** (2:20)
It is. Every single prompt, code review or even just a background task costs money.
For decades, the industry just paid a set price per user per month, but not anymore.

**SPEAKER_4** (2:33)
It's like paying for an all-you-can-eat buffet, but suddenly the restaurant starts charging you per byte.

**SPEAKER_6** (2:38)
Per byte, yeah. That's exactly it.

**SPEAKER_4** (2:39)
As these AI agents get smarter and run multiple parallel tests autonomously, the bytes just multiply invisibly in the background.

**SPEAKER_6** (2:47)
Because a token is essentially just a fragment of a word, right? So when a developer uses an AI agent to read a code base, the system processes millions of these tokens, and the raw numbers from the sources show exactly how fast that accelerates. One mysterious corporation actually burned through half a billion dollars in a single month because they forgot to say usage limits on Claude. Wow.
Yeah, half a billion. Uber exhausted a $3.4 billion AI budget in just four months, with individual engineers racking up thousands of dollars a month in tokens.

**SPEAKER_4** (3:19)
If you are running a business, managing unpredictable software costs that just scale exponentially with employee usage creates an impossible budgeting scenario.

**SPEAKER_6** (3:27)
Oh, absolutely.

**SPEAKER_4** (3:27)
I mean, a CFO relies on predictable operating expenses, you know. When you give your workforce a tool that essentially charges by the thought, and those thoughts happen thousands of times a second, your financial forecasts are useless.

**SPEAKER_6** (3:41)
Which connects directly to the internal politics at Microsoft. Internal tracking actually documented that their engineers explicitly preferred Claude Code over Microsoft's own Copilot CLI.

**SPEAKER_4** (3:54)
The CLI being the command line interface.

**SPEAKER_6** (3:56)
Right, exactly. But Microsoft shut the preferred tool down anyway under the banner of tool chain unification.

**SPEAKER_4** (4:03)
Tool chain unification. I mean, that's such a classic corporate override. Moving everyone to Copilot keeps the data in-house and avoids sending engineer keystrokes to a competitor backed by Amazon.

**SPEAKER_6** (4:13)
Yeah, which makes sense of a security standpoint. But that dynamic totally skews the industry data we see, because Microsoft will inevitably report massive internal adoption numbers for GitHub Copilot.