Post-Cutoff

Review

Microsoft Slashes Spending on Anthropic AI Claude - AI is Too Expensive

Eli the Computer GuyYouTube100,604 views as of 12 October 2026

Watch on YouTubePlay loads YouTube’s player from youtube-nocookie.com.

Why it is here

Commentary on reports that Microsoft is cutting spending on Anthropic’s Claude because AI is too expensive; ~100k views. Which report it covers is not stated in the description (possibly The Information’s story in The Information: Meta and Microsoft work to wean employees off Claude; not verified).

Description

Description written by Gemini from the videoGemini 3.8 Flash, 11 October 2026

Summary
In this video, tech commentator Eli the Computer Guy discusses reporting from The Information detailing how major enterprise clients—specifically Microsoft and Meta—are drastically cutting internal employee spending on Anthropic’s Claude and Claude Code. Eli argues that the era of unconstrained corporate “tokenmaxxing” is coming to an end, highlighting how unsustainable token burn rates and corporate shift to in-house or open-weight models will impact the business models and multi-trillion-dollar valuations of frontier labs like Anthropic and OpenAI.

What is shown

  • [00:00] Screen display of The Information article headline: “Microsoft Slashes Internal Claude Spending by a Third”, featuring photos of Anthropic CEO Dario Amodei and Microsoft CEO Satya Nadella.
  • [00:13] Eli speaking directly to camera, analyzing corporate AI system architecture, the artificial inflation of ARR via unmonitored employee spending, and Michael Burry’s skeptical view of AI IPO timing.
  • [05:25] Scrolling and highlighting through The Information article text detailing:
    • Microsoft slashing projected internal Claude spending by over a third from an initial $1 billion projection [05:50].
    • Meta halving internal Claude Code users from ~60,000 to ~30,000 [06:27].
    • Meta’s internal AI tools, Muse Code and MetaCode, replacing third-party assistants [06:42].
    • Meta spending over $105 million on Claude Code within a single 28-day window [12:23].
    • Anthropic’s revenue concentration, noting two customers accounted for nearly a quarter of its revenue [12:40].
  • [14:30] Highlighting Microsoft Cloud and AI division policy changes restricting per-engineer monthly model spending allowances from $100,000 down to $10,000 [14:57].
  • [16:15] Eli breaking down multi-agent loops (“while true” loops), continuous repository scanning, and why runaway agentic token burn does not translate to proportional business value.

Claims & numbers

  • According to The Information and reporting cited in the video:
    • Microsoft was initially on track to spend at least $1 billion on internal Anthropic use, but has reduced that figure by more than a third (to roughly $666 million) [05:48].
    • Microsoft’s Cloud and AI group employs over 60,000 people, where individual employees previously had allowances up to $100,000 per month on AI models, which leadership capped to around $10,000 per month [14:33].
    • Meta’s internal Claude Code users dropped from roughly 60,000 to approximately 30,000 [06:27].
    • In a recent 28-day period, Meta spent over $105 million on Claude Code (over $1.2B–$1.3B annualized) [12:24].
    • Meta’s in-house AI coding tool MetaCode has over 30,000 internal users, while Muse Code reached over 6,000 employee users [06:53].
    • Anthropic’s draft IPO prospectus reportedly indicated two customers accounted for nearly a quarter of its revenue last year [12:40].
    • Anthropic reported roughly 6,000 enterprise customers spending $100,000+ over the 12 months through June, with over 100 customers spending over $10 million each [12:49].
  • The presenter claims Michael Burry has pointed out that Anthropic must rush to IPO around November 2026 to capitalize on peak annualized revenue run rates before post-Q3 enterprise spending cuts appear in financial statements [01:05, 04:27].

Notable quotes

  • [02:02] “In what other era of technology have CEOs literally come to the IT department... with fucking wheelbarrows full of cash saying, ‘You have to spend it’?”
  • [15:23] “Having your engineers burn $100,000 in tokens per month... per month! is never going to be sustainable.”
  • [16:40] “Essentially, it is literally burning tokens at the speed your systems can run at.”

Assessment
This is an independent analysis and reaction video reviewing published reporting from The Information. No software or hardware is demonstrated firsthand; the video consists of reading and providing commentary on leaked corporate memos and industry financial dynamics.

Described by gemini-3.8-flash on 2026-10-11 from the video’s audio and frames.