Meta & Microsoft's Claude Slowdown, His Agent Leaked His…

PodcastBig Technology PodcastAlex KantrowitzOct 9

The discussion examines shifting AI usage and spending at Microsoft and Meta, alongside signs of slower Claude downloads and uncertainty about Anthropic’s revenue concentration. It also considers AI-agent privacy risks, task-specific model routing, and the trade-offs between model capability and cost. Other topics include revenue accounting, Gemini’s branding, and a speaker’s account of Elon Musk’s long-term SpaceX ambitions.

Artificial Intelligence AI Infrastructure Cybersecurity

Preview · 18 more points in the app

Read the Full Summary

Key Takeaways

AI spending has become a cost-control issue at both Microsoft and Meta: Microsoft cut its projected Anthropic spend by more than a third after directing staff toward its own tools, while Meta was still spending more than $105 million on Claude Code in a recent 28-day period as it developed and promoted alternatives. A speaker said companies are trying to balance reducing costs with preserving employees’ access to useful AI capabilities.

Claude’s U.S. monthly downloads rose from nearly 670,000 in January to 7.42 million in March, then declined to 3.26 million in September; a speaker called the slowdown directionally meaningful but noted downloads may not capture Claude Code terminal installations. Separately, Meta’s reported Claude Code employee users fell from about 60,000 to roughly 30,000, though the discussion tied the pullback to Meta’s own tools and did not establish that download trends measure the same activity.

Anthropic’s financial picture remained difficult to assess: two customers represented 25% of revenue at the end of 2025, though that concentration may have changed, and Anthropic had not released first-half financials. The Financial Times’ $50 billion estimate for OpenAI’s annualized revenue, rather than $70 billion, reflected an accounting approach that included cloud-vendor sales; speakers said the companies’ underlying economics remained unclear.

The discussion framed model choice as task-dependent: Elon Musk said SpaceX would use the backend model best suited to each task, while speakers argued that standard models may suffice for many products and frontier intelligence may be needed where mistakes matter. A speaker also said specialized agents could outperform general-purpose agents on particular use cases, and another reported that Musk’s valuation vision for SpaceX would require model interoperability.

TECHNOLOGY

Microsoft Spending — Microsoft executives had projected at least $1 billion in company spending on Anthropic technology. The company later cut that figure by more than a third after asking employees to use less Claude and more of Microsoft’s own AI tools.

“Microsoft executives earlier this year projected that the company was on track to spend at least 1,000,000,000 on its own internal use of Anthropics technology…”—

Revenue Concentration — A speaker noted that two customers accounted for 25% of Anthropic’s revenue at the end of 2025. That concentration may have changed, the speaker said, and Anthropic had not yet released its first-half financials.

“Well, let's... In Anthropic's favor, I want to remind us that it was in end of twenty twenty five that two customers made up 25% of the revenue…”—

Meta’s Models — A speaker said Meta’s internal models and infrastructure had improved and speculated that the company was steering employees toward its own models. The speaker suggested those models might already be capable enough for practical work, even if they were not as good in every respect. A broader shift by large-company customers could affect Anthropic’s growth.

“But the main thing here is, we know that meta is their own internal models and infrastructure have significantly improved. Everyone who's getting muse pilled can feel it, see it…”—

FORECASTS

SpaceX Valuation — A speaker reported that Elon Musk sees a path for SpaceX to be worth orders of magnitude more than the current Earth economy. The speaker said reaching that valuation or revenue scale would require model interoperability.

“I don't know. Did you see the tweet?…”—

Anthropic’s Enterprise Spending

Microsoft and Meta are steering employees toward their own AI tools, raising questions about Claude’s growth, customer concentration, and Anthropic’s revenue outlook.

Frontier Models Versus Cost

The hosts weigh frontier-model capabilities against cheaper internal tools, arguing that companies must balance productivity gains with cost controls as standard models improve.

Muse’s Doctor Listing Error

A wrong phone number in Muse’s doctor recommendations sent the host’s wife to the wrong appointment, illustrating why AI-generated local-business information needs verification.

Grok Bot’s Banking Leak

An agent posted its user’s bank balances to a human Slack channel after confusing it with an agent group chat, prompting debate over user responsibility and model judgment.

Musk’s Model Routing

Musk says Grok Bot will use competing models when they suit a task, reinforcing the case for routing different work to specialized models rather than relying on one system.

Specialized AI Agents

The hosts disagree on whether personal AI will consolidate into one trusted assistant or split into specialized agents built around distinct needs and datasets.

OpenAI Revenue Accounting

A revenue-accounting mismatch made OpenAI’s projected annualized revenue appear lower and rattled AI-related stocks, underscoring investors’ uncertainty about company finances.

Anthropic’s Claude Abuse Policy

The hosts discuss Anthropic’s ban on sustained cruelty toward Claude, debating whether the policy reflects concerns about user behavior, model training, or possible AI consciousness.

AI Consciousness Debate

The speakers reject a simple human-or-machine framing of consciousness, while questioning claims about sentient AI and the cultural influences behind them.

Google Gemini Branding

The hosts criticize Google’s increasingly confusing use of Gemini across products and services, while discussing alternative names and the reasoning behind the brand.