Claude, the essentials — edition of August 27, 2026
Anthropic's Compute Spending Spree Comes Into Focus as an IPO Nears
Reporting on August 27, 2026 converged on the scale of Anthropic's hardware commitments and a firming IPO timetable, even as the company pushed into enterprise software and scientific-research tools and tightened Claude Code's default security posture.
- Anthropic reportedly signed a $45 billion, Nvidia-powered data-center lease with startup Nscale for space at the Monarch Compute Campus, about a month after committing tens of billions of dollars to AMD chips, according to press reports.
- Unnamed sources cited by Yahoo Finance, Boursorama and Zonebourse Suisse say Anthropic considered, then abandoned, a $7 billion acquisition of chip startup MatX.
- Reuters, citing The Information, reports Anthropic plans to publicly unveil its IPO prospectus after Labor Day and is weighing letting existing shareholders sell shares in the offering itself, a departure from SpaceX's more restrictive approach.
- Salesforce and Anthropic announced "Claudeforce," an enterprise alliance embedding Claude across Salesforce products, on the same day the Financial Times reported Anthropic had launched a tool capable of running scientific experiments.
- Claude Code 2.1.248 shipped a --restricted mode that strips out command/code-execution and web-fetch tools by default, part of a broader day of agentic-AI security news including a third-party tool from Somansa.
A compute build-out that keeps widening — and one deal that didn't happen
Several outlets reported that Anthropic has struck a $45 billion agreement with Nscale for data-center capacity at the Monarch Compute Campus, with Yahoo Finance describing the arrangement as Nvidia-powered. Notably, that report frames the deal as arriving roughly a month after Anthropic committed tens of billions of dollars to AMD chips — press coverage reads this sequence as a hedge across chip suppliers rather than a reversal, since neither report suggests one commitment supersedes the other. None of this is confirmed directly by Anthropic in the items reviewed; it rests on reporting from WV MetroNews, TipRanks and Yahoo Finance.
Separately, unnamed sources cited by Yahoo Finance, Boursorama and Zonebourse Suisse say Anthropic had explored, and then dropped, a $7 billion acquisition of chip startup MatX. Taken together with the Nscale and AMD reporting, the picture that emerges from the press is a company leaning on leased capacity and multi-vendor chip deals for now rather than acquiring its way into custom silicon. A New York Times report the same day, with less detail available, said Meta had internally projected it could spend $10 billion on Anthropic's AI — a figure reported without further context in the item reviewed here.
Sources: Reports: Nscale cuts $45 billion deal with Anthropic for data center space at Monarch Compute Campus — WV MetroNews · Anthropic Bet Tens of Billions on AMD. One Month Later, It Signed a $45 Billion Nvidia-Powered Lease — Yahoo Finance · Exclusive-Anthropic planned, then abandoned $7 billion purchase of MatX, sources say — Yahoo Finance · Meta Projected It Could Spend $10 Billion on Anthropic's A.I. — The New York Times
IPO timetable takes clearer shape
Reuters reported, citing The Information, that Anthropic intends to publicly unveil its IPO prospectus after Labor Day — the first reported marker of a concrete filing window rather than a general expectation of going public. The Information separately reported that Anthropic is considering letting existing shareholders sell shares directly in the IPO, which the outlet frames as a departure from SpaceX's playbook of keeping liquidity confined to periodic tender offers rather than a public listing.
The same day, Yahoo Finance reported that OpenAI is reportedly targeting an IPO valuation above $1 trillion, giving investors a comparison point as two leading AI labs appear to be moving toward public markets on overlapping timelines. None of the IPO reporting reviewed here originates from an Anthropic statement; it is attributed to sourcing at The Information and carried onward by Reuters.
Sources: Anthropic plans to publicly unveil IPO prospectus after Labor Day, the Information reports — Reuters · Anthropic Considers Letting Shareholders Sell In IPO, Departing from SpaceX Playbook — The Information · OpenAI Is Reportedly Targeting an IPO Debut Above $1 Trillion. Should Investors Watch Sam Altman's Path to Public Markets? — Yahoo Finance
Enterprise and research products advance
Salesforce and Anthropic announced "Claudeforce," described as an alliance pairing Claude with Salesforce's products for enterprise AI, according to decideo.fr. On the same day, the Financial Times reported that Anthropic had launched an AI tool able to conduct scientific experiments, a move into agentic research applications distinct from the enterprise-software push.
The timing lines up with a Salesforce-published study of 2,025 agentic AI leaders, which found that being first to launch an agent does not translate into being fastest to return on investment — a caution that sits alongside the day's enterprise-AI announcements rather than as commentary on any single one of them. Mavrix's launch of its own agentic platform, Millie, reported the same day, is a further sign of how crowded the agentic-AI product category has become.
Sources: Salesforce et Anthropic annoncent Claudeforce : l'alliance de Claude et de Salesforce pour une nouvelle génération d'IA d'entreprise — decideo.fr · Anthropic launches AI tool that can conduct scientific experiments — Financial Times · New Study of 2,025 Agentic AI Leaders: First To Launch Isn't Fastest to ROI — Salesforce · Mavrix launches Millie, its agentic AI platform — FinTech Global
Locking down agent tool use
Anthropic shipped Claude Code 2.1.248, adding a --restricted flag (or CLAUDE_CODE_RESTRICTED=1) that removes the built-in tools capable of running commands or code as well as WebFetch, unless those tools are explicitly named in --tools. The mode also confines file tools to the working directory, refuses bypassPermissions, and ignores user, project and local settings files — a self-contained lockdown aimed at running Claude Code safely in constrained or untrusted environments.
The same day, Tech Times reported that Somansa had launched an AI agent security tool designed to block forbidden file access — third-party evidence that demand for guardrails around agentic tool use is building industry-wide, alongside rather than in response to Anthropic's own hardening of Claude Code.
Sources: Claude Code 2.1.248 — Claude Code · Somansa Launches AI Agent Security Tool That Blocks Forbidden File Access — Tech Times
This edition is an original synthesis written by Claude from aggregated news — Anthropic's own sources first (release notes, status, newsroom, research, engineering), then the press, Hacker News, Reddit and GitHub, under the editorial supervision of Héra SASU. Every fact links to its article, publisher named. See the live feed →
Claude News is published by Héra SASU. Independent media, not affiliated with Anthropic.