🧭 Anthropic Signs $10B Compute Deal with Nvidia-Backed Volta — 133 MW in Norway

Anthropic has inked a six-year, $10 billion contract with Volta Infra Holdings, a months-old infrastructure startup backed by Nvidia and Dell at a $2.4 billion valuation. The deal secures 133 megawatts of AI data-centre capacity at a facility being built in Norway, powered by Nvidia's next-generation Vera Rubin chip architecture. The site will be co-developed with Bitdeer, a crypto-mining company that is converting its hardware expertise to AI infrastructure. Bloomberg first reported the deal on August 4.

Why this deal is unusual

• Brand-new counterparty: Volta was founded earlier in 2026. Contracting $10B with a company that is months old signals how constrained conventional compute supply has become — Anthropic is effectively co-creating suppliers rather than selecting from an existing market.
• Norway location: Norway's abundant hydroelectric power and cold climate make it attractive for energy-intensive AI workloads. The facility will benefit from some of the lowest industrial electricity prices in Europe.
• Vera Rubin architecture: Nvidia's Vera Rubin is the successor to Blackwell. Securing capacity on this generation now, before mainstream availability, locks in a performance advantage for Anthropic's next frontier training runs.
• Stacking infrastructure bets: This follows Anthropic's earlier compute agreements with Google/Broadcom TPUs, Amazon Web Services, and reported discussions with Meta. The pattern is deliberate multi-source diversification — no single vendor dependency for the training compute needed to build Claude 5-generation models.

🧭 Former CA Supreme Court Justice Tino Cuéllar Joins Anthropic as First Global Affairs Chief

Anthropic has appointed Mariano-Florentino (Tino) Cuéllar as its first-ever Chief Global Affairs Officer. Cuéllar reports to President Daniela Amodei and will lead Anthropic's policy, strategic international engagement, and government relationships worldwide. The appointment — announced August 4 — comes as Anthropic navigates intensifying regulatory scrutiny across the US, EU, UK, and Asia, and as Dario Amodei has been publicly engaging on AI policy questions from open-weights regulation to national-security competitiveness.

Who is Tino Cuéllar?

• Justice, Supreme Court of California — authored landmark opinions on technology and privacy, international agreements, and the separation of powers before stepping down in 2022.
• President, Carnegie Endowment for International Peace — led one of the world's largest independent foreign-policy research institutions, with scholars in 20 countries.
• Stanford Law School professor and senior fellow at the Stanford Institute for Human-Centered Artificial Intelligence — he takes a leave of absence to join Anthropic.
• Anthropic Long-Term Benefit Trust — Cuéllar has served as a Trustee since January 2026, giving him a head start on the company's governance structure before starting as an executive.

Why this hire matters beyond the résumé

• Creating a Chief Global Affairs Officer role signals that Anthropic now views policy and government relations as a first-class function, not a PR add-on. This is a structural bet that regulatory outcomes will materially affect what Anthropic can build and deploy.
• Cuéllar's legal and judicial background — particularly on tech/privacy law — is well-matched to the EU AI Act compliance and US executive-order interpretation work that will dominate the next 12–18 months.
• His Carnegie background means he arrives with existing relationships in foreign ministries and multilateral bodies — relevant as Anthropic expands its international partnerships.

🧭 Claude Code 2.1.221: Focus View Declutters Chat, Two Shell Security Fixes

Claude Code shipped version 2.1.221 on August 4, introducing a VSCode Focus View that collapses tool activity into compact per-turn summaries, a new sandbox credential masking mode for Linux/WSL, and patches for two shell permission-check bypasses that could let hidden commands run without prompting.

Focus View — what it does and how to enable it

Working inside VS Code, heavy Claude Code sessions accumulate tool-call blocks — file reads, shell commands, search results — that push the actual conversation out of view. Focus View hides all of that behind an expandable summary with a live indicator while the tool is still running.

• Toggle: Ctrl+Alt+F (Windows/Linux) or the Command Palette → "Claude Code: Toggle Focus view"
• Scope: VS Code only for now; the JetBrains extension and terminal UI are not affected.
• Expansion: Click any collapsed turn to expand the full tool transcript — nothing is lost, just hidden by default.

Sandbox credential masking (Linux/WSL)

The new mode: "mask" option for sandbox credential files prevents secrets from leaking into the sandboxed environment. Sandboxed commands read a sentinel value; the sandbox proxy substitutes the real credential on egress. On macOS, file masking is not yet supported and falls back to deny.

Security fixes

• Bash bypass (zsh): A permission-check bypass allowed zsh to execute hidden commands inside [[ ]] regex conditionals without triggering Claude Code's permission prompt. Fixed in 2.1.221 — both shells now require permission for such constructs.
• PowerShell path-quoting bypass (Windows): A path-quoting flaw on Windows allowed commands to execute via PowerShell without permission. Also patched.