Anthropic locks up SpaceX's Colossus 1, doubles Claude Code limits, and ships a 10-agent Wall Street pack

What landed, and when

Three Anthropic announcements stacked across roughly 36 hours:

May 5 (Tuesday) — Anthropic releases 10 finance-agent templates targeting labor-intensive Wall Street work: pitchbook construction, earnings analysis, KYC screening, month-end close, statement audit, valuation review, market research, meeting prep, model building, ledger reconciliation. Each template ships as a plugin in Claude Cowork and Claude Code, or as a Claude Managed Agent for firms running nightly schedules.

The same announcement names JPMorgan Chase, Goldman Sachs, Citi, AIG, and Visa as live deployments. Jamie Dimon appeared on stage with Anthropic CEO Dario Amodei — their first shared appearance — and disclosed personal use of Claude Code: “In 20 minutes, it created a huge dashboard…very accurate about what I wanted.” A $1.5 billion joint venture between Anthropic, Blackstone, Hellman & Friedman, and Goldman Sachs is creating an AI-native enterprise services firm targeting mid-market deployment.

May 6 (Wednesday) — Anthropic and SpaceX announce a compute partnership. Anthropic takes all of Colossus 1 — SpaceX/xAI’s Memphis data center — at 300+ MW with 220,000+ NVIDIA GPUs coming online within a month. Both sides also disclosed “interest” in working together on multiple gigawatts of space-based compute over a longer horizon.

Same-day operational changes:

• Claude Code’s 5-hour rate limits doubled across Pro, Max, Team, and seat-based Enterprise plans
• Peak-hour rate-limit reductions removed for Pro and Max accounts
• Claude Opus API rate limits “raised considerably”

May 7 (Thursday) — Microsoft 365 integration goes generally available. Claude Excel, PowerPoint, and Word add-ins reach GA; Outlook enters public beta. The integration maintains “one conversation across four apps” with automatic context synchronization between open files. Managed Agents add three new capabilities: Dreaming (research preview — agents review past sessions to self-improve), Outcomes (public beta — agents work toward developer-defined rubrics), and multiagent orchestration (public beta — lead agents delegate to specialist subagents).

The three announcements are connected: the capacity unlock made the rate-limit doubling possible; the Wall Street agents are what justify the capacity spend.

What it means for Claude Code users

For developers paying for Claude Code — the most directly user-impacting change in 2026 so far — the changes ship this week:

• Pro ($20/month) — Claude Code 5-hour limits doubled; no more peak-hour reductions
• Max 5x ($100/month) — Same doubling; no more peak-hour reductions
• Max 20x ($200/month) — Same doubling; no more peak-hour reductions
• Team Premium ($125/seat/month, min 5 seats) — Same doubling
• Enterprise (seat-based) — Same doubling
• API Opus customers — Rate limits “raised considerably” (specific numbers in Anthropic’s release notes)

The actual user-impact framing: many Claude Code users hit the 5-hour limit during active work sessions. Doubling that limit removes the most-cited friction in the product. Pro users who never hit the cap won’t notice; heavy users will notice immediately.

For users on lower tiers (free chat-only Claude.ai), no changes — the capacity unlock is being routed to paying developers and Claude Code first.

What it means for the enterprise story

Anthropic’s Claude Opus 4.7 launched April 16. Three weeks later, it has been deployed at five named Fortune 100 financial institutions, with Anthropic publishing the agent templates that wrap it. The fast turnaround tells you something about the buy cycle: large enterprises are no longer doing 6-month evaluations on frontier models. They are deploying the moment a model clears their internal quality bar and the surrounding agent infrastructure is shippable.

The Microsoft 365 GA is the other strategic move. Most enterprise work happens in Excel, PowerPoint, Word, and Outlook. Putting a single Claude conversation across those four apps — with context flowing automatically between them — is the kind of integration that becomes table-stakes once one vendor ships it. Microsoft Copilot has had this footprint for over a year. Anthropic now has it with a frontier-model-grade reasoning engine behind it.

For Moody’s, the deal puts credit ratings and risk data for 600+ million companies natively inside Claude, so finance analysts can query Moody’s directly without leaving the conversation. This is the shape of integration enterprise buyers actually pay for.

What it means for the compute story

The Colossus 1 deal joins a long list of Anthropic compute commitments now visible:

• Amazon — 5 GW (announced earlier in 2026)
• Google / Broadcom — 5 GW
• Microsoft / NVIDIA / Azure — $30 billion in committed Azure capacity
• Fluidstack — $50 billion infrastructure investment
• SpaceX / Colossus 1 — 300+ MW, all of one data center, 220,000+ GPUs

The pragmatic read on this stack: Anthropic is building a multi-vendor compute supply chain that doesn’t depend on any single provider, including AWS. This matters for two reasons. First, capacity is the binding constraint on Claude’s user growth — the rate-limit doubling that landed May 6 was waiting on hardware, not software. Second, the multi-vendor approach reduces the risk that any one cloud provider can squeeze pricing or block expansion.

The space-based-compute “expression of interest” is more speculative than concrete and shouldn’t be read as a near-term capacity source. Treat it as long-horizon optionality, not an announcement.

The broader pattern in May 2026

Anthropic’s May 5-7 stack lands the same week OpenAI made GPT-5.5 Instant the default ChatGPT model and Google DeepMind published its AI co-mathematician research preview. The competitive shape in May 2026: every major lab is shipping smaller, more focused updates — not headline frontier-model launches.

Anthropic’s particular move this week is the one most directly aimed at enterprise procurement. The five-named-customer announcement, the JV with Blackstone and Goldman Sachs, the Microsoft 365 footprint, and the agent templates targeting specific Wall Street workflows are a coordinated push to convert Opus 4.7’s quality lead into committed enterprise contracts before the next OpenAI or Google enterprise push.

For broader context on the AI competitive landscape, see the Claude review, the Claude Code review, and the head-to-head Claude vs ChatGPT comparison.

The pragmatic read

Three connected announcements, one strategic story: Anthropic now has the capacity to remove the user-facing friction (rate limits) that was capping Claude Code growth, and it has the enterprise commitments to justify both the capacity spend and the model investment. Whether the company can convert the announced enterprise momentum into recurring revenue at the scale the funding rounds imply is still an open question — but the May 6 rate-limit doubling is real and lands in users’ accounts this week regardless of how the longer-horizon story plays out.

For developers already on Claude Code: the friction you hit on May 5 is gone on May 6. For finance teams evaluating AI agents: the named-customer references and Microsoft 365 GA changed the procurement story this week. For everyone else: the capacity unlock is the structural news; the rate-limit doubling is the immediate one; the enterprise push is the strategic one.

Questions or corrections? Email Pick Right. Want the full list? See all news.