Version 8

On May 6, 2026, Anthropic held its 'Code w/ Claude 2026' developer conference in San Francisco, announcing a compute deal with SpaceX's Colossus 1 data center (300+ MW, 220,000+ GPUs) and doubled Claude Code rate limits for major subscription tiers.[2] The Verge subsequently reported the deal costs Anthropic approximately $15 billion per year, making the partnership's financial scale concrete for the first time.[3] A GitHub issue against the claude-code repository revealed that Max 20x plan users had not received the promised rate-limit doubling despite the announcement, while a separate Reddit post reported a 50% increase in weekly limits through July 13 — complicating the picture of what actually changed and for whom.[12][13] The event series expanded internationally with a Code with Claude London 2026 event on May 19, and MIT Technology Review published a notably ambivalent post-conference analysis on May 21.[20][23]

Anthropic is converting record-scale compute investment into a developer platform play, but two fault lines have emerged: a reported $15 billion annual price tag for the SpaceX deal raises hard questions about unit economics and long-term viability, and an announced rate-limit doubling that didn't reach all plan tiers undercuts the credibility of capacity claims. The Musk/Anthropic partnership's irony — Musk publicly called Anthropic anti-Western before SpaceX became its primary compute supplier — has entered mainstream commentary, adding reputational friction to what Anthropic framed as a geopolitically principled infrastructure decision.[5][6]

On May 6, 2026, Anthropic held its 'Code w/ Claude 2026' developer conference in San Francisco. Simon Willison attended and live-blogged the morning keynote sessions, serving as a primary public relay for announcements.[1] The same day, Anthropic disclosed it had signed an agreement with SpaceX to access the full Colossus 1 data center — over 300 megawatts and more than 220,000 NVIDIA GPUs — with access expected within a month.[2] Alongside the compute deal, Anthropic doubled Claude Code's five-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans and removed peak-hours rate-limit reductions for Pro and Max subscribers, framing both changes as a consequence of the new infrastructure capacity.[2] The SpaceX arrangement sits within a broader 10+ GW multi-provider compute portfolio spanning Amazon, Google/Broadcom, Microsoft/NVIDIA, and Fluidstack, and includes stated interest in orbital AI compute.[2]

The financial terms of the SpaceX deal quickly attracted scrutiny. The Verge reported the agreement costs Anthropic approximately $15 billion per year — a figure that, if accurate, is enormous relative to the company's current revenue and puts the sustainability of the partnership in sharper focus.[3] CNBC covered the deal as encompassing space development dimensions beyond ground-based compute.[4] The partnership's political framing generated pointed commentary: Elon Musk had previously tweeted that Anthropic 'hates Western Civilization,' making SpaceX's subsequent role as Anthropic's primary compute supplier a source of pointed irony across social media.[5][6] LinkedIn commentary questioned whether Anthropic had lost its moral high ground, and YouTube content speculated about Musk's motivations for reversing course.[7][8] Wired's headline 'Anthropic Gets in Bed with SpaceX' captured the skeptical editorial register, while Ars Technica and The Hill covered the deal factually.[9][10][11] Anthropic's own framing — citing partner selection criteria of 'democratic countries' with secure supply chains — invited further scrutiny given these circumstances.

On the developer side, the rate-limit announcement encountered an early credibility problem: a GitHub issue filed against the claude-code repository reported that Max 20x plan users had not seen the promised doubling of their five-hour limits despite the May 6 announcement.[12] A separate Reddit post reported that Claude Code weekly limits were increasing 50% through July 13, a distinct adjustment that complicated the picture of exactly what was changing and for whom.[13] Developer community reactions ranged broadly: some praised real-world productivity gains — site migrations, Java LSP tooling, creative projects[14][15] — while others noted that Claude Code still produced generic UI designs comparable to simpler tools.[16] The 'dual-wielding' practitioner meta — using Cursor for rapid inline autocomplete alongside Claude Code for heavier agentic tasks — emerged as a community-level pattern in the weeks following the conference.[17] One notable feature highlighted from the conference was CI auto-fix: when tests fail, Claude automatically generates a fix and opens a pull request for human review.[18] On the open-source front, a May 19 project demonstrated that structured guardrails could boost an 8B model from 53% to 99% accuracy on agentic tasks, signaling continued progress in the alternatives space.[19]

The Code with Claude event series expanded internationally: on May 19, 2026, Anthropic held a 'Code with Claude London 2026' event with its own opening keynote, drawing live viewers and generating social engagement across multiple languages including Japanese and English.[20][21][22] MIT Technology Review published a post-conference analysis on May 21 with the notably ambivalent framing 'Anthropic's Code with Claude showed off coding's future — whether you like it or not.'[23] Claude Code continued rapid versioning through mid-May, with version 2.1.139 introducing Agent View and Background Sessions features.[24] Developer community activity extended well beyond San Francisco — Bangkok-based meetups,[25] international tool-comparison threads,[17] and first-person productivity accounts reflect growing geographic breadth of Claude Code adoption.