AgenticMarket
CONTEXT WINDOW — Issue 004: Anthropic's 80x Growth, SpaceX GPUs, and the Battle for AI Infrastructure
CONTEXT WINDOW — Issue 004
Week of May 11–17, 2026
Signal from the MCP ecosystem. No noise.
What Shipped
Anthropic grew 80x in Q1 2026 and secured every GPU at SpaceX's Colossus 1 to keep up
On May 6 at the Code with Claude developer conference in San Francisco, Anthropic CEO Dario Amodei disclosed the number that explains everything about this company's past six months: Q1 2026 revenue grew 80 times year over year.
Anthropic had planned for 10x growth. It got 80x. That gap — 8x the expected demand against infrastructure built for a different scale — is what caused the rate limits, the peak-hour throttling, and the degraded Claude Code experience that users reported through March and April.
The fix announced the same day: Anthropic signed a deal to use all compute capacity at SpaceX's Colossus 1 data center in Memphis, Tennessee. The facility hosts more than 220,000 Nvidia GPUs — H100s, H200s, and Blackwell-generation GB200 accelerators — with 300 megawatts of raw power. Anthropic adds this on top of already-announced commitments: up to 5GW with Amazon, 5GW with Google and Broadcom, $30 billion of Azure capacity with Microsoft and Nvidia, and a $50 billion U.S. infrastructure investment with Fluidstack.
Immediate user impact as of May 6:
- Claude Code 5-hour rate limits doubled for Pro, Max, Team, and seat-based Enterprise plans
- Peak-hour limit reductions removed for Pro and Max accounts
- Claude Opus API rate limits increased considerably for all tiers
The most important sentence from Amodei on stage: "That is the reason we have had difficulties with compute. We are sorry if sometimes it takes time, but we will continue to acquire as much capacity as possible." Executives at Anthropic had reportedly considered removing Claude Code from the $20/month Pro plan entirely due to infrastructure pressure. The SpaceX deal closed that option off, at least for now.
The Musk angle is impossible to ignore. Musk called Anthropic "misanthropic and evil" in February 2026. He spent the week before the deal announcement with Anthropic's leadership team and posted that he was "impressed." SpaceX, in which Musk holds a 42% stake, then leased its flagship AI compute facility to the company Musk had been attacking publicly. The compute race makes for strange alliances.
Separately: Anthropic also reportedly signed a five-year, $200 billion compute commitment with Google Cloud, per The Information.
Anthropic's valuation crosses $900 billion in latest round
The compute deals happened in context. Anthropic's annualized revenue run rate crossed $30 billion in Q1, up from $9 billion at the end of 2025. The company closed a $50 billion fundraising round at a reported valuation above $900 billion — and secondary market trades are pricing shares at an implied $1 trillion or higher.
To put the 80x growth number in perspective: Anthropic's Q1 2026 revenue likely exceeded its full-year 2025 revenue. At this growth rate, maintained for even two more quarters, Anthropic would be one of the fastest-growing companies in technology history by revenue.
The reported IPO timeline remains October 2026.
OpenAI launched a $4B deployment company and acquired Tomoro
On May 11, OpenAI announced the OpenAI Deployment Company — a new majority-owned entity designed to embed Forward Deployed Engineers directly inside customer organizations.
The structure: $4 billion in initial investment from 19 global firms, consultancies, and system integrators. Lead investors include TPG, Bain Capital, Brookfield, Goldman Sachs, SoftBank Corp., and Warburg Pincus.
The acquisition: alongside the launch, OpenAI agreed to buy Tomoro, an applied AI consulting and engineering firm with approximately 150 engineers and clients including Virgin Atlantic, Tesco, and Supercell. Tomoro gives the deployment company an immediate team rather than building from zero.
The model: FDEs embed inside organizations, redesign critical workflows around AI, and build durable systems. This is explicitly not "here's an API and an SDK." It is enterprise implementation work long dominated by Accenture, Deloitte, and McKinsey — all of whom are now competing against the company whose models they were previously implementing.
The competitive read: Anthropic announced its $1.5 billion private equity joint venture with Blackstone, Goldman Sachs, and Hellman & Friedman in the week prior. Both frontier labs are now building not just model infrastructure but professional services arms. The race is no longer only about model quality — it is about who owns the implementation layer inside enterprise customers.
MCP crossed 97 million installs — Linux Foundation takes governance
The Model Context Protocol reached 97 million installs in March 2026, per reporting from Crescendo AI. The Linux Foundation announced it will take over MCP governance, moving the protocol from Anthropic-controlled to a neutral multi-stakeholder body.
This is a significant structural shift. When Anthropic launched MCP in November 2024, it was a proprietary standard. When IBM ships a production MCP server for watsonx.data (as announced at Think 2026 last week) and the Linux Foundation is the governance body, MCP is foundational infrastructure — as standardized as HTTP, not owned by any one company.
The 97 million install number matters for a second reason: it validates the developer adoption thesis. MCP is not a niche protocol anymore. It is the default mechanism by which AI agents connect to external tools, data sources, and APIs. Every MCP server published on AgenticMarket is now part of infrastructure that developers are accessing 97 million times.
Apple is breaking ChatGPT's 17-month Siri exclusive at WWDC June 8
Apple plans to formally unveil iOS 27 at WWDC on June 8. The most consequential AI change in iOS 27: an Extensions system that ends ChatGPT's exclusive integration with Siri.
Under iOS 27, iPhone users will be able to assign any third-party AI model — Claude, Grok, Gemini, or others — to handle Siri requests, Writing Tools, and Image Playground. Each provider gets its own distinct Siri voice. The change addresses xAI's August 2025 antitrust lawsuit alleging Apple-OpenAI exclusivity was anticompetitive.
For the MCP ecosystem, this matters in a specific way: Claude on iOS with Siri-level integration creates a new surface for MCP tool calls from a mobile context. Users asking Siri a question that triggers a Claude response can, in theory, have Claude call MCP servers as part of that response chain. The mobile MCP surface has not existed in a meaningful way until now.
Google's pre-I/O week — agentic Android, Gemini Omni video, and Chrome AI
Google I/O 2026 hasn't happened yet, but the pre-announcement week has been dense.
Android agentic features — Gemini-powered capabilities allowing AI to complete multistep tasks across apps, browse the web, fill out forms, dictate speech, and generate custom widgets through natural language prompts. Users can trigger workflows like copying grocery lists into shopping carts or completing online transactions with screen-context awareness.
Gemini Omni — an unreleased video generation system tested inside Gemini, reportedly building on Veo technology with conversational video generation and editing. Testing is underway; expect a formal I/O announcement.
Figma and GitHub import for AI Studio Build — Google is working on direct imports from both platforms, which would allow developers to pipe their design files and codebases directly into AI Studio workflows.
Gemma 4 under Apache 2.0 — Google released its most capable fully permissive open-weight model, explicitly positioned to fill the gap created by Meta's transition away from open-source flagship models.
Anthropic is reportedly acquiring Stainless for $300M+
The Information reported that Anthropic is in advanced talks to buy Stainless, the AI-powered SDK generator that builds and maintains SDKs for OpenAI, Google, Meta, and dozens of other AI companies.
If it closes, Anthropic would own the tooling used to generate SDKs for its direct competitors. Stainless generates idiomatic client libraries from OpenAPI specifications — the infrastructure layer that makes API integrations feel native in Python, TypeScript, Go, and other languages. The strategic value: owning Stainless means Anthropic controls a shared dependency that the entire AI industry relies on.
ElevenLabs hits $500M ARR. OpenAI launches Voice API.
Two audio infrastructure developments worth noting.
ElevenLabs crossed $500 million in annualized revenue — reported May 8. The voice AI company is now larger by revenue than most B2B SaaS companies that have been operating for a decade.
OpenAI Voice API launched the same week, offering programmatic access to GPT-5.5-quality voice synthesis and transcription. The timing relative to ElevenLabs' ARR milestone is probably not accidental.
Claude Code rate limits: what actually changed for developers
Since this directly affects anyone building with Claude Code, here are the precise changes in effect as of May 6, 2026:
| Plan | Previous 5-hour limit | New 5-hour limit | Peak-hour throttling |
|---|---|---|---|
| Pro ($20/mo) | Baseline | 2x baseline | Removed |
| Max ($100/mo) | Baseline | 2x baseline | Removed |
| Max ($200/mo) | Baseline | 2x baseline | Removed |
| Team ($30/seat) | Baseline | 2x baseline | Unchanged |
| Enterprise (seat) | Baseline | 2x baseline | Unchanged |
| API (Opus models) | Previous limit | Considerably increased | N/A |
From the Platform
The MCP 97 million install milestone is the number AgenticMarket's existence is built on. Every install is a developer who chose MCP as the integration path — and who is now in the market for more MCP servers to extend what their AI agents can do.
The Linux Foundation governance transition is unambiguously positive for the ecosystem. Anthropic-controlled MCP was always a governance risk for enterprise adoption — enterprises are reluctant to standardize on infrastructure owned by a single commercial entity they might compete with. Linux Foundation governance removes that risk. IBM shipped a production MCP server for watsonx.data within weeks of the governance announcement. The correlation is not a coincidence.
For server builders on AgenticMarket: the iOS 27 Siri Extensions announcement is a surface worth designing for. If Claude gains Siri-level access on iOS devices, MCP tool calls from mobile contexts become real. Tool descriptions that are clear and action-oriented matter more on mobile, where users give shorter, more conversational instructions. If you're building a server today, write your tool descriptions as if the calling context is someone saying "hey Siri, can you check if my site is up" rather than a developer typing a precise tool invocation.
Current @agenticmarket catalog: 9 production servers. All HTTPS. No external API keys required.
agenticmarket install agenticmarket/url-status
One Take
The compute race has become the product race — and the labs that lose compute access are losing users today, not at some future point.
The story behind the SpaceX deal is not the deal itself. It is the admission that Anthropic grew 80x when it planned for 10x, and that the gap manifested as a product regression users felt in real time. The Claude Code degradation post-mortem from three weeks ago documented what that looked like from the model side — silent reasoning effort changes, broken caching. The rate limits and peak-hour throttling were what it looked like from the infrastructure side. Both were symptoms of the same constraint.
This creates a new competitive dynamic that was not present in 2024. In 2024, the competition was about model capability: which lab's model scored highest on MMLU, on HumanEval, on GPQA. Developers benchmarked models and chose the best one. Infrastructure was assumed to be available; it was background.
In 2026, infrastructure availability is foreground. Claude Code's five-hour rate limit was a direct constraint on how much real work developers could complete per session. When Anthropic doubled it, developers noticed immediately — not because the model got smarter, but because they could finish more without hitting a wall. The user experience improvement came from compute, not capability.
This pattern is going to intensify. OpenAI's Codex workspace agents, which run long-horizon tasks while users are offline, are infrastructure-limited in the same way. Google's agentic Android features require sustained on-device inference that competes with cloud inference for compute budget. Every agentic product the labs are shipping has a compute floor it cannot go below without degrading the user experience.
The labs that solve compute access — through data center deals, orbital compute partnerships, TPU buildouts — will win the agentic product race not because their models are better, but because their products work reliably at scale. The SpaceX deal is Anthropic's version of that bet. The $200 billion Google Cloud commitment is the same bet, longer term.
For developers building on top of these platforms: the compute constraint is not over. It shifted. 220,000 GPUs buys months, not years, at 80x growth rates. Watch what happens to limits again in Q3.
Signal
Anthropic: Higher usage limits for Claude and a compute deal with SpaceX
Primary source for the limit changes and the compute deal structure. The section listing all of Anthropic's infrastructure commitments is the most useful part for understanding total compute strategy.
Engadget: Anthropic doubles Claude Code rate limits after SpaceX deal
Best mainstream summary of what changed for users and the Musk relationship context.
OpenAI: OpenAI launches the Deployment Company
Primary source. The list of 19 founding investment and consulting partners is buried but important — it tells you which advisory firms are now incentivized to recommend OpenAI implementations to their clients.
Let's Data Science: Anthropic Takes All of SpaceX's Colossus 1
The most technically detailed breakdown of the Colossus 1 facility, the GPU breakdown by generation, and why 80x growth against 10x planned infrastructure creates the specific product problems users experienced.
Crescendo AI: MCP 97 million installs milestone
The source for the 97 million install number. Also carries the Linux Foundation governance confirmation. Treat the exact figure as a directional estimate — the growth curve is what matters.
FAQ
What is the Model Context Protocol and why does it matter in May 2026?
The Model Context Protocol (MCP) is an open standard that lets AI agents connect to external tools, APIs, and data sources through a standardized interface. As of March 2026, MCP has crossed 97 million installs and is now governed by the Linux Foundation. Every major AI platform — Anthropic Claude, OpenAI Codex, Google Gemini, and IBM watsonx — ships MCP-compatible tooling. It is the default integration layer for AI agents connecting to external systems.
Why did Anthropic partner with SpaceX for compute?
Anthropic's Q1 2026 revenue grew 80x year over year against internal projections of 10x. The demand mismatch exhausted available GPU capacity and caused product-level degradations including rate limits and peak-hour throttling. The SpaceX Colossus 1 deal provides access to 220,000 Nvidia GPUs and 300 megawatts of power in Memphis, Tennessee, giving Anthropic infrastructure headroom while longer-term compute partnerships with Amazon, Google, and Microsoft come online.
What changed for Claude Code users after the SpaceX deal?
As of May 6, 2026, Claude Code 5-hour rate limits doubled for Pro, Max, Team, and Enterprise plans. Peak-hour throttling was removed for Pro and Max accounts. Claude Opus API rate limits increased considerably for all tiers.
What is the OpenAI Deployment Company?
The OpenAI Deployment Company, announced May 11, 2026, is a majority-owned subsidiary that embeds Forward Deployed Engineers directly inside customer organizations to implement AI systems. It launched with $4 billion from 19 investment firms and consultancies, and acquired Tomoro, an AI consulting firm, for approximately 150 engineers. It is OpenAI's direct entry into enterprise professional services, competing with Accenture, Deloitte, and McKinsey.
How do I install MCP servers for Claude Code or Cursor in one command?
Use AgenticMarket's CLI to install any MCP server directly into your IDE. After running npm install -g agenticmarket and authenticating with agenticmarket auth, run agenticmarket install username/server-name. The CLI auto-detects VS Code, Cursor, Claude Desktop, Claude Code, and other supported IDEs and writes the correct configuration automatically.
CONTEXT WINDOW publishes Monday mornings. MCP ecosystem. Developer tooling. AgenticMarket platform updates. No sponsored content.
Install MCP servers in one command: agenticmarket.dev
Publish your MCP server and earn on every call: agenticmarket.dev/creators