Anthropic April 10 2026 Announcement: Custom Chips, CoreWeave Deal, and Mythos Fallout
Anthropic April 10, 2026 Announcement: Custom Chips, CoreWeave Deal, and Mythos Fallout
April 10, 2026 was one of the busiest single days in Anthropic's history. The company confirmed it is exploring custom AI chip design, signed a multi-year cloud infrastructure deal with CoreWeave, and continued managing the fallout from its Claude Mythos Preview release earlier that week. If you are searching for what Anthropic announced on April 10, 2026, this post covers every major development with full context on what came before it.
Key Takeaways
- Anthropic is exploring building custom AI chips to reduce dependency on third-party silicon, driven by Claude demand surging past a $30 billion annual run rate
- CoreWeave signed a multi-year deal with Anthropic to provide GPU cloud infrastructure for Claude workloads, sending CW stock up 10%
- Claude Mythos Preview, released days earlier as part of Project Glasswing, triggered cybersecurity risk warnings from Bloomberg and government officials
- The week of April 7 to 10 saw Anthropic announce a Google/Broadcom TPU deal, Managed Agents, Claude Cowork GA, and a $380B valuation
Timeline: Anthropic's Full Week of Announcements (April 7 to 10, 2026)
| Date | Announcement | Details |
|---|---|---|
| April 7 | Project Glasswing | Cybersecurity consortium with AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks |
| April 7 | Claude Mythos Preview | Gated frontier model for defensive security research; found thousands of unknown vulnerabilities in major OS and browser codebases |
| April 7 | Google/Broadcom TPU Deal | 3.5 gigawatts of next-gen TPU capacity secured starting 2027 |
| April 7 | $30B Revenue Run Rate | Tripled from $9B at end of 2025; over 1,000 enterprise customers each spending $1M+ annually |
| April 7 | $380B Valuation | Series G funding round closed |
| April 8 | Claude Managed Agents | Public beta with sandboxing, built-in tools, and the ant CLI |
| April 8 | Service Outage | Second consecutive day of connectivity issues affecting Claude users |
| April 9 | Claude Cowork GA | Generally available on macOS and Windows in Claude Desktop with analytics API and role-based access |
| April 10 | Custom Chip Exploration | Reuters reported Anthropic is in preliminary stages of evaluating in-house chip design |
| April 10 | CoreWeave Multi-Year Deal | GPU cloud deal for Claude production workloads on Nvidia hardware in U.S. data centers |
Custom AI Chip Exploration
The biggest headline on April 10 came from Reuters: Anthropic is weighing building its own AI chips. The move is driven by two factors. First, the global shortage of AI accelerators continues to constrain scaling. Second, Claude's demand has grown so fast (from a $9B run rate in late 2025 to $30B by April 2026) that vertically integrating silicon makes economic sense.
Currently, Claude runs on a multi-architecture stack:
- Amazon AWS Trainium chips via the Amazon Bedrock partnership
- Google TPUs via the newly expanded Broadcom deal (3.5GW starting 2027)
- NVIDIA GPUs via CoreWeave and other cloud providers
The custom chip effort is in early stages. Anthropic has not committed to a specific design, and there is no dedicated chip team yet. But the direction mirrors what Amazon (Trainium/Inferentia), Google (TPU), and Meta (MTIA) have already done: designing silicon optimized for their own model architectures to reduce per-inference cost and improve supply chain control.
CoreWeave Multi-Year Cloud Deal
Also on April 10, CoreWeave announced a multi-year agreement with Anthropic to power Claude workloads. Key details:
- Anthropic will use Nvidia chips deployed in CoreWeave's U.S. data centers
- Compute begins coming online later in 2026, with phased rollout
- CoreWeave did not disclose the financial terms
- CW stock closed up 10% on the news
- With this deal, nine of the ten largest AI model providers now use CoreWeave's platform
The timing is notable. CoreWeave announced a separate $21 billion expansion with Meta just one day earlier. For Anthropic, diversifying beyond AWS and Google TPUs gives them more leverage and redundancy in the increasingly competitive market for GPU compute.
Claude Mythos and the Cybersecurity Debate
While April 10 was dominated by infrastructure news, the ripple effects of the April 7 announcements continued. Bloomberg published a detailed analysis titled "Mythos: Why Anthropic's New AI Has Officials Worried," and Fortune ran its own piece arguing that the era of AI-driven hacking was already here regardless of Anthropic's safeguards.
The core tension: Claude Mythos Preview discovered thousands of previously unknown vulnerabilities in every major operating system and web browser. Anthropic chose to restrict the model to 40+ organizations (including Microsoft, Apple, CrowdStrike, and AWS) rather than release it publicly. Critics argued this creates a two-tier security landscape. Defenders argued a public release would hand offensive capabilities to attackers before patches could be developed.
Project Glasswing, the consortium built around Mythos, includes eleven organizations:
- AWS
- Apple
- Broadcom
- Cisco
- CrowdStrike
- JPMorganChase
- Linux Foundation
- Microsoft
- NVIDIA
- Palo Alto Networks
What This Means for Developers and AI Teams
The April 10 announcements signal three things for teams building on Claude:
More compute diversity. Between AWS Trainium, Google TPUs, NVIDIA GPUs via CoreWeave, and eventually custom Anthropic silicon, the infrastructure options for running Claude are expanding. This should mean better availability and potentially lower costs as competition increases.
Managed Agents are production-ready. The April 8 launch of Claude Managed Agents in public beta, combined with the ant CLI, means teams can now deploy autonomous agents with proper sandboxing and state management without building that infrastructure themselves.
Security tooling is becoming a first-class use case. Project Glasswing and Mythos Preview demonstrate that Anthropic views defensive cybersecurity as a primary application of frontier models. Teams building security products should watch the Glasswing consortium for early access opportunities.
How Anthropic's April 2026 Week Compares to Previous Launch Weeks
For context, here is how the density of Anthropic's April 7 to 10 announcement week compares to previous major release periods:
| Period | Key Releases | Revenue Context | |---|---|---| | March 2025 | Claude 3.7 Sonnet, Claude Code | ~$2B run rate | | October 2025 | Claude 4.0 family, Computer Use GA | ~$6B run rate | | January 2026 | Claude 4.5 Opus, Extended Thinking | ~$15B run rate | | April 7-10, 2026 | Mythos, Glasswing, Managed Agents, Cowork, CoreWeave deal, custom chips | $30B run rate |
The pace of announcements correlates directly with revenue growth. As Anthropic's customer base and compute needs have scaled, the cadence of product and infrastructure announcements has increased to match.
Sources
- Reuters: Anthropic weighs building its own AI chips (via CNBC)
- CoreWeave Announces Multi-Year Agreement With Anthropic
- Bloomberg: Anthropic's New AI Model Prompts Cyber Risk Warning
- Fortune: Anthropic's Mythos is a wake-up call
- SiliconANGLE: Anthropic tries to keep its new AI model away from cyberattackers
- TechCrunch: Anthropic ups compute deal with Google and Broadcom
- The Hill: Anthropic restricts release of new AI model
- Benzinga: Anthropic Eyes Custom AI Chips