Anthropic’s AI Push Signals Major Shift for Channel Partners

Anthropic is rapidly scaling both the infrastructure and security footprint of its AI business, signaling a shift that could reshape how enterprises—and their channel partners—approach both compute and cyber risk.

The company this week announced a massive expansion of TPU capacity through Google and Broadcom while simultaneously launching Project Glasswing, a sweeping industry collaboration aimed at countering AI-driven software vulnerabilities. 

Together, the moves highlight how frontier AI is accelerating demand for compute while exposing new security realities.

Anthropic secures massive TPU capacity with Google and Broadcom

Anthropic this week announced a new agreement with Google and chip designer Broadcom for “multiple gigawatts” of next-generation TPU capacity, the equivalent of a small power station dedicated to running AI. 

According to a Broadcom regulatory filing spotted by The Register, that figure is specifically 3.5 gigawatts, coming online from 2027.

This builds on earlier agreements and deepens Anthropic’s ties with Google Cloud, while Broadcom continues supplying the underlying chip and networking technology.

Anthropic hits $30B run rate, doubles down on US AI infrastructure

Anthropic framed the move as necessary to keep up with explosive growth. The company said its run-rate revenue has surpassed $30 billion, up from about $9 billion at the end of 2025. 

It also noted that over 1,000 enterprise customers are now spending more than $1 million annually, a figure that has doubled in just a few months.

“This groundbreaking partnership with Google and Broadcom is a continuation of our disciplined approach to scaling infrastructure: we are building the capacity necessary to serve the exponential growth we have seen in our customer base while also enabling Claude to define the frontier of AI development,” CFO Krishna Rao said in a company statement.

Most of this new compute capacity will be based in the US, expanding Anthropic’s previously announced $50 billion commitment to US AI infrastructure.

Despite the Google and Broadcom ties, Anthropic emphasized it remains multi-cloud, running Claude across Amazon Web Services, Google Cloud, and Microsoft Azure, using a mix of Trainium chips, TPUs, and NVIDIA GPUs.

Project Glasswing brings tech giants together to tackle AI-driven cyber threats

While the hardware news is massive, the announcement of Project Glasswing is arguably more significant for channel partners and the security community. 

Anthropic has assembled an “Avengers-style” lineup of tech giants, including Microsoft, Google, NVIDIA, AWS, Broadcom, Cisco, CrowdStrike, JPMorgan Chase, Palo Alto Networks, and the Linux Foundation, to tackle a looming cybersecurity crisis.

Along with these firms, more than 40 additional organizations involved in maintaining critical infrastructure software are participating.

Mythos Preview alarms its creators with software vulnerability discovery

The project, Anthropic says, exists because of something disturbing it noticed about Mythos Preview: the model had become extraordinarily good at finding and exploiting software vulnerabilities, not because the company trained it to, but as a side effect of its raw coding ability. 

“We haven’t trained it specifically to be good at cyber,” Anthropic CEO Dario Amodei said in a Project Glasswing launch video. “We trained it to be good at code, but as a side effect of being good at code, it’s also good at cyber.”

Mythos Preview’s existence had actually leaked weeks earlier, when unpublished information about the model was found in a publicly accessible database, a rather ironic data exposure for a model Anthropic was quietly positioning as a security powerhouse. 

Dianne Penn, a head of product management at Anthropic, told The Verge that the company is “taking steps in terms of solidifying our processes” and confirmed the leak “was not related to software vulnerabilities in any way.”

Mythos model uncovers critical vulnerabilities across major systems

Over the past several weeks, Anthropic used Mythos Preview to hunt vulnerabilities across major software. The results were jarring. 

The model flagged thousands of previously unknown, high-severity security flaws across every major operating system and web browser, many of which it identified and developed exploits for entirely on its own, without human guidance, according to Anthropic’s blog post.

In internal tests, it found a 27-year-old bug in OpenBSD and a 16-year-old flaw in FFmpeg that had survived millions of previous automated tests. It even managed to autonomously chain together several vulnerabilities in the Linux kernel to take full control of a machine.

All of the above have been reported to maintainers and patched. Anthropic says it has provided cryptographic hashes for additional vulnerabilities it has not yet disclosed, and will release the specifics once fixes are in place.

Controlled rollout reflects urgency around AI-powered security risks

Unlike typical AI launches, Mythos Preview will not be made broadly available, at least for now. Anthropic says the model’s capabilities are too sensitive, citing the risk that it could be misused to accelerate attacks.

Instead, the company is taking a staged approach, first working with a closed group of partners, then sharing findings more broadly with the industry. A public report on early results is expected within 90 days. 

Partners are also expected to collaborate on new standards for secure software development in the AI era, including vulnerability disclosure, patching automation, and supply chain security.

What Anthropic’s moves mean for channel partners

For channel partners, managed service providers, and systems integrators, these moves signal a shift that is both an opportunity and a potential disruption. 

On the bright side, Anthropic is essentially handing partners a high-powered security scanner that can do the heavy lifting of finding ancient bugs in a fraction of the time. 

This allows MSPs to move from being basic troubleshooters to high-value security advisors, helping clients secure their systems against a new wave of automated threats before they even happen.

In March, Anthropic announced its formal entry into the channel with a partner network of its own, signaling the company’s push to bring Claude to more enterprises through partnerships and tech alliances.

Channel partners must adapt quickly to the AI-driven security shift

But there’s a catch: the speed of this technology could quickly outpace those who aren’t ready to pivot. If a model like Claude Mythos can autonomously find vulnerabilities, the traditional, manual approach to security audits might become a relic of the past. 

Partners will need to invest heavily in training their teams to work alongside these AI agents, shifting their focus from hunting for flaws to verifying and patching them at scale. 

It’s a sink-or-swim moment, where the winners will be those who can blend human expertise with the raw speed of these new-frontier models.