AI’s HPC Era: When Access Becomes the Product
Anthropic just launched Claude Fable 5 and Mythos 5, and the way they’re handling access reminded me of something from my research days.
Back in college, our lab used TACC (Texas Advanced Computing Center) for high-performance computing. You couldn’t simply sign up and start running large-scale workloads. Labs had to justify their use case, secure funding, and gain access through their institution.
Not because the technology was secret. But because it was powerful and needed to be used responsibly.
Anthropic seems to be taking a similar approach with frontier AI.
Fable 5 and Mythos 5 are the same underlying model. The difference isn’t intelligence — it’s access.
Fable 5 is publicly available, but sensitive cybersecurity and biology-related requests are automatically routed to a less capable model. Mythos 5 removes many of those restrictions, but access is limited to vetted government, research, and critical infrastructure organizations through Project Glasswing.
This got me thinking: AI may be entering its HPC era.
For years, we’ve assumed that every generation of AI would eventually become available to everyone. Now we’re seeing a different possibility: some AI systems may become so capable that access itself becomes a managed resource.
In the HPC world, researchers applied for access to supercomputers. In the AI world, organizations may apply for access to frontier models.
The reasons are different. TACC was primarily about scarce compute. Project Glasswing is about safety and misuse risk. But the structure feels surprisingly familiar.
What’s changing isn’t just model capability. It’s the idea that the most advanced AI systems may increasingly be treated less like software products and more like critical infrastructure.
If that’s true, the future of AI may be shaped not only by who can build the most powerful models, but by who gets access to them.