On September 13, 2026, Anthropic CEO Dario Amodei published an essay titled "We Must Pace the Frontier," calling on every frontier AI lab to slow the rate at which they push model capabilities forward, so safety and alignment work can keep up. Within days, reporting revealed that Anthropic itself had quietly signed roughly $517 billion in compute commitments over the preceding eleven months — one of the largest infrastructure bets in tech history, disclosed the same week its CEO was asking the industry to ease off the gas.
For businesses building on AI — agencies, automation shops, FlutterFlow and no-code teams, and any company leaning on frontier models to run their operations — the gap between what AI labs say and what they do is not just an interesting contradiction. It's a signal about how to plan.
Amodei's essay argued that the pace of capability gains is outrunning the industry's ability to understand and align the systems it's building. He proposed that labs voluntarily slow capability increases and floated opening Anthropic's models to permanent, independent third-party safety evaluation. It wasn't a call to halt AI development — it was a call to decouple "faster" from "better," and to buy more room for alignment research before the next leap in capability.
The reaction was fast. Reporting indicates OpenAI's Sam Altman, Google DeepMind's Demis Hassabis, and xAI's Elon Musk each signaled agreement with the broad thrust of the essay within about a day. That kind of quick, cross-company nod on a safety framing is unusual in an industry that otherwise competes ferociously on capability and speed.
Not everyone was satisfied with "pace, don't stop." Senator Bernie Sanders responded by introducing legislation that would go much further: an outright ban on developing artificial superintelligence, with steep penalties — reportedly including prison time — for violators. His argument, paraphrased: when you're racing toward a cliff, easing off the accelerator isn't the same as hitting the brakes.
Here's where it gets uncomfortable. The same week Amodei was urging restraint, The Information reported that Anthropic had committed roughly $517 billion across more than a dozen compute providers — including AWS, Google, and SpaceX — covering nearly 15 gigawatts of capacity through 2026. That figure is nearly triple the $180 billion in commitments that had been disclosed just months earlier, and most of these deals are structured as take-or-pay contracts: Anthropic is on the hook for the spend whether or not it ends up using all that capacity.
To be fair, "pacing capability increases" and "securing compute for the next decade" aren't strictly the same thing — a lab can argue it's building the runway for the future while still choosing to release capability more deliberately. But the optics are hard to miss: a company asking the industry to slow down was, in the same breath, locking in infrastructure at a scale that only makes sense if it expects to keep scaling aggressively.
If you're running an agency, an automation practice, or a no-code shop that depends on frontier models, three things fall out of this story worth sitting with:
Public safety commitments and internal roadmaps are two different documents. Vendor messaging about caution, alignment, and responsible deployment is real and often sincere — but it's not a reliable predictor of how fast the underlying models you depend on will change. Build your workflows and prompts to be resilient to model updates, not dependent on a particular pace of releases staying calm. Compute economics are getting locked in at a scale that will show up in your pricing. When a single lab commits over half a trillion dollars to compute, that cost eventually flows somewhere — either subsidized by investor capital chasing the next funding round, or passed through in API pricing down the line. Businesses with AI baked into their cost structure should watch compute and pricing trends as closely as they watch model capability announcements. The regulatory conversation is no longer theoretical. A sitting U.S. senator proposing a ban on superintelligence development, paired with Anthropic's own policy chief publicly stating this week that AI companies "can't be checking our own homework," suggests binding rules are getting closer, not further away. Agencies and automation businesses that treat AI governance as someone else's problem — a lab's problem, a regulator's future problem — are the ones most likely to be caught flat-footed by a compliance requirement that shows up with little warning.
Our cutting-edge features simplify collaboration and creativity, making your workflow intuitive and efficient. Transform your vision into reality effortlessly with Hadidiz Flow.
