AI Slowdown Calls Collide With a Shipping Spree
3 min readTwo things happened in the same week. Senior figures at OpenAI and Anthropic pressed publicly for easing off the frontier AI race. Days later, reports had GPT-6 Sol and Claude Opus 5.2 both moving through staged testing, and Huawei published a 2035 forecast that assumes no brakes at all.
A Pacing Argument That Keeps Getting Louder
The slowdown debate has been building for weeks. Anthropic’s leadership called for cooling the race, Beijing dismissed the idea as a Cold War tactic aimed at Chinese developers, and Nvidia’s Jensen Huang rejected the premise outright at Dreamforce. The concern behind it is specific rather than vague: agents that operate hardware on their own, acquire resources, and resist supervision. Recent incident disclosures from frontier labs gave that concern documented examples instead of hypotheticals.
What the Roadmaps Actually Say
Huawei’s new report, Intelligent World 2035: Turning Vision into Action, does not argue against slowing down. It simply never raises the question. The company projects that autonomous agents will account for more than 90 percent of global AI token traffic by 2035, building on an earlier forecast of 900 billion active agents worldwide, roughly a hundred times the projected human population. Security and privacy protection for autonomous agents appears in the report as one of ten enabling technologies, an engineering problem to solve along the way rather than a reason to wait. Guo Ping, chairman of Huawei’s supervisory board, told staff this week that AI is the company’s “biggest opportunity” and that he wants its computing infrastructure to become China’s answer to Nvidia, according to Reuters.
The labs calling for restraint were busy too. Chinese outlet 36Kr reported that GPT-6 Sol and Opus 5.2 entered limited staged testing within days of the weekend’s brake-pumping, with Opus 5.2 surfacing first through Claude Code. OpenAI separately confirmed that the GPT-5.5 series retires on October 14, six months after launch. Anthropic is preparing an IPO and can ill afford a quiet quarter.
Why It Matters
Nobody here is being dishonest, exactly. An executive can believe the industry as a whole should move more carefully and still believe their own lab should lead, since unilateral restraint mostly hands the frontier to someone less cautious. But that reasoning only holds if the coordination it depends on actually arrives, and right now there is no mechanism, no regulator, and no counterparty willing to sign. Huawei’s report is the clearest evidence of the gap: a major player mapping out a decade of agent deployment without treating pacing as an open question at all.
Watch what gets shipped, not what gets said. If the next few weeks bring a full GPT-6 family launch and a general release of Opus 5.2, the slowdown conversation will have produced a great deal of agreement and no observable change in speed.