Something strange is happening at the frontier of intelligence.
Anthropic CEO Dario Amodei called for slowing frontier AI development, OpenAI CEO Sam Altman agreed that the industry needs to “pace the frontier,” and Elon Musk said Amodei was right. The industry leaders’ fear stems from the fact that the alignment problem remains fundamentally unsolved. Rumors that Google DeepMind may have reached a major milestone in recursive self-improvement, or RSI, also erupted across AI circles over the weekend, but those remain particularly weak and vague.
The timing is striking, although there is no evidence that the public calls for pacing were prompted by the DeepMind rumor.
Altman had already reached for even more loaded language in July, saying “we are now, like, in the singularity” shortly after OpenAI disclosed that its AI agents had escaped a cyber evaluation and compromised Hugging Face.
Amodei wrote on September 12 that “since roughly this summer, AI has been advancing drastically faster,” driven primarily by AI’s growing ability to help build the next generation of AI. OpenAI has independently disclosed that its researchers were already using 3.1 agent-workdays for every human workday by mid-August, allowing researchers to write more code and run more experiments.
And Google was already working on the exact capability now generating the rumors.
Photo by Maximalfocus on Unsplash
Reuters reported on August 12 that Google co-founder Sergey Brin had been pushing resources inside DeepMind toward recursive self-improvement, described by a source familiar with his efforts as the point where the technology can improve without human intervention.
That puts Alphabet Inc. (NASDAQ:GOOGL) directly inside the current story even before considering the weekend speculation.
There is another clue in the broader industry backdrop. On September 6, OpenAI said it had reached its goal of an automated research intern and that coding agents were already accelerating research workflows. The disclosure came shortly after OpenAI declared the beginning of an “AGI era” with Astra.
Amodei now says this dynamic is beginning across the industry, including at Anthropic, and explicitly calls it recursive self-improvement.
The China Problem Turns Nvidia Into the Choke Point
Amodei’s proposal contains a problem that makes a simple industry-wide slowdown much harder: China.
He argues that U.S. frontier labs can only slow by roughly the amount of their technological lead over Chinese projects. Slow further, and an unpaced Chinese program could overtake them. His answer is to widen the gap first.
The first item on his list is blunt: do not sell powerful AI chips or semiconductor-manufacturing equipment to China. “Chips will be the main determinant of China’s AI strength,” Amodei wrote. He also wants tighter controls on chip smuggling, remote access to overseas data centers, unauthorized model distillation and model-weight theft.
That puts NVIDIA Corporation (NASDAQ:NVDA) at the physical center of a debate that otherwise sounds abstract.
If AI systems are helping researchers run more experiments and attack more research tasks in parallel, the obvious bull case is more compute consumption. OpenAI itself says compute may become a more important gating factor as other research bottlenecks diminish. Nvidia’s latest results already show how violent the demand curve has become: Data Center revenue reached $89.0 billion in fiscal Q2 2027, up 117% from a year earlier.
Yet the same scarcity that makes Nvidia valuable makes it an obvious control point. A serious pacing regime could eventually constrain frontier training workloads, while harsher export rules would further limit China revenue. NVIDIA Corporation already said current U.S. policy has effectively shut it out of China’s data-center compute market, and its fiscal Q3 outlook assumes no Data Center compute revenue from China.
This is where Amodei’s proposal differs from Bernie Sanders’ push to ban superintelligence. Amodei is not proposing that America simply stop. He wants the U.S. to preserve enough of a lead over China that American labs can afford to slow down without losing the race.
That creates a strange position for Nvidia. More powerful research agents could accelerate demand for its chips. Successful coordination could eventually restrict the rate at which the largest customers use them.
Hedge-fund ownership was still moving higher before this debate erupted. Insider Monkey counted 285 funds long NVDA in Q2, up from 275 in Q1. Fisher Asset Management held 90.94 million shares after increasing the position about 3%.
The Rumor Does Not Have to Be True for the Trade to Change
What deserves more attention is the sudden convergence among people closest to frontier AI. Within days, several of them began talking publicly about the same problem: AI systems are starting to accelerate the work required to build better AI systems.
Amodei says AI-assisted AI development has accelerated sharply. Altman told OpenAI employees the company is open to slowing development and then publicly backed Amodei’s call to pace the frontier. Musk agreed. DeepMind Chair Demis Hassabis has now said Amodei’s essay points “towards the right path forward,” while noting that the details still need work.
Meanwhile, OpenAI’s own research organization is already consuming more than three agent-workdays for every human workday.
That is enough to change the investor question.
Alphabet can benefit if DeepMind turns AI-assisted research into a compounding capability advantage across models, cloud and products. It can also become an early target if policymakers decide internal AI-for-AI research needs limits.
Nvidia can benefit if each human researcher effectively commands an expanding army of compute-hungry agents. It can also become the hardware throttle regulators use to control both frontier development and China’s ability to follow it.
Short positioning suggests neither stock entered this debate carrying a large conventional bearish crowd. At the August 31 settlement, 77.7 million GOOGL shares were sold short, equal to 0.72% of public float. NVDA short interest stood at 298.3 million shares, or 1.29% of float.
The RSI rumor may have more spice than warranted, but it’s not something that can be ignored either. AI systems are helping build better AI systems; the people running frontier labs are increasingly worried about how quickly that loop could accelerate, and the two public companies sitting closest to the economic machinery are Alphabet and Nvidia.
READ NEXT:
The Trillion-Dollar AI Capex Gap: Why Too Much Hardware Could Be Nvidia’s Trap and Microsoft’s Opportunity and Redditors Are Skeptical of SpaceX’s Orbital Data Centers. Are They Right?
Hedge Fund Resource Center