The debate over whether artificial intelligence development should be slowed for safety reasons has intensified, with Anthropic CEO Dario Amodei calling for a more cautious pace at the frontier of model training. His intervention has reopened a key question for markets: will AI investment and expansion actually slow in response?
CNBC’s answer is largely no. The outlet argues that a meaningful slowdown in AI spending remains unlikely in the near term, mainly because the geopolitical contest between the U.S. and China continues to push both sides to move fast.
In his essay, “We Must Pace the Frontier,” Amodei laid out a three-step framework. First, he called for third-party safety evaluators to be embedded in each major lab. He then argued for democratic countries to align on the issue and eventually coordinate with authoritarian governments on safety standards and “limits on the rate of unchecked AI progress.” He acknowledged that some forms of coordination would be legally difficult and would require government backing.
That stance drew public support from OpenAI’s Sam Altman and SpaceX’s Elon Musk. OpenAI also published six examples of “unexpected or concerning model behavior,” saying the industry has not yet solved key safety problems well enough to keep scaling at maximum speed for much longer. Musk said “Dario is right” and later suggested that leading American and Chinese companies should run safety tests on each other’s models.
But prominent figures have pushed back. Nvidia CEO Jensen Huang said new laws or antitrust waivers are “completely unnecessary” for companies to do the engineering work required before releasing products. In his view, safety and testing are engineering challenges. Technology investor and former White House AI and crypto czar David Sacks accused Amodei of trying to raise the regulatory bar so high that it becomes a barrier to entry. He argued that companies concerned about their own unreleased models can slow down on their own, without presenting that choice as a universal requirement.
Meta Platforms CEO Mark Zuckerberg also questioned the need for broad coordination, saying market forces already reward safer behavior. He pointed to Meta’s decision to delay the release of its Muse model by several months in order to focus on safety and security.
To cut through the debate, CNBC turns to game theory and the classic prisoner’s dilemma. In that framework, the most beneficial result for both sides would be coordinated restraint. Applied to AI, that would mean the U.S. and China both agreeing to slow development in a credible and enforceable way, producing a safer outcome for all.
However, the outlet argues that such cooperation is too difficult to enforce in practice. That makes the most likely outcome what game theory describes as a Nash equilibrium: not the best collective result, but the one rational actors are most likely to choose. The U.S. is unlikely to slow if it fears China may continue advancing at full speed and narrow or erase America’s lead. China faces the same incentive in reverse.
Amodei himself has acknowledged how difficult global coordination with China would be. He wrote that any agreement would need either ironclad verifiability or limits narrow enough that defection would not create an existential military risk. CNBC sees that as further evidence that a binding global slowdown would be extremely hard to achieve.
Palantir CEO Alex Karp added on CNBC that leading American AI labs are still underestimating the geopolitical consequences of slowing down. He said competition with China remains central and suggested that expectations for a near-term breakthrough between U.S. President Donald Trump and Chinese President Xi Jinping should remain modest.
CNBC’s bottom line is that investors should not base their AI stock strategy solely on calls for slower development. Without enforceable international cooperation, the U.S. and China are unlikely to ease off the pace, which in turn means a significant slowdown in AI development and spending still does not look like the market’s most probable outcome.
