AI speed control theory loses power and competition accelerates... Antropic is also aiming for an IPO with a new model ahead of time

(Seoul = Yonhap News) Reporter Hwang Cheol-hwan = The 'AI speed control theory' raised by heads of global artificial intelligence (AI) companies such as Antropic, OpenAI, SpaceX, and Google DeepMind appears to be rapidly losing power.

This is because of the question of whether it is possible to slow down the pace in a ‘winner-takes-all’ structure in which a very small number of winners take all the fruits. Some people even seem to be stepping on the accelerator pedal.

According to the financial investment industry on the 25th, Reuters, citing multiple sources familiar with the situation, recently reported that Antropic, which is scheduled to go public (IPO) in November of this year, is considering a plan to accelerate the release of a new AI model.

It is said that as competitor OpenAI showed a move to regain the industry lead with the new model 'GPT-6 Astra' released earlier this month, it was fumbling with the 'IPO after unveiling the new model' card.

If true, this is the exact opposite of the remarks made by Antropic CEO Dario Amodei, who proactively advocated for AI speed control.

In an article posted on his blog on the 12th, CEO Amodei warned that the speed of AI development could overwhelm human understanding and control capabilities, saying, “We must slow down the speed of performance improvement of AI models.”

Open AI CEO Sam Altman, DeepMind founder Demis Hassabis, and Space

Park Jae-hwan, a researcher at Eugene Investment & Securities, pointed out, "In addition, Musk also mentioned that he recently completed pre-training for 'Grock 4.8' with 2.5 trillion parameters, and Altman is predicting the launch of a number of new products around the 29th of this month."

He pointed out, “Despite the theory of slowdown, model competition is accelerating,” adding, “It has been confirmed in just one week that voluntary speed control is virtually impossible to establish in the current situation where model competition is intensifying.”

Jang Da-hyun, a researcher at Mirae Asset Securities, also said, "A reduction in compute has not yet been seen in the actual investment behavior of companies claiming speed control. Antropic has secured up to 5GW of next-generation tensor processing unit (TPU) capacity with Amazon and 5GW with Google and Broadcom, and recently secured additional capacity of more than 300MW from SpaceX's Colossus 1 and more than 220,000 NVIDIA graphics processing units (GPUs)." reported.

Recently, Meta, which had been relatively lagging in the AI ​​competition, is causing a sensation with its new AI agent 'Muse' and is rapidly narrowing the gap with the leaders.

In relation to this, some say that AI companies are falling into the 'prisoner's dilemma' found in game theory in economics.

Kim Ye-seul, a researcher at IBK Investment & Securities, pointed out, "If everyone adjusts the speed, AI safety can be increased and the risk of control failure can be lowered, but if only one company adjusts the speed, there is a greater possibility that competitors who have continued development will dominate the market share."

He said, “In the current AI market, losses from falling behind in the competition are recognized as a greater risk,” and “Unless this structure changes, AI competition is expected to unfold in a direction where each other’s investments encourage investment again.”