Recently, several highly anticipated AI flagship models have failed to launch as scheduled, a trend that is increasingly becoming more widespread.

According to reports, last summer, Elon Musk, the billionaire founder and CEO of AI company xAI, announced that the company's next major AI model, Grok 3, would be released by the end of 2024. Grok 3 was intended to be a response to OpenAI's GPT-4 and Google's Gemini, capable of analyzing images, answering questions, and providing various functionalities for Musk's social network, X.

In a post on X in July last year, Musk mentioned that after training on a cluster of 100,000 H100 GPUs, Grok 3 should have "very special performance" by the end of the year. In another post in mid-December, he stated, "Grok 3 will be a significant leap forward."

However, as of January 2nd, Grok 3 has not been released, and there are no signs of its imminent launch.

In fact, AI insider Thibault Bracher discovered that some code on the xAI website suggests an intermediate version, "Grok 2.5," might be released first.

This is not the first time Musk has set ambitious goals and failed to meet them. His statements about product release dates are often seen as unrealistic, a fact well-known to many.

In August last year, during an interview with podcaster Lex Fridman, Musk said that if they were "lucky," Grok 3 "hopes to be launched in 2024."

However, the absence of Grok 3 is noteworthy as it reflects a current trend.

Last year, AI startup Anthropic failed to release its top model, Claude 3 Opus, on schedule. Months after announcing that the next-generation model, Claude 3.5 Opus, would be released by the end of 2024, Anthropic removed all mentions of the model from its developer documentation. (Reports suggest that Anthropic did complete the training of Claude 3.5 Opus last year but decided that releasing it was not economically viable.)

It is also reported that the flagship models of Google and OpenAI have faced setbacks in recent months.

This may indicate that the current AI scaling laws—methods used by companies to enhance model capabilities—are reaching their limits. Until recently, it was still possible to achieve significant performance improvements by using large amounts of computational power and increasingly large datasets. However, the gains from each generation of models have started to diminish, forcing companies to seek alternative technologies.

Musk hinted at this in his interview with Fridman. When asked if he hoped Grok 3 would become the most advanced model, he replied, "I hope so. That's the goal. We might fail. It's just an expectation."

There may be other reasons behind the delay of Grok 3. For example, the xAI team is significantly smaller than many of its competitors. Nevertheless, the delayed release further demonstrates the challenges facing traditional AI training methods.