Elon Musk Quietly Follows DeepSeek! Grok's Honest Response Wins Praise from Netizens

A netizen discovered that Elon Musk himself quietly followed the official account of Chinese AI startup DeepSeek on the social platform X this past Saturday (1st). This action coincides closely with DeepSeek's recent product developments. Just the day before, on Friday (31st), DeepSeek announced via its API documentation changelog that the official version API of its next-generation model, DeepSeek-V4-Flash, is now open for public beta testing—once again igniting heated discussions in the AI community with its highly competitive pricing strategy.

In response to Musk's move, a user directly asked Grok, the AI model under Musk's company, for its opinion on the 'boss following a competitor.'

Grok's response was notably candid, calling the action both reasonable and intriguing. It pointed out that DeepSeek has achieved extreme optimization in efficiency in recent years, delivering model performance that approaches—or in some cases surpasses—top Western standards, all while using less computing power and at lower cost. For xAI and Tesla (TSLA-US), such a competitor is not one to be ignored, but rather a key benchmark that must be closely monitored.

In other words, following an account is, to some extent, the most direct form of 'market watching.'

Grok further analyzed that Musk had previously expressed skepticism toward DeepSeek, particularly questioning its claims about training costs and chip usage. Now, by proactively following DeepSeek's account, Musk is signaling—at the very least—that he acknowledges DeepSeek has reached a stage where it can no longer be dismissed.

Grok emphasized that the AI field is rapidly evolving, and rather than working in isolation, it's better to incorporate competitors' movements into one's own information landscape.

This response earned widespread praise from netizens, with many calling it 'interesting.'

Looking back at Musk's shifting stance toward DeepSeek, it's nothing short of a dramatic arc. In early 2025, the launch of DeepSeek R1 shocked Silicon Valley, but Musk remained skeptical.

When the founder of Scale AI claimed that DeepSeek secretly possessed 50,000 NVIDIA H100 chips despite not disclosing it, Musk immediately chimed in on the comment thread with 'obviously so,' scoffing at the widely circulated narrative of low-cost development. At one point, he was even dubbed DeepSeek's 'number one hater' by netizens.

However, Musk began quietly shifting his stance starting in February this year. When Anthropic accused Chinese AI firms like DeepSeek of 'distilling' its Claude model technology, Musk was the first to push back with sarcasm, questioning whether Anthropic's own model wasn't also 'stolen' from human engineers.

By May, Musk's attitude had further evolved into open praise, acknowledging that DeepSeek had built a model rivaling global top-tier performance using fewer than 2,000 NVIDIA chips and a budget of approximately $5.5 million—an astonishing feat of efficiency.

In July, he repeatedly expressed optimism about China's AI development potential. He retweeted a post on X noting that Microsoft was testing models like DeepSeek as potential replacements for OpenAI. Later, in an exclusive interview with the UK's The Economist, he stated outright that China possesses strong capabilities in robotics and AI, and could very well take the lead at a certain point in the future.

Notably, the newly launched public beta version of DeepSeek-V4-Flash maintains identical specifications to its predecessor in terms of model architecture, total parameter count, and active parameter count. It uses a MoE (Mixture of Experts) architecture with 284 billion total parameters and 13 billion active parameters—exactly the same as the previous generation. The significant improvement in the model's agent capabilities was achieved solely through re-optimization in post-training calibration.

Additionally, there have been market rumors that Tesla's Chinese in-car system will integrate DeepSeek. According to Tesla China's official website's 'In-Car Voice Assistant Terms of Use,' the system will adopt a dual-model collaborative operation scheme using both ByteDance's Doubao and DeepSeek, with both models interfaced via Volcano Engine.