China-based DeepSeek made waves in the generative artificial intelligence (GenAI) sector earlier this year with the release of its R1 model, a budget-friendly yet high-performing alternative that directly challenged OpenAI and other industry giants.
Since late 2022, a few AI assistants, such as OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini, had dominated the market due to massive investments in engineers, data centers, and advanced AI chips.
However, DeepSeek’s R1 model, reportedly developed for just $6 million, disrupted the industry. Built using less advanced chips, its release ignited widespread debate regarding cost efficiency in AI development.
Read More: Liang Wenfeng, the founder and CEO of DeepSeek
Although some experts speculated that DeepSeek’s actual expenses were higher than stated, the model’s introduction intensified discussions on the increasing commoditization of GenAI assistants, propelled by rapid innovation and shifting market trends.
“The first company to train models must invest significant resources,” explained Angelo Zino, senior equity analyst at CFRA. “But second movers can achieve similar results more affordably and efficiently.”
At the HumanX AI conference in Las Vegas this week, Thomas Wolf, co-founder of Hugging Face, highlighted the declining costs of launching GenAI models, suggesting that users now had more model choices than before. “We’re moving towards a multi-model landscape, which is beneficial,” Wolf said, referencing the lukewarm reception of ChatGPT’s latest iteration.
In contrast, Kevin Weil, OpenAI’s chief product officer, dismissed the notion that all AI models were now equivalent. “That’s simply not true,” Weil stated. “While our lead time may have shortened from a year to three to six months, that still holds immense value.” He emphasized OpenAI’s advantage in refining its models through extensive user data, leveraging a vast base of 400 million users.
“OpenAI enjoys the same mental association as Google—it’s the first thing people think of,” said Fen Zhao, research director at Alpha Edison.
Jeff Seibert, CEO of AI startup Digits, acknowledged OpenAI’s ongoing edge in specialized applications but predicted that the performance gap among top competitors would eventually shrink. “For advanced use cases, OpenAI will maintain advantages,” he noted. “But for general applications, the choice of model won’t matter as much.” Seibert advised businesses to structure their technology to remain adaptable as AI models evolve.
Advancements in chip efficiency and optimization techniques have significantly lowered the costs of developing large language models (LLMs) such as ChatGPT and Gemini. Some models have also embraced open-source frameworks, accelerating innovation by allowing free modifications and enhancements.
Zino suggested that the valuations of closed-model startups like OpenAI and Anthropic might have peaked as their first-mover advantage weakened. In February, Japanese investment firm SoftBank injected $40 billion into OpenAI, nearly doubling its valuation to $300 billion compared to the previous year.
“If OpenAI is burning a billion dollars a month, as I suspect, it must continuously raise funds,” said Jai Das, managing director at Sapphire Ventures. “I struggle to see how their revenue will surpass their expenditure.”
Meanwhile, in early March, Anthropic secured $3.5 billion in funding, bringing its valuation to $61.5 billion.
