Is OpenAI doomed? Open-source models may crush it, warns expert

ZDNET

OpenAI, which is losing billions of dollars, may struggle in the new landscape of top generative artificial intelligence programs, according to a noted AI scholar.

"The issue isn't really whose [AI] model is 1% better; I think they're all very good," said AI scholar and entrepreneur Kai-Fu Lee in an interview on Bloomberg Television Thursday morning. "The issue is: Is OpenAI's [business] model even sustainable?"

Harder to compete

Lee said as AI "foundation," or "pioneer" models, such as OpenAI's GPT, become commodities, it becomes harder for OpenAI to compete with the cheaper economics presented by DeepSeek AI.

Also: What is DeepSeek AI? Is it safe? Here's everything you need to know

"You're spending $7 billion or $8 billion a year, making a massive loss, and here you have a competitor coming in with an open-source model that's for free," said Lee, comparing OpenAI's financials with those of DeepSeek AI.

"If you think about OpenAI's cost of $7 billion of operating costs, in 2024, DeepSeek probably operated with 2% of the operating expense."

Also: DeepSeek's new open-source AI model can outperform o1 for a fraction of the cost

"And that company also is infinitely lasting," Lee said of DeepSeek, "because this founder has enough money to fund it at the current level, and has reduced the cost of computing by a factor of five to ten; with that kind of a competitor, I think Sam Altman is probably not sleeping very well."

Bloomberg

The cost has not yet been justified

DeepSeek AI, which caused a crash in the stock market in January of this year, is the creation of a unit of a Chinese hedge fund run by Liang Wenfeng. The company has claimed its flagship AI model, R1, can do what OpenAI's does for a tenth of the cost.

Lee, who served as founding director of Microsoft Research Asia before working at Google and Apple, founded his current company, Sinovation Ventures, to fund startups such as 01.AI.

Also: Are ChatGPT Plus or Pro worth it? Here's how they compare to the free version

Lee's startup company makes a generative AI search engine called BeaGo. Lee said that the company is now building an "operating system" of sorts to run on top of AI models such as DeepSeek's R1, specifically to serve China.

That effort is in alignment with China's big push to make DeepSeek AI's open-source software, and other models such as Alibaba's Qwen, pervasive throughout China's economy.

"The Chinese government's new big direction is called new quality productivity," said Lee, which involves, "using high tech to create productivity, efficiency, and really turn all the traditional industries into more profitable and competitive ones."

Also: Google Gemini just made two of its best features available for free

Lee's remarks about OpenAI are a continuation of an economic warning he has been making for some time now.

Lee has previously pointed out that the economics of the AI industry are unsustainable for many firms. Companies are investing massively in data centers, and chips from Nvidia, to build ever-larger models. But the cost has not yet been justified by the payoff, he has said.

Open-source will be the winner

Thursday, Lee expanded on those thoughts, telling the program that "the underlying foundational model is commoditized: it costs a ton of money, and it's hard to monetize."

"Clearly, the pre-training of a giant model has consolidated," added Lee. "It's becoming clear that open-source will be the winner" because it's cheaper to make and to operate, he said, as exemplified by DeepSeek AI's open-source R1 versus closed-source models such as those of OpenAI and Anthropic.

Also: OpenAI wants to trade gov't access to AI models for fewer regulations

Lee indicated OpenAI and Anthropic are in denial about that economic reality. Both "build their businesses thinking they can build a closed-model that's better than anyone else, and they got shocked" by DeepSeek AI's economics, he said.

The real indication of the AI economics, said Lee, is that most venture capitalists won't fund new AI foundation-model businesses. Instead, startup investors are throwing money at follow-on applications that ride on top of the foundation, or "pioneer," models, such as his own O1.ai. That startup capital for AI apps "is going gangbusters," said Lee.

Also: Open AI, Anthropic invite US scientists to experiment with frontier models

Asked to predict the future of the foundation models, Lee did not call for OpenAI's imminent demise. "I think China will have three winners, and the US might have four winners, at the end," said Lee. But, "dozens will want to compete."

ZDNET

Frequent releases will continue

The AI industry overall is "still ultra-competitive," said Lee. "I expect frequent releases to continue" of new large language models, from "the fast-moving companies," he added.

Among those fast-moving companies, said Lee, Elon Musk's xAI's Grok models, and DeepSeek AI, are the fastest-moving innovators. "I think DeepSeek currently has the momentum," he said.

Also: Why Anthropic's latest Claude model could be the new AI to beat - and how to try it

OpenAI, Alibaba, Google, and Anthropic "are all making respectable moves," he added.

One potential winner not fully appreciated is TikTok owner ByteDance, said Lee.

"I listed ByteDance because we know how much they're spending, and one way you can justify spending a ton of money on pioneer models is because you have so many customers […] ByteDance has more ways to monetize than anybody."

Also: What is sparsity? DeepSeek AI's secret, revealed by Apple researchers

Chinese e-commerce giant Baidu could also be a contender, said Lee. The company has historically had "that kind of vision," meaning the same visionary sensibility as OpenAI, but, "the execution is currently lagging" in terms of producing new foundation-model technology, he said.

One winner and a runner-up

Even a vibrant market such as AI models may eventually see big firms go away, said Lee.

Also: Could this be the reason you cancel your ChatGPT subscription?

"In most markets, you end up with only two [competitors]," noted Lee of technology markets, "the winner, who makes money, and the runner-up, who breaks even, and everyone else, who dies."

Get the morning's top stories in your inbox each day with our Tech Today newsletter.