Chinese technology leaders rarely speak publicly, making DeepSeek CEO Liang Wenfeng's latest remarks particularly significant. In a rare insight into the philosophy driving one of China's leading AI companies, Liang made it clear that DeepSeek is not focused on profits or revenue. Instead, its ultimate objective is achieving artificial general intelligence (AGI) — AI capable of reasoning and learning in ways comparable to humans.
A transcript of a nearly four-hour discussion between Liang and DeepSeek's investors was recently made public. During the conversation, he outlined the company's long-term vision and positioned DeepSeek as a challenger not only to Silicon Valley's AI models but also to its dominance in advanced AI chips.
No IPO, profits not the priority
While companies such as OpenAI and Anthropic are reportedly pursuing public listings at massive valuations, Liang dismissed the idea of taking DeepSeek public.
He said generating wealth had never been the company's primary objective.
"Our original intention was not that I ultimately wanted to make a certain amount of money, to go to the capital markets, to IPO, or anything like that. So we didn't have that original intention," Liang said.
He compared short-term revenue to "sesame seeds," saying DeepSeek would collect them only when convenient without making them a central focus.
"The sesame seeds in front, we'll pick some up, but we just pick up a bit casually — we won't stop, or treat it as an important matter."
Instead, he said the company was focused on pursuing the much larger "watermelons" that could emerge in the future.
According to Liang, DeepSeek's purpose is to create technology that benefits humanity rather than maximise profits.
"So overall, we're doing this with tremendous goodwill toward the world, and then we feel this is useful for humanity—this is something beyond money."
He also stressed that a company's vision is reflected through its actions rather than slogans.
"What is vision? Vision is not a slogan hanging on the wall—vision is how you do things, not how you talk, that is, how you actually operate," he said.
AGI remains DeepSeek's long-term goal
While many US AI firms are increasingly focusing on enterprise customers and commercial products, Liang said DeepSeek's long-term objective remains AGI.
"I think our goal should be AGI."
Although he acknowledged there is no universally accepted definition of AGI, he said that would not prevent DeepSeek from pursuing it.
"Everyone's definition of AI isn't necessarily the same, but that doesn't stop us from treating AGI as our goal."
Liang compared AI development to climbing a staircase, with each breakthrough serving as the foundation for the next.
"Why is it a staircase? Because each subsequent step is based on the previous foundation. Agent uses CoT (chain of thought), and CoT also uses the previous step, which is the language model – so no step is wasted."
He believes AI agents will continue improving until systems become capable of continuously learning and eventually reaching what he described as technological singularity.
"It could then develop its own version, could research on its own, then develop its own next version, developing more advanced AI models. So it would reach a singularity, able to achieve its own iteration."
Liang added that DeepSeek wants its AI models to become capable of helping build future generations of models, accelerating progress toward AGI.
According to him, the biggest remaining challenge is not creativity or intuition but enabling AI systems to learn continuously.
US restrictions could accelerate China's chip industry
Training advanced AI models requires powerful graphics processing units (GPUs), and US export restrictions have limited China's access to Nvidia's most advanced chips.
Liang argued that these restrictions may ultimately strengthen China's domestic semiconductor industry.
"If it were in a normal business environment where I could buy Nvidia cards, then domestic substitution would be relatively hard, but when Nvidia cards can't be bought, everyone is forced into it – everyone has to do domestic chips."
He even suggested that Nvidia's restrictions were encouraging competitors to catch up.
"I think on this point, Nvidia is digging its own grave."
Liang claimed Huawei's latest AI computing systems could eventually become viable alternatives to Nvidia's offerings.
"Huawei's supernode, Huawei's 950 supernode, in performance and price can completely substitute for Nvidia's GB200, GB300."
DeepSeek is already collaborating closely with Huawei on AI chip development.
"We currently mainly cooperate with Huawei. Huawei does their own adaptation, but we ourselves will participate in this ecosystem, and will deeply participate within Huawei."
He also revealed that DeepSeek has increasingly reduced its reliance on Nvidia's software ecosystem by developing its own compiler technology.
"When DeepSeek V3 was trained, it still used Nvidia cards, but it already didn't use Nvidia's ecosystem. We first wrote a high-level compiler called TileLang, then, based on TileLang's ecosystem completed everything else."
However, Liang acknowledged that China's hardware still trails Nvidia's technology.
"The only cost is that four Huawei cards equal one NVIDIA card, and simultaneously lagging by two years."
Betting on cheaper Chinese AI
Liang believes Chinese AI companies will eventually compete by offering models that match global rivals in performance while costing significantly less.
"Chinese people will make this product the cheapest."
He argued that Chinese AI could follow the same path as several other industries where Chinese manufacturers offer comparable products at lower prices.
"In the future AI might be like this too, but Chinese-made AI might have a cheaper price. This cheapness might be systematically low, just like in other industries where China provides cheaper services."
Unlike OpenAI and Anthropic, whose proprietary models typically operate on paid, token-based pricing, DeepSeek offers open-weight AI models that are considerably more affordable to use. Other Chinese AI firms, including Moonshot AI with Kimi K3 and Zai with GLM 5.2, have also adopted the open-weight approach.
