Now, many companies in China are increasingly building on that foundation to develop products that look to go beyond a chatbot.
Baidu, best known for its search engine and Ernie chatbot, said Tuesday its generative AI-integrated Wenku platform for quickly creating powerpoints and other documents had reached 40 million paying users, with revenue up 60% from a year ago as of the end of last year. Updated features, such as using AI to generate a presentation based on a company’s financial filing, started being rolled out to users in the last week.
On the corporate side, Gartner data and analytics director analyst Ben Yan estimates more than 10% of businesses in China are using AI, up from 8% about six months ago. That would be a pickup in pace — the last 2 percentage point increase in adoption took more than a year, he said Wednesday.
“With our clients, we hear more and more success stories,” he said in Mandarin translated by CNBC. Yan noted that so-called AI agents will help speed up corporate implementation of the new tech.
AI models focus on specific functions such as search and generating summaries, whereas AI agents are more advanced — they can automate entire processes from searching to booking. One example is OpenAI’s new “Operator” function that claims to be able to make restaurant reservations on behalf of a ChatGPT user.
AI agents are also on the verge of coming to the Chinese market at scale.
Tencent plans to soon integrate AI agents with its messaging and social media app WeChat, CEO Pony Ma told staff in a Jan. 13 speech, according to a copy of the annual address seen by CNBC.
“We believe that China’s AI sector is advancing at a pace comparable to that of the United States,” Jo Huang, head of private equity at Raffles Family Office, said in an email. She said the firm is considering investing in a leading China AI deep tech fund in order to capture the local opportunity.
The development of Chinese AI applications creates features that are being integrated with domestic smartphones. Apple’s AI intelligence functions have yet to come to iPhone users in China.
She pointed out that Chinese smartphone companies such as Honor, Xiaomi and Vivo have been able to improve user experience of AI features, thanks to efforts to improve the efficiency of AI models that can run on the device without relying heavily on an internet-connected cloud service.
Compliance hurdles
The latest developments also reflect a difference in regulatory scrutiny with the U.S., and with the kind of technology being created.
While AI models must get official certification for use in China, using them in applications is much easier, said Alex Lu, founder of Shanghai-based LSY consulting. On the side, he is working with a small team on an AI-powered tool for giving companies targeted daily insights on industry trends and global regulations, similar to the work of a human consultant.
Half a year after development began in June 2023, Lu said the team began testing a product for free with potential customers, including a manufacturer of car batteries. That has provided the feedback for a product the team hopes to charge 70,000 yuan to 100,000 yuan ($9,660 to $13,790) in annual license fees, Lu said.
But a bigger challenge can be getting companies to give AI access to proprietary data, or using AI-generated content commercially.
“I think [multinational corporations are] much more cautious than Chinese brands because of copyrights and legal issues,” Chris Reitermann, CEO of Ogilvy Asia-Pacific and Greater China, told CNBC late last year. He is also president of WPP China.
He said clients attempted to use AI for campaigns, only to run into compliance issues that prevented the projects from being launched. “Local brands, they may be a little less worried about these issues, more trial and error,” he said.
AI for global users
Some China-created AI applications are also being used overseas. Alibaba‘s international arm announced earlier this month that Accio, its AI-powered search engine for product sourcing, had reached 500,000 small business users.
Launched in November, Accio lets businesses use a few text or image prompts to find wholesale products — and provides them with analysis on their popularity with consumers and projected profit.
Accio cut the research time down from weeks to a day or so, said Mike McClary, who got early access to Accio and has sold camping lanterns and other products online for more than 10 years. McClary, CEO of amazing.com, claims e-commerce sales of more than $1 million a year and is based outside of St. Louis, Missouri.
He said Alibaba.com and Amazon, which he previously used, involved going through hundreds or thousands of results, and then individually negotiating with five to 10 suppliers before settling on one. The “next gamechanger,” McClary said, would be to use AI to put an image of a product into any scenario to create an advertisement.
A young Chinese AI startup DeepSeek sparked a massive rout in U.S. technology stocks on Monday as its highly competitive — and potentially shockingly cost-effective — models stoked doubts about the hundreds of billions that America’s biggest companies are spending on artificial intelligence.
DeepSeek’s emergence is shaking up investor confidence in the AI story that has been lifting the U.S. bull market the past two years. It calls into question the hype around Nvidia’s chips and rippled all the way through the market to hit shares of power producers who were set to get a boost from AI data center demand.
Here’s how the DeepSeek-triggered market sell-off on Wall Street unfolded:
New Reasoning Model
DeepSeek was founded in May 2023 by Liang Wenfeng, who partly funded the company by his AI-powered hedge fund. In late December, the AI developer launched a free, open-source large language model that it says took only two months to develop and less than $6 million to build.
On Jan. 20, the Hangzhou, China-based DeepSeek released R1, a reasoning model that outperformed Open AI’s latest o1 in many third-party tests.
DeepSeek is seeking to differentiate from its competitors with its reasoning capabilities, meaning that before delivering the final answer, the model first generates a “chain of thought” to enhance the accuracy of its responses.
Top-performing Model
The buzz around DeepSeek’s R1 seemingly picked up steam after Alexandr Wang, CEO of ScaleAI, touted its competitiveness against the best products from the U.S. megacap tech giants, which most had thought were leading the AI war.
“What we found is that deep seek, which is the leading Chinese AI Lab, their model, is actually the top performing, or roughly on par with the best American models,” Wang, whose company provides data to help companies train their AI tools, said on CNBC from the World Economic Forum in Davos, Switzerland last week.
Wang noted that DeepSeek actually has more H100 chips from Nvidia than expected — about 50,000 of them. Those chips are the processor of choice for AI firms in the U.S. such as OpenAI and the U.S. has banned the sale of advanced AI chips to China.
Nvidia shares took a 3% hit on Friday as chatter about DeepSeek started to pick up.
No.1 App
But it wasn’t until the past weekend when the hype surrounding DeepSeek reached a fever pitch on social media.
Marc Andreessen, co-founder and general partner of venture capital firm Andreessen Horowitz, sang DeepSeek’s praises on X, saying the R1 model is “one of the most amazing and impressive breakthroughs” he’s ever seen. Andreessen’s portfolio includes Airbnb and dozens of AI companies.
Tech investor Chamath Palihapitiya on X pointed to DeepSeek’s “very good” report, which said its R1 model “essentially cracked one of the holy grails of AI: getting models to reason step-by-step without relying on massive supervised datasets.”
The Chinese AI app DeepSeek in Apple’s us App Store on an iPhone 12. In the ranking of free apps, DeepSeek was even ahead of ChatGPT from OpenAI.
Picture Alliance | Getty Images
By this point, DeepSeek’s mobile app surged to the top of Apple’s appstore download charts over the weekend in the U.S., marking a tangible threat to pricier models such as OpenAI’s ChatGPT.
U.S. futures were down big overnight Sunday and investors woke up to a sea of red Monday morning.
Nasdaq Composite, 1-day
AI darling and chip producer Nvidia saw shares tumbling more than 12% Monday, on track for its worst day since March 2020.
Nvidia, 1-day
Other chip makers as well as power providers were hit big. The tech-heavy Nasdaq Composite is down as much as 3.6% Monday, dragged down by megacap names.
Check out the companies making headlines in premarket trading. Nvidia — The artificial intelligence darling saw shares tumbling more than 11% in premarket trading, on track for its worst day since March 2020. The sell-off occured after Chinese startup DeepSeek launched a free open-source large-language model in late December, saying it was developed in just two months at a cost of under $6 million. The move called into question the need for Nvidia’s fastest chips and its competitiveness. Other chipmakers also declined, with Broadcom losing 12% and AMD shedding 4%. Microsoft , Amazon — Other megacap stocks also sold off sharply after the news of DeepSeek stoked questions about the large amounts of money big tech companies have been investing in AI models and data centers. Shares of Microsoft shed 5%, while Amazon dropped by more than 4% and Meta Platforms slid by 2.5%. Vertiv Holdings — Shares of the datacenter services company tumbled 16% as the artificial intelligence trade in the U.S. reversed. If the emergence of DeepSeek leads to less AI infrastructure spending than expected in the future, that could hurt Vertiv’s business outlook. Constellation Energy , Vistra — Power providers for AI infrastructure were also hit hard on concerns that there would be less energy needed if Deepseek can perform using fewer chips. Constellation Energy, Vistra, Nuscale Power and Oklo all slid at least 10% in premarket trading. AT & T — The telecommunications giant added 2.3% after earnings for the fourth quarter came in higher than Wall Street expected. The company earned 54 cents per share, excluding items, exceeding the consensus forecast of analysts polled by StreetAccount by 4 cents. AT & T also posted $32.3 billion in revenue, while analysts expected $32.02 billion. D.R. Horton – Shares of the homebuilding company fell nearly 1% after receiving a downgrade to neutral from buy at Bank of America. The bank said a “challenging” housing market backdrop and rising input costs could weigh on the company from here. – CNBC’s Alex Harring, Sean Conlon and Jesse Pound contributed reporting.
Tariffs or not, Chinese markets are still waiting for earnings to turn around, analysts point out. “Regardless of what the number of the tariffs are for China, it comes back to the domestic stimulus for China and whether China can alleviate the deflation pressures,” Aaron Costello, head of Asia at Cambridge Associates, said Thursday. Beijing “has clearly shown a desire” to stimulate the economy, Costello said, noting details are due out at an annual parliamentary meeting in March. “The potential for Chinese equities to rebound sharply is there, so we don’t want to be underweight China, we want to be neutral,” he said. Chinese stocks closed higher Friday after U.S. President Donald Trump ‘s latest comments indicated reluctance to raise tariffs , despite threatening a day earlier that 10% duties could come as soon as Feb. 1. The mainland market also got a lift Thursday after financial regulators effectively mandated state-backed insurers to buy more stocks . While the directive offers longer-term support for stocks, “we reiterate our preference for the A-share market, and for stocks with stable cash returns and decent dividend yields ,” Morgan Stanley’s Chief China Equity Strategist Laura Wang said in a note Thursday. She referred to the firm’s report on Jan. 20 for a list of “well positioned” names. Morgan Stanley surveyed its analysts for Chinese stocks for which they expected to see solid earnings growth in the year ahead. The stocks must be rated overweight or equalweight, have a market capitalization of more than $2 billion and average daily trading turnover of more than $2 million. The three names with the highest expected earnings growth for 2025 are: Espressif Systems — The Shanghai-listed company develops chip sets for home appliances. Earlier this month it said its net profit more than doubled in 2024 . SICC — Founded in 2010, the Shanghai-listed company produces silicon carbide substrate, used in semiconductors. It said in December it plans to list in Hong Kong at an unspecified date . Zijin Mining — The Hong Kong-listed mining company, which extracts metals such as copper, gold, zinc and lithium, said net profit in the third quarter rose by more than 50% from a year ago. Morgan Stanley expects each company can grow earnings per share by at least 40% in 2025. “Quality earnings beats becoming a proven alpha generator in the China equity space and should continue to be so,” the analysts said in the Jan. 20 report. They said Chinese stocks have missed earnings expectations for 13 straight quarters since late 2021. But in their historical analysis of stock performance between 2021 and 2024, they found that earnings beats and upward revisions led to significant outperformance versus companies that missed or had earnings estimates cut. Overseas revenue has increasingly become a growth driver for Chinese companies as they face a slower economy at home. And despite worries about geopolitics hitting cross-border e-commerce, Bernstein analysts pointed out in a Wednesday note that the market outside the U.S. is “as big, if not bigger than the U.S. one.” Total e-commerce gross merchandise value in the U.S. was $1.1 trillion in 2023, while the next 29 markets for which eMarketer has data had a total GMV of $1.5 trillion, Bernstein said. Bernstein analysts expect PDD and Alibaba earnings to grow in the year ahead, but the only one they rate outperform is the Temu parent. They have a price target of $150 a share on PDD, for upside of more than 40% from Thursday’s close. “From an investing standpoint, our sense is global (and in particular US) investors take a very US-centric view of Temu, and what it means for PDD’s shares,” the analysts said. “In contrast, we’d argue that Temu’s US experience in the past 12-18 months — showing a large jump in profitability once new user acquisition was de-emphasised — demonstrates the path to profitability elsewhere.” — CNBC’s Michael Bloom contributed to this report.