Virtual singer Luo Tianyi performing with world renowned pianist Lang Lang in 2019 at the Mercedes-Benz Arena in Shanghai, China. Launched in 2012, Luo Tianyi has nearly 3 million fans and even performed at the Winter Olympics opening ceremony in Beijing this year.
Visual China Group | Getty Images
BEIJING — From customer service to the entertainment industry, businesses in China are paying big bucks for virtual employees.
Tech company Baidu said the number of virtual people projects it’s worked on for clients has doubled since last year, with a wide price range of as little as $2,800 to a whopping $14,300 per year.
Virtual people are a combination of animation, sound tech and machine learning that create digitized human beings who can sing and even interact on a livestream. While these digital beings have appeared on the fringes of the U.S. internet, they’ve been popping up more and more in China’s cyberspace.
Some buyers of virtual people include financial services companies, local tourism boards and state media, said Li Shiyan, who heads Baidu’s virtual people and robotics business.
As the tech improves, costs have dropped by about 80% since last year, he said. It costs about 100,000 yuan ($14,300) a year for a three-dimensional virtual person, and 20,000 yuan for a two-dimensional one.
Li expects the virtual person industry overall will keep growing by 50% annually through 2025.
Searching for scandal-free icons
From a business perspective, much of the focus is on how virtual people can generate content.
Brands in China are looking for alternative spokespeople after many celebrities recently ran into negative press about tax evasion or personal scandals, said Sirius Wang, chief product officer and head of marketplace Greater China at Kantar.
Dancers perform with virtual digital people at the Future Life Festival 2022 in Hangzhou, China, on Nov. 4, 2022.
At least 36% of consumers had watched a virtual influencer or digital celebrity perform in the last year, according to a survey published by Kantar this fall. Twenty-one percent had watched a virtual person host an event or broadcast the news, the report said.
Looking ahead to next year, 45% of advertisers said they might sponsor a virtual influencer’s performance or invite a virtual person to join a brand’s event, according to the Kantar report.
Growing development of virtual people
Many of China’s large tech companies have already been developing products in the virtual humans industry.
Video and game streaming app Bilibili was one of the earliest to take the concept of virtual people mainstream.
The company acquired the team behind virtual singer Luo Tianyi, whose image and sound are fully created by tech. This year, the developers focused on improving the texture of the virtual singer’s voice by using an artificial intelligence algorithm, according to Bilibili.
Launched in 2012, Luo Tianyi has nearly 3 million fans and even performed at the Winter Olympics opening ceremony in Beijing this year.
Bilibili also hosts many so-called virtual anchors, which are the direct avatars of people using special technology to reach their audience. The company said 230,000 virtual anchors started broadcasting on its platform since 2019, and the virtual anchors’ broadcasting time this year surged by about 200% from last year.
Tencent said in its latest earnings call that Tencent Cloud AI Digital Humans provide chatbots to sectors such as financial services and tourism for automated customer support. The company’s Next Studios also developed a virtual singer and virtual sign language interpreter.
Far smaller companies are also getting into the industry.
Startup Well-Link Technologies — whose cloud rendering tech support for Chinese video game developer miHoYo brought it success in the gaming industry — announced this year it has developed yet another model of a virtual person in a joint venture with Haixi Media.
Oracle CEO Clay Magouyrk appears on a media tour of the Stargate AI data center in Abilene, Texas, on Sept. 23, 2025.
Kyle Grillot | Bloomberg | Getty Images
Oracle on Friday pushed back against a report that said the company will complete data centers for OpenAI, one of its major customers, in 2028, rather than 2027.
The delay is due to a shortage of labor and materials, according to the Friday report from Bloomberg, which cited unnamed people. Oracle shares fell to a session low of $185.98, down 6.5% from Thursday’s close.
“Site selection and delivery timelines were established in close coordination with OpenAI following execution of the agreement and were jointly agreed,” an Oracle spokesperson said in an email to CNBC. “There have been no delays to any sites required to meet our contractual commitments, and all milestones remain on track.”
The Oracle spokesperson did not specify a timeline for turning on cloud computing infrastructure for OpenAI. In September, OpenAI said it had a partnership with Oracle worth more than $300 billion over the next five years.
“We have a good relationship with OpenAI,” Clay Magouyrk, one of Oracle’s two newly appointed CEOs, said at an October analyst meeting.
Doing business with OpenAI is relatively new to 48-year-old Oracle. Historically, Oracle grew through sales of its database software and business applications. Its cloud infrastructure business now contributes over one-fourth of revenue, although Oracle remains a smaller hyperscaler than Amazon, Microsoft and Google.
OpenAI has also made commitments to other companies as it looks to meet expected capacity needs.
In September, Nvidia said it had signed a letter of intent with OpenAI to deploy at least 10 gigawatts of Nvidia equipment for the San Francisco artificial intelligence startup. The first phase of that project is expected in the second half of 2026.
Nvidia and OpenAI said in a September statement that they “look forward to finalizing the details of this new phase of strategic partnership in the coming weeks.”
But no announcement has come yet.
In a November filing, Nvidia said “there is no assurance that we will enter into definitive agreements with respect to the OpenAI opportunity.”
OpenAI has historically relied on Nvidia graphics processing units to operate ChatGPT and other products, and now it’s also looking at designing custom chips in a collaboration with Broadcom.
On Thursday, Broadcom CEO Hock Tan laid out a timeline for the OpenAI work, which was announced in October. Broadcom and OpenAI said they had signed a term sheet.
“It’s more like 2027, 2028, 2029, 10 gigawatts, that was the OpenAI discussion,” Tan said on Broadcom’s earnings call. “And that’s, I call it, an agreement, an alignment of where we’re headed with respect to a very respected and valued customer, OpenAI. But we do not expect much in 2026.”
“This is the wrong approach — and most likely illegal,” Sen. Amy Klobuchar, D-Minn., said in a post on X Thursday.
“We need a strong federal safety standard, but we should not remove the few protections Americans currently have from the downsides of AI,” Klobuchar said.
Trump’s executive order directs Attorney General Pam Bondi to create a task force to challenge state laws regulating AI.
The Commerce Department was also directed to identify “onerous” state regulations aimed at AI.
The order is a win for tech companies such as OpenAI and Google and the venture firm Andreessen Horowitz, which have all lobbied against state regulations they view as burdensome.
It follows a push by some Republicans in Congress to impose a moratorium on state AI laws. A recent plan to tack on that moratorium to the National Defense Authorization Act was scuttled.
Collin McCune, head of government affairs at Andreessen Horowitz, celebrated Trump’s order, calling it “an important first step” to boost American competition and innovation. But McCune urged Congress to codify a national AI framework.
“States have an important role in addressing harms and protecting people, but they can’t provide the long-term clarity or national direction that only Congress can deliver,” McCune said in a statement.
Sriram Krishnan, a White House AI advisor and former general partner at Andreessen Horowitz, during an interview Friday on CNBC’s “Squawk Box,” said that Trump is was looking to partner with Congress to pass such legislation.
“The White House is now taking a firm stance where we want to push back on ‘doomer’ laws that exist in a bunch of states around the country,” Krishnan said.
He also said that the goal of the executive order is to give the White House tools to go after state laws that it believes make America less competitive, such as recently passed legislation in Democratic-led states like California and Colorado.
The White House will not use the executive order to target state laws that protect the safety of children, Krishnan said.
Robert Weissman, co-president of the consumer advocacy group Public Citizen, called Trump’s order “mostly bluster” and said the president “cannot unilaterally preempt state law.”
“We expect the EO to be challenged in court and defeated,” Weissman said in a statement. “In the meantime, states should continue their efforts to protect their residents from the mounting dangers of unregulated AI.”
Weissman said about the order, “This reward to Big Tech is a disgraceful invitation to reckless behavior by the world’s largest corporations and a complete override of the federalist principles that Trump and MAGA claim to venerate.”
In the short term, the order could affect a handful of states that have already passed legislation targeting AI. The order says that states whose laws are considered onerous could lose federal funding.
One Colorado law, set to take effect in June, will require AI developers to protect consumers from reasonably foreseeable risks of algorithmic discrimination.
Some say Trump’s order will have no real impact on that law or other state regulations.
“I’m pretty much ignoring it, because an executive order cannot tell a state what to do,” said Colorado state Rep. Brianna Titone, a Democrat who co-sponsored the anti-discrimination law.
In California, Gov. Gavin Newsom recently signed a law that, starting in January, will require major AI companies to publicly disclose their safety protocols.
That law’s author, state Sen. Scott Wiener, said that Trump’s stated goal of having the United States dominate the AI sector is undercut by his recent moves.
“Of course, he just authorized chip sales to China & Saudi Arabia: the exact opposite of ensuring U.S. dominance,” Wiener wrote in an X post on Thursday night. The Bay Area Democrat is seeking to succeed Speaker-emerita Nancy Pelosi in the U.S. House of Representatives.
Trump on Monday said he will Nvidia to sell its advanced H200 chips to “approved customers” in China, provided that U.S. gets a 25% cut of revenues.
Coinbase is gearing up to launch an in-house prediction market, powered by Kalshi, a source close to the matter told CNBC — a strategic play to expand the number of asset classes available on the cryptocurrency exchange at a time some investors are shying away from digital assets.
The source said Coinbase and Kalshi will “soon” formally announce the prediction market, with news on the matter potentially coming as early as next week.
Rumblings of the prediction market launch have swirled for nearly a month. An alleged screenshot of Coinbase’s prediction markets dashboard shared by Silicon Valley researcher Jane Manchun Wong in an X post dated Nov. 18 offered some clues about the new product.
The Information first reported on Nov. 19 that Coinbase planned to launch prediction markets powered by Kalshi, adding that the exchange would unveil the new product at its “Coinbase System Update” event on Dec. 17. Bloomberg published a similar report on Thursday, citing a source familiar with the matter, adding that Coinbase would also announce a tokenized stock offering at the showcase.
Coinbase declined to confirm the reports to CNBC, but said to tune into its event next week. The firm did not comment on a timeline for when its prediction markets would go live for its users.
Coinbase’s upcoming product launches underscore its push to refashion itself into an “everything exchange,” or a one-stop shop for trading all kinds of assets, including crypto tokens, tokenized stocks and event contracts. In May, CEO Brian Armstrong articulated that “everything exchange” vision to investors, saying Coinbase would aim to become a top financial services app within the next decade.
The trading platform is setting its sights on that goal as it faces intensifying competition from rivals such as Robinhood,Gemini and Kraken. All three have launched tokenized equity offerings to users outside of the U.S. within the past year, in addition to exploring prediction markets to varying extents.
Coinbase’s moves to expand the financial instruments available to its users also come as investor sentiment on digital assets cools. A series of liquidations of highly leveraged digital asset positions in mid-October triggered several pullbacks in the crypto market, prompting investors to rotate out of tokens and into gold and other safe-have assets.
Bitcoin fell as low as around $85,000 in early December, hitting its lowest level since last March. The token was last trading at $89,951, down 23% in the past three months. Coinbase has also fallen more than 16% over the past three months.
The deal also underscores U.S.-based prediction markets operator Kalshi’s push to embed its event contracts into various brokerages, widening its reach as the prediction markets space becomes increasingly competitive.
This year, Kalshi embedded several of its prediction markets into trading platform Robinhood, as part of a non-exclusive partnership between the companies. Kalshi has also engaged in talks with several other major brokerages, including those in the crypto industry, with the aim of closing more deals like the ones it has struck with Robinhood and now Coinbase, a source familiar with the matter told CNBC.