By Krystal Hu

Feb 18 (Reuters) - (Artificial Intelligencer is published every Wednesday. Think your friend or colleague should know about us? Forward this newsletter to them. They can also subscribe here or email me to share any thoughts.)

Have ads started popping up in your conversations with ChatGPT yet?

More than 30 advertisers have run ads on ChatGPT this month, according to estimates from Sensor Tower. Unsurprisingly, a third of them are retailers — including Best Buy, Urban Outfitters, Target and Home Depot — looking to reach consumers inside chatbot conversations.

Best Buy has promoted products ranging from video games to home appliances, while brands like Chewy and Petco have advertised pet supplies directly through ChatGPT. With more than 800 million weekly active users, the platform has become a rich new channel for brands experimenting with AI-driven visibility.

Still, the strategy is evolving. Kimberly Shenk, CEO of Novi, a startup that helps brands optimize their presence in chatbot answers, told me many companies are still figuring out how paid ads should complement organic mentions — especially when their products already surface naturally in AI responses.

Not everyone is eager to follow OpenAI’s lead. Google DeepMind CEO Demis Hassabis said he was “a bit surprised” by OpenAI's move into advertising, and that he is under no pressure to do so. Perplexity is also dropping ads for now, following early experiments, to focus on building trust and accuracy.

In this week’s issue, we look at how China used its biggest annual holiday to showcase advances in AI hardware and software — and how xAI’s controversial image-generation feature boosted its popularity and unleashed a backlash. Scroll on.

OUR LATEST REPORTING IN TECH AND AI:

• Microsoft to keep buying enough renewable energy to match all its electricity needs

• AI adoption already hitting Irish graduate jobs, finance department says

• Nvidia to sell Meta millions of chips in multiyear deal

• California builds AI oversight unit and presses on xAI investigation

• Waymo defends use of remote assistance workers in US robotaxi operations

• India tells university to leave AI summit after presenting Chinese robot as its own, sources say

CHINA RINGS IN THE NEW YEAR WITH AN AI SHOW-OFF

Can you imagine a Super Bowl halftime show where, instead of human dancers, all the performers are humanoid robots?

That’s essentially what China staged at this year’s Spring Festival Gala — the annual state television variety show that most Chinese families watch to ring in the Lunar New Year. The gala, a glossy mix of pop stars, comedy sketches, dance and magic tricks, is part entertainment and part national temperature check, and remains the world’s most-watched TV program.

Chinese robotics firms, from Unitree and Galbot to Noetix, put their humanoids in the spotlight. Early in the show, a comedy sketch featured a grandmother proudly doting on her four small humanoid robots, joking that they take better care of her than her grandson, who forgets to visit. For now, they can’t actually care for her. But they can flip, march, hold props and execute choreographed movements — even simulated laughter — with uncanny precision.

Later, more humanoids performed extended martial arts routines, with over a dozen Unitree robots demonstrating synchronized fight sequences, waving swords, poles and nunchucks near human child performers.

The humanoid showcase quickly went viral both inside and outside China. It highlighted advances in motor control, coordination and safer human-robot interaction — and underscored Beijing’s push to dominate humanoid robotics and the future of manufacturing.

The attention comes as major players, including AgiBot and Unitree, prepare for potential IPOs this year.

But China’s AI ambitions aren’t confined to hardware. Over the Lunar New Year holiday, domestic AI startups released a wave of advanced open-source models, reshaping the global AI conversation from Beijing to Silicon Valley.

Since DeepSeek broke out and shook the Western tech world a year ago, Chinese open-source models such as Alibaba’s Qwen and Kimi from Moonshot AI have gained attention and adoption in the U.S. Some now rival leading models on reasoning and coding benchmarks. Others emphasize efficiency — models small enough to run cheaply and widely.

Recent updates have made waves among U.S. developers for delivering strong performance at lower cost, fueling startup experimentation and putting pricing pressure on top American labs.

Unlike the more closed, API-centric strategies of OpenAI and Anthropic, many Chinese labs are releasing models with open weights. The strategy is deliberate. Open models attract developers, lower barriers to entry, and accelerate adoption globally — especially in emerging markets. On Hugging Face, Alibaba’s Qwen family, the most downloaded model series in 2025, has overtaken Meta’s Llama models in cumulative downloads. Chinese open-source models have surpassed U.S. models in total downloads, according to a recent MIT study.

For policymakers debating export controls on advanced chips, the message is nuanced. Even with limited access to cutting-edge GPUs, Chinese firms are learning to do more with less — and using open distribution to win developer mindshare worldwide.

While humanoids flipped and bowed on the Spring Gala stage, one of the robots featured is already listed for sale on Amazon for over $17,000. While impressive on stage, these machines are still far from replacing human caregivers. In software, the shift is already here. The chatbots we use daily, the cost of running them, and who shapes the next generation of models — China’s AI push is influencing all of it.

CHART OF THE WEEK

The AI chatbot race is entering a more competitive phase. While ChatGPT remains the dominant player in the U.S., rivals are steadily gaining ground. According to Apptopia data, ChatGPT’s market share fell to 52.9% last month among the top three chatbots, down from 80.9% in January 2025. Over the same period, Google’s GOOGL.O Gemini expanded its share to 29.4% from 17.3%, while xAI’s Grok climbed to 17.8% from 1.9% a year ago.

Gemini’s growth is tied to viral consumer features and deeper integration across Google’s ecosystem. Grok’s rise, meanwhile, has been linked to its integration into X, but that growth has come with controversy. Last month, Grok generated a wave of AI-altered, sexually explicit images, including children, in response to user prompts, triggering global backlash and regulatory scrutiny.

California’s attorney general told Reuters he sent a cease-and-desist letter to xAI, accusing the company of deliberately enabling harmful outputs and child sexual abuse material and vowing to hold the Elon Musk firm accountable. In January, the company said it had added measures to reject user requests for sexualized images of real people, for instance, editing them to be in a bikini. In jurisdictions where such images are illegal, xAI has also said it blocks users from creating that content.

(Reporting by Krystal Hu; Editing by Lisa Shumaker)

((krystal.hu@thomsonreuters.com, +1 917-691-1815))