Alibaba Teams Up With Zelos on $2 Billion Robovan Business — Update

By Tracy Qu

Alibaba Group has secured several high-profile entities, including State Grid Corp. of China, XPeng and the Chinese Academy of Sciences, as clients for its new AI chip, according to a person familiar with the matter.

The company's Zhenwu 810E chip, which is designed for both training and inference, has more than 400 clients. It has also been used internally for training Alibaba's Qwen models, according to the source.

The Wall Street Journal first reported on Alibaba's development of the new chip in August.

This comes as Chinese tech companies step up efforts to develop homegrown technologies amid a more challenging geopolitical environment.

Investor interest in domestic chip development is high. Shares of Chinese AI chip startups Moore Threads and MetaX have soared since their blockbuster debuts in December.

Write to Tracy Qu at tracy.qu@wsj.com

By Adam Clark

Nvidia was rising early on Wednesday. Good news on sales in China and potential funding for ChatGPT-developer OpenAI bode well for the chip maker.

Nvidia shares were up 1.6% at $191.45 in premarket trading. It looks set to break out its trading range of between $170 and $190, where it has largely been since early November.

China has approved purchases of Nvidia's H200 artificial-intelligence chip for the first time, allowing companies including Alibaba Group to buy a total of several hundred thousand chips worth around $10 billion, The Wall Street Journal reported Wednesday, citing people familiar with the matter.

Nvidia didn't immediately respond to a request for comment early on Wednesday. Nvidia has agreed to pass on a 25% cut of Chinese sales to the U.S. government.

As Barron's has written, Chinese developers are keen to get their hands on Nvidia's hardware as they face bottlenecks in development and deployment of their AI models. The H200 outperforms any Chinese-produced chip, with a roughly 32% processing-power advantage over Huawei's Ascend 910C, according to the Institute for Progress think tank.

Among other chip makers, Advanced Micro Devices was up 2.2% and Broadcom was rising 1.2% in premarket trading. AMD also hopes to sell chips in China, while Broadcom works with China's ByteDance on custom AI chips.

Meanwhile, Nvidia could get a further boost if OpenAI achieves its target of up to $100 billion in additional funding. Japan's SoftBank Group is in talks to invest up to $30 billion more in OpenAI, the WSJ reported.

SoftBank and OpenAI didn't immediately respond to requests for comment from Barron's.

Nvidia and OpenAI's fates are entwined. Nvidia has agreed to invest up to $100 billion in OpenAI. Meanwhile, OpenAI has agreed to lease up to five million of Nvidia's chips at a value of $350 billion according to reporting in October by the WSJ.

Barron's owner News Corp has a content-licensing partnership with OpenAI.

Write to Adam Clark at adam.clark@barrons.com

This content was created by Barron's, which is operated by Dow Jones & Co. Barron's is published independently from Dow Jones Newswires and The Wall Street Journal.

By Raffaele Huang

SINGAPORE--China has approved purchases of Nvidia's popular H200 artificial-intelligence chip for the first time, giving authorization to several of Nvidia's Chinese customers, people familiar with the matter said Wednesday.

The long-awaited move came during a trip to China by Jensen Huang, chief executive of the American chip giant. The Trump administration said late last year that Nvidia could sell the AI chip to Chinese companies, but it had been uncertain whether Beijing would allow the sales to go through.

The first approval covers several hundred thousand H200 chips, which would be worth around $10 billion, the people said. Major technology companies including Alibaba and ByteDance received the initial batch of approvals, and authorities are expected to greenlight more imports in coming weeks, they said.

Write to Raffaele Huang at raffaele.huang@wsj.com

By Adam Clark

It's a year on from the "DeepSeek moment" that sparked a $1 trillion market panic.

The artificial-intelligence trade has proved resilient since then because a stream of more capable models continues to justify vast spending by U.S. companies. But Chinese AI models are threatening to capture more of the market in 2026, making it harder for U.S. leaders such as Alphabet's Google, OpenAI, and Anthropic to cash in.

In January 2025, investors misread DeepSeek's purported $5.6 million training cost as evidence that expensive Nvidia chips were becoming obsolete and that large U.S. technology companies had overinvested in AI. Instead, what followed was a demonstration of Jevons' Paradox: More capable and efficient models greatly increased demand for AI services, ultimately justifying the data-center build-outs by Microsoft, Amazon.com, and Google.

However, markets may now be overlooking another factor that makes DeepSeek and its domestic peers dangerous for Western rivals. Their models are often "open source" or "open weight," meaning they are free for users to download and modify, and cheap to use. That contrasts with U.S. companies, which generally maintain full control of their models and charge higher subscription fees.

While U.S. models such as Google's Gemini, Anthropic's Claude or OpenAI's GPT remain leaders in complex reasoning, Chinese open-source models have captured around 30% of the "working" market — specifically in programming and roleplay — where cost efficiency and flexibility are prioritized, according to a report by OpenRouter, an AI model marketplace.

Opening Up

China's open approach also accelerates AI development by allowing users to inspect the code and suggest improvements.

"China seems to be taking a different strategy where if you open the weights and let this diffuse across society and allow it to accelerate research, that can help compensate for the fact that you may not be able to compete directly with companies like OpenAI or Anthropic," says Kyle Miller, a research analyst at Georgetown's Center for Security and Emerging Technology, in an interview with Barron's.

Estimates of how far China's models lag behind U.S. capabilities vary, but the average time frame tends to be around seven months before Chinese developers can match a new American release, according to the research institute Epoch AI. The gap tightened around the time of DeepSeek's R1 release early last year but has since widened again.

The U.S. lead has been sustained by enormous investment. Consensus estimates for capital expenditure by AI hyperscalers — almost all American — put spending at around $400 billion in 2025 and more than $520 billion in 2026, according to Goldman Sachs Research. UBS analysts estimate that the combined capital spending of China's internet leaders was equivalent to around $57 billion last year.

Powering Ahead

Sustaining that pace faces obstacles, most notably power supply. New data-center designs require more than one gigawatt of power, equivalent to the capacity of a nuclear reactor. China now produces more than twice as much power as the U.S., and its centralized planning system can direct energy capacity toward AI faster than the decentralized American model.

"We continue to favour China's approach to AI over that of the U.S. China's open-source models, combined with access to almost unlimited cheap power, make it a formidable competitor," wrote Christopher Woods, global head of equity strategy at Jefferies, in a research note this month. "In this respect, the message from the DeepSeek moment early last year remains highly relevant, even though the U.S. stock market seems, for now at least, to have forgotten about it."

U.S. companies' key advantage remains access to advanced AI chips from Nvidia. DeepSeek is developing the next version of its flagship model; the speculation is that it will arrive in mid-February, around China's Lunar New Year. After experimenting with chips from Huawei Technologies and other domestic vendors, it found the results unacceptable and turned to Nvidia GPUs for some of the training, The Wall Street Journal reported, citing people familiar with the matter. DeepSeek didn't respond to a request for comment from Barron's.

Chinese firms are nevertheless finding ways to advance within chip constraints. DeepSeek published research this month outlining a method of training larger models using fewer chips through a more efficient memory design.

"We view DeepSeek's architecture as a new, promising engineering solution that could enable continued model scaling without a proportional increase in GPU capacity," wrote UBS analyst Timothy Arcuri.

Trump Tightrope

Export controls haven't prevented Chinese companies from training advanced models, but challenges emerge when the models are deployed at scale. Zhipu AI, which released its open-weight GLM 4.7 model in December, said this month it was rationing sales of its coding product to 20% of previous capacity after demand from users overwhelmed its servers.

"I don't see compute constraints limiting [Chinese companies'] ability to make models that are better and compete near the U.S. frontier," Miller says. "I would say compute constraints hit on the wider ecosystem level when it comes to deployment."

That is where President Donald Trump's plan to allow Nvidia to sell its H200 chips to China could be pivotal. Alibaba Group and ByteDance, TikTok's parent company, have privately indicated interest in ordering more than 200,000 units each, Bloomberg reported.

With access to those chips, Chinese labs could build AI-training supercomputers as capable as American ones at 50% extra cost compared with U.S.-made ones, according to the Institute for Progress. Subsidies by the Chinese government could cover that differential, leveling the playing field, the institute says.

The think tank, based in Washington, D.C., notes the H200 outperforms any Chinese-produced chip, with a roughly 32% processing-power advantage over Huawei's Ascend 910C.

A combination of open-source innovation and loosened chip controls could create a cheaper, more capable Chinese AI ecosystem. The possibility is emerging just as OpenAI and Anthropic consider public listings and U.S. hyperscalers such as Microsoft and Meta Platforms face pressure to justify heavy spending.

Rather than a new "DeepSeek moment," the bigger threat is likely to be a slow realization that Chinese companies are structurally undercutting American rivals, triggering a revaluation of U.S. technology stocks.

Write to Adam Clark at adam.clark@barrons.com

This content was created by Barron's, which is operated by Dow Jones & Co. Barron's is published independently from Dow Jones Newswires and The Wall Street Journal.

Local investors in China are excited about a new artificial intelligence stock trend.

As trading volumes in the retail investor-dominated mainland Chinese stocks surged to record highs this month, one of the big themes centered on generative engine optimisation, or GEO. It's the idea that advertisers will spend more on getting brands to show up in AI-generated chatbot results.

Part of the shift is driven by concerns about a bubble in AI chips and other computing power infrastructure, said Wei Wang, a researcher at Tianjin University of Commerce who also runs 10 investment-focused Chinese group chats with more than 3,000 members.

News in late December of Meta Platforms' deal to buy Manus has also bolstered Chinese investors' interest in agentic AI, Wang said, referring to generative AI tools that can automatically make a series of decisions in order to produce better output.

Manus started developing its agentic AI tools — for tasks such as market research and data analysis — as a start-up in China, before relocating to Singapore.

"AI agents and its monetisation opportunity ... is likely to be the top investment theme in 2026," Bank of America analysts said in a report earlier this month. "We believe established China Internet ecosystems, namely Tencent, Alibaba and ByteDance, have natural advantage on incorporating AI agents."

"Within China internet sector, we see Alibaba (Buy) as the best proxy for 'AI in China' theme," the BofA analysts said, noting the e-commerce giant also has a strong AI cloud business. They rate the stock a buy, with a price target of $180.

This month, Alibaba upgraded its Qwen AI app so that users interacting with the chatbot can shop, order food and pay without leaving the app — thanks to integration with the company's existing e-commerce platforms. Qwen claims more than 100 million monthly active users.

Gaming giant Tencent, which operates the WeChat app with around 1.4 billion users, has developed its own AI chatbot and AI-based advertising tools.

"We continue to see Tencent as the key AI application beneficiary in China Internet, given AI has helped empower long runways of growth across all of its major business lines, games, advertising, fintech and cloud," Goldman Sachs analysts said in a Jan. 19 report. They rate the stock a buy, with a price target of 752 Hong Kong dollars.

Adapting to changing ad spend

ByteDance, which owns TikTok and isn't publicly traded, is driving the industry race in China with the country's most popular AI app, Doubao, multiple analysts have pointed out. Late last year, ByteDance also started testing how it might integrate Doubao's AI capabilities into a smartphone.

"We see 2026 as a strategic pivot year for China internet mega-caps with stepping up of AI to-[consumer] investments," the Goldman analysts said. They predict it could be the first year for markets to realise potential disruptions to user habits — such as more entertainment screen time if AI assistants take care of mundane tasks.

"Similar to trends seen in the US, we expect more brands/advertisers in China to adopt ROI-based ads in eCommerce/local services," the analysts said, anticipating advertising budgets shift from traditional search engine optimisation (SEO) strategies to one combined with GEO and a similar AI-focused strategy called AEO, or answer engine optimisation.

Chinese research firm Analysys this month predicted that China's GEO market would skyrocket in value to 3 billion yuan ($430 million) this year, up from 250 million yuan ($35.9 million) in 2025. They expect further growth to 9 billion yuan ($1.29 billion) in 2027.

Even then, it's still a fraction of the Chinese social media online advertising industry, which Goldman Sachs estimates will grow to well over 600 billion yuan this year.

While AI's rapid development doesn't guarantee which tools will end up producing the most commercial value, large companies that adapt can likely stay ahead. The mainland China investment capital that piled into locally traded stocks is also pouring into those plays in Hong Kong.

Alibaba and Tencent shares traded in Hong Kong ranked among the three most-popular stocks by net buys from mainland-based investors over the last seven days, according to Wind Information.

Those "southbound" flows from the mainland to Hong Kong have become a "meaningful price influencer, if not price setter, for many China internet stocks," the BofA analysts pointed out. For stocks such as Alibaba that are also listed in the U.S., that liquidity is rivaled or exceeded by trading in Hong Kong, the report said.

—CNBC's Michael Bloom contributed to this report.

By Adam Clark

Nvidia stock was rising early Friday following a report that Chinese authorities will permit sales of its hardware in the country.

Nvidia shares were up 1.6% at $187.74 in premarket trading. The stock rose 0.8% on Thursday.

Shares were climbing after Bloomberg reported that Beijing has authorized Chinese firms including Alibaba to prepare orders for Nvidia's H200 chips.

Chinese companies would be willing to buy around 1.5 million of the H200 chips, representing roughly $30 billion in revenue, according to KeyBanc analyst John Vinh. Nvidia has agreed to pass on a 25% cut of the sales to the U.S. government.

Manufacturers in the supply chain for Nvidia's H200 chips — designed specifically for the Chinese market — had previously paused production after customs officials blocked shipments of the hardware, the Financial Times reported. However, Beijing's stance now appears to be softening.

Among other chip makers, Advanced Micro Devices was rising 2.7% and Broadcom was down 0.3% in premarket trading. AMD also hopes to sell chips in China, while Broadcom works with ByteDance on custom AI chips.

Write to Adam Clark at adam.clark@barrons.com

This content was created by Barron's, which is operated by Dow Jones & Co. Barron's is published independently from Dow Jones Newswires and The Wall Street Journal.