Nvidia’s plans to stage a comeback in China’s lucrative AI market seem to have hit a brick wall.

According to The Information, Nvidia has instructed its component suppliers to halt production related to its H20 AI chip, a model specifically designed for China, after Washington’s sweeping export restrictions barred Nvidia from selling its most powerful AI processors to the country.

The production pause reportedly follows a strong warning from Beijing, urging Chinese companies to refrain from deploying the H20 over fears of potential security risks. Authorities in Beijing are concerned the chips could contain backdoors that might allow the U.S. government to access sensitive data. In line with its broader push for technological self-sufficiency, China has been encouraging its firms to switch to domestically developed processors instead.

The setback comes barely a month after Nvidia, alongside other U.S. chipmakers, was given approval to sell AI chips tailored to meet U.S. export rules in the Chinese market. The H20, along with other restricted models like the L20, was Nvidia’s attempt to maintain its presence in the world’s second-largest AI economy while sidestepping Washington’s sanctions.

In a statement to TechCrunch, an Nvidia spokesperson pushed back on Beijing’s allegations, saying: “We constantly manage our supply chain to address market conditions. Cybersecurity is critically important to us. NVIDIA does not have ‘backdoors’ in our chips that would give anyone a remote way to access or control them. The market can use the H20 with confidence.”

Beijing’s skepticism and Nvidia’s attempts at reassurance

Beijing’s concern over the H20 chip stems from growing suspicion that foreign-designed hardware could serve as a conduit for espionage. Chinese regulators, citing warnings from cybersecurity experts, suggested that Nvidia’s AI chips might contain hidden functions capable of transmitting sensitive data back to the U.S. or even shutting down systems remotely. This fear was amplified after Washington placed strict export controls on high-performance chips in 2022 and 2023, aimed at curbing China’s progress in advanced AI and military applications.

The H20, a lower-performance version of Nvidia’s flagship AI chips, was specifically tailored to comply with U.S. export restrictions while still allowing the company to maintain a foothold in China’s lucrative AI market. Nvidia marketed the chip as safe, efficient, and within Washington’s legal framework. However, Beijing’s warnings have cast a cloud over its future, highlighting the delicate balance Nvidia must strike between satisfying U.S. regulators and appeasing Chinese authorities.

Last month, China’s Cyberspace Administration reportedly summoned the company over alleged “serious security issues” in its chips, echoing earlier claims from U.S. AI experts that Nvidia’s hardware contained location-tracking capabilities and remote shutdown features. Although Nvidia has strongly denied these allegations, the skepticism persists in Beijing.

Now, Nvidia is signaling a cautious response to China’s concerns by halting production of the H20, while also buying time to address the political and technical fallout. The move underscores the difficult position the chipmaker finds itself in, with Washington’s restrictions limiting what it can sell to China, and Beijing’s mistrust threatening to shut it out of the very market it has been trying to re-enter.

Additionally, the dispute highlights the increasingly fraught tech rivalry between Washington and Beijing, with semiconductors sitting at the heart of the contest. U.S. officials argue that restricting Nvidia’s most advanced chips is critical to preventing China from achieving military and surveillance capabilities powered by cutting-edge AI. Beijing, in turn, has responded by pushing for greater reliance on local firms such as Huawei, which has already begun rolling out competitive AI chips that are being rapidly adopted across China.

Some analysts believe that Nvidia’s halt is significant because the H20 was expected to be part of its main revenue driver in China for 2025. The company generated over $10 billion annually from Chinese sales before the U.S. export bans, making the market one of its most important globally. With Beijing now tightening its stance, Nvidia risks ceding ground to Huawei’s Ascend series of AI chips, which are quickly gaining traction as domestic replacements.

While Nvidia relies heavily on China for its revenue, China, for its part, is accelerating efforts to cut dependence on U.S. suppliers altogether. That has created a window of opportunity for domestic players who are not just filling the void but positioning themselves as long-term competitors.

Meta has struck a high-profile partnership with Midjourney — one of the fastest-growing startups in the AI image and video generation space, marking a step deeper into the artificial intelligence arms race.

The announcement, made Friday by Meta’s Chief AI Officer Alexandr Wang on Threads, signals Meta’s willingness to collaborate rather than compete outright, even as it spends billions building its own AI infrastructure.

Wang explained that the partnership is part of a broader strategy to secure every possible advantage.

“To ensure Meta is able to deliver the best possible products for people it will require taking an all-of-the-above approach,” he wrote. “This means world-class talent, ambitious compute roadmap, and working with the best players across the industry.”

The deal allows Meta to license Midjourney’s highly popular image and video generation technology, which has gained a cult following for producing artistic, realistic visuals. For Meta, it is a chance to rival models such as OpenAI’s Sora, Google’s Veo, and Black Forest Lab’s Flux — all of which are considered industry leaders in generative media. Meta itself has already experimented with tools like Imagine, an image-generation feature baked into Facebook, Instagram, and Messenger, as well as Movie Gen, which allows users to create videos from text prompts. But licensing Midjourney’s models could supercharge those efforts.

This partnership adds to Meta’s growing list of aggressive AI moves in 2024. Earlier this year, CEO Mark Zuckerberg authorized a massive hiring spree for AI talent, dangling compensation packages as high as $100 million to lure researchers from rivals. The company also poured $14 billion into Scale AI, a San Francisco-based data-labeling firm, and snapped up Play AI, a voice startup, to round out its AI portfolio.

Zuckerberg’s expansionist streak has extended even further. He has held talks with other AI labs about potential acquisitions. He was reportedly approached at one point by Elon Musk about joining his $97 billion takeover bid for OpenAI — a bid Musk ultimately failed to complete, and which OpenAI denied. Although Meta declined to join, the conversation underscored just how central AI has become to the future of the world’s largest tech companies.

While the financial terms of the Midjourney deal remain undisclosed, the startup is already a remarkable outlier in the AI space. Its CEO, David Holz, said in a post on X that Midjourney remains independent and has never taken on outside investment. That independence has made the company stand out in a market dominated by venture-backed AI labs burning through capital. At one point, Meta explored acquiring Midjourney outright, according to Upstarts Media, but Holz has so far resisted selling.

Founded in 2022, Midjourney quickly became a revenue powerhouse, reportedly on track to generate $200 million by 2023 from subscriptions that start at $10 per month and go as high as $120 for premium access. In June, the company released its first video-generation model, V1, signaling its ambitions beyond still imagery.

But in June, Disney and Universal sued Midjourney, alleging that its models were trained on copyrighted material — a legal battle that mirrors broader disputes facing nearly every leading AI company, including Meta. So far, courts have largely sided with tech firms, ruling that training on publicly available data falls under fair use, but the litigation remains a dark cloud over the industry.

The stakes are financial as much as technological for Meta. Its $14 billion bet on AI and the billions more it is spending on infrastructure and talent underscore a gamble that generative media will be a central part of its future products. Analysts say the Midjourney licensing deal will save Meta both time and resources in building image models from scratch, while providing a bridge to compete with OpenAI, Google, and others.

The deal also points to a broader trend: partnership is becoming as important as algorithms in the AI race. Meta announced a hiring freeze earlier this week, suggesting that its partnership with Midjourney is not just about artful images — it’s about shoring up its position in a rapidly consolidating industry. Meta platforms reportedly struck a six-year cloud computing deal with Google worth more than $10 billion