NVIDIA, AMD may soon start selling new AI chips in China to comply with US restrictions
NVIDIA plans to sell a stripped-down AI GPU, code-named "B20," while AMD is looking to target AI workload needs with its new Radeon AI PRO R9700 workstation GPU, Digitimes reported, adding that the companies will likely start selling these AI chips in China from July.
Earlier this week, Reuters reported that NVIDIA is working on a new budget AI chip built on its Blackwell architecture for China that is expected to be priced at $6,500-$8,000. In comparison, the company sells its H20 GPUs for $10,000-$12,000 each.
NVIDIA on Wednesday said it had incurred a $4.5 billion charge in Q1 due to licensing requirements impacting its ability to sell its H20 AI chip to companies in China, and it couldn't ship an additional $2.5 billion of H20 chips in the quarter due to the restrictions. The company forecast that licensing requirements would result in an $8 billion hit to the company's revenue in Q2.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Yahoo
2 hours ago
- Yahoo
US reportedly plans to curb sales of AI GPUs to Malaysia and Thailand to prevent smuggling to China
When you buy through links on our articles, Future and its syndication partners may earn a commission. The U.S. government is preparing a new set of export rules that would tighten control over the exports of advanced Nvidia AI GPUs to Malaysia and Thailand, in a bid to prevent the re-export of these components to China amid existing bans, according to Bloomberg. A preliminary version of the new export rule, reported by Bloomberg, states that the U.S. Commerce Department would require companies to obtain a U.S. government export license before sending AI GPUs to the two Southeast Asian nations. The plan has not been finalized and may change, yet it may represent another step towards limiting Chinese entities' access to high-performance Nvidia AI GPUs. Malaysia and Thailand are not major suspected hubs for the smuggling of Nvidia's GPUs, unlike Singapore, which is officially listed as one of Nvidia's primary sources of revenue, raising questions about whether the products sold to Singapore-based entities eventually end up in China. Indeed, Nvidia denies that its AI GPUs formally sold to Singapore-based entities could end up in China, arguing that they are sold to entities officially based in Singapore, but they are destined elsewhere. Nonetheless, it is widely believed that Singapore is a hub for smuggling high-end Nvidia GPUs to China and other sanctioned countries. Yet, Malaysia seems like a different thing. The country is not listed as one of Nvidia's primary sources of revenue, so we do not know how much the company earns selling products to the country. However, Malaysia has emerged as a significant destination for computing equipment and components — including CPUs and GPUs essential for AI applications — sourced from Taiwan in recent quarters. Moreover, Malaysia is reportedly used to avoid U.S. import tariffs for China-made goods, which might be a concern for the U.S. Commerce Department. Thailand is among the countries suspected of being a hub for smuggling Nvidia's AI GPUs to China. However, without formal data, we can only wonder whether the suspicion has merit. Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.
Business Insider
4 hours ago
- Business Insider
NVDA, TSLA, PLTR: Retail Investors Traded a Record $6.6 Trillion of Stocks in Year's First Half
Uncertainty, volatility and relief led retail investors around the world to trade a record $6.6 trillion worth of stocks during the first half of 2025. Don't Miss TipRanks' Half-Year Sale Take advantage of TipRanks Premium at 50% off! Unlock powerful investing tools, advanced data, and expert analyst insights to help you invest with confidence. Make smarter investment decisions with TipRanks' Smart Investor Picks, delivered to your inbox every week. Tariffs, market volatility, ongoing wars in Europe and the Middle East, and political tensions within the U.S. all conspired to get investors to buy and sell a record amount of equities in the year's first half. According to data from Nasdaq (NDAQ), individual retail investors cumulatively bought $3.4 trillion worth of stocks over the first six months of 2025. At the same time, the retail crowd sold $3.2 trillion worth of stocks, bringing the total traded to $6.6 trillion, a previously unheard-of sum. Among the most widely bought and sold securities between January and June were the stocks of chipmaker Nvidia (NVDA), electric vehicle maker Tesla (TSLA), and data analytics company Palantir (PLTR). Bear to Bull Even more dramatic, most of the buying and selling that occurred in the year's first half happened in April and May after the tariff announcements from U.S. President Donald Trump rattled global markets and ignited fears of a global trade war and economic recession. Those fears have since abated as the Trump administration has struck trade deals with countries such as the United Kingdom and China, and backed down on many of its retaliatory tariff threats. Nasdaq says that some retail investors found this past spring 'the toughest investment climate' they ever experienced. Yet the initial concerns have now given way to bullish sentiment as tariff and trade war fears subside. Is NVDA Stock a Buy? The stock of Nvidia has a consensus Strong Buy rating among 40 Wall Street analysts. That rating is based on 35 Buy, four Hold, and one Sell recommendations assigned in the last three months. The average NVDA price target of $175.69 implies 10.26% upside from current levels.
Forbes
5 hours ago
- Forbes
The Blackwell Successor: What Nvidia's Rubin Means For Your Phone
23 February 2024, USA, Santa Clara: The logo of the chip company Nvidia can be seen at its ... More headquarters in Silicon Valley. The company's market capitalization exceeded the two trillion dollar mark for the first time at the close of trading on Friday. Photo: Andrej Sokolow/dpa (Photo by Andrej Sokolow/picture alliance via Getty Images) It doesn't take too much digging to see that the hardware world is heating up. One company remains dominant in some key ways. Nvidia, led by Jensen Huang, is a big new star in the tech world, and recently became the largest company by market cap on the U.S. stock market, eclipsing both Apple and Microsoft. How did this happen? Nvidia has become adept at setting the trends when it comes to hardware. Its GPUs are in the biggest data centers, like XAI's Colossus and other projects operating at enormous scales, as when Musk kept capriciously adding 100,000 more GPUs to the aforementioned project. Nvidia has also garnered a large share of the foundry operations of TSMC, the Taiwanese company that primarily manufacturers chips, providing a wide range of top clients with processing power. Bold New Architectures Along with its track record in production and market dominance, Nvidia has also been announcing a lot of new models for its processors. It wasn't too long ago we were talking about Nvidia Blackwell, named after David Blackwell, and the Grace Blackwell superchip, that had groundbreaking capacity and a customer base around the world. At a recent tech event in 2024, though, Huang announced the company's intention to start offering a new kind of microarchitecture called 'Rubin,' named after physicist Vera Rubin. It would have a GPU called Rubin and a CPU called Vera, for a powerful one-two punch in terms of clock speed. In addition, there will be an alternative called Rubin Ultra developed after that, and a hardware model called Feynman (after physicist Richard Feynman) in 2028. As for memory bandwidth, Rubin is expected to have around 900 GB per second, with Rubin Ultra operating around 1200 GB per second. Feynman is expected to operate around 1000 GB per second, being optimized for certain kinds of data handling and precise mathematical operations. There's also a Rubin AI platform, the company's ecosystem, set up with Rubin chips. This will have ultra fast HBM4 memory, with NV Link 6 switches, to deliver up to 3600 GB per second in bandwidth. It's all made for data centers; it's made for AI. Putting LLMs on Devices Now, let's talk about the market effect of Nvidia's Rubin. Experts point out that it's not necessarily that you'll get an entire model working entirely off-line – but the architectures themselves will drive more edge computing models for robust systems. That means the AI entities that you do have on your local devices will become more competent, more capable, and able to reason at a more profound depth. So we are likely to soon see the effect of these hardware revolutions in the consumer world with our wearables and mobile device devices. Everything from fitness and education to retail, medicine and law is getting re-Imagined with powerful LLMs that are portable and decentralized from big data centers. 'These devices encompass a wide range, from powerful edge servers to resource-constrained IoT sensors, and include familiar examples like smartphones, smart home appliances, autonomous vehicles, and even industrial robots,' writes Bhavishya Pandit at Datacamp, explaining some of the values of edge computing and its potential effects on our tech world. Presumably, there's still important work going on in those big data centers, but increasingly, it will be relegated to research and esoteric IT, while we'll continue to see more implementation at the user device level. We'll probably see more impact on the average user, too. That may come in the form of higher user adoption for applications, or more respect for the powers that AI has, or both. How do you see the future of AI when everyone's carrying powerful artificial intelligence engines in their pockets? Let's keep an eye on this throughout the rest of 2025 and in 2026 as Rubin first comes to the hardware market.
