Yazılar

Nvidia’s New AI Chips Slash Training Times for Massive AI Models

/in /tarafından

Nvidia’s latest generation of AI chips is making significant advances in training some of the world’s largest artificial intelligence systems, according to new benchmark data released on Wednesday by MLCommons, a nonprofit organization that tracks AI system performance.

The results show a dramatic drop in the number of chips required to train large language models (LLMs), highlighting Nvidia’s growing technological lead in this critical area of AI development. While much of the financial market’s current focus is on the booming sector of AI inference—where AI models answer user queries—training remains a core competitive battleground, especially for developing next-generation models with trillions of parameters.

Blackwell Chips Outperform Previous Generations

Nvidia’s new Blackwell chips demonstrated superior performance over its previous Hopper generation. In tests involving Meta Platforms’ open-source Llama 3.1 405B model, which is complex enough to simulate some of the most demanding AI training workloads, Nvidia’s Blackwell chips completed training tasks with more than double the speed per chip compared to Hopper.

In one benchmark, a system using 2,496 Blackwell chips completed the training run in just 27 minutes. By comparison, even though more than three times as many Hopper chips were used in previous tests, they only achieved faster results due to sheer scale rather than efficiency.

Nvidia and its partners were the only ones to submit data for models of this size, giving Nvidia a clear demonstration of its leadership in training capabilities for multi-trillion parameter models.

Changing Industry Trends in AI Training

Chetan Kapoor, chief product officer of CoreWeave, which collaborated with Nvidia on the results, noted that AI companies are moving away from building vast, homogenous data centers with 100,000 or more identical chips. Instead, they are increasingly assembling smaller, specialized subsystems that handle different aspects of the training process. This modular approach allows companies to speed up training times and manage extremely large model sizes more efficiently.

“Using a methodology like that, they’re able to continue to accelerate or reduce the time to train some of these crazy, multi-trillion parameter model sizes,” Kapoor explained at a press briefing.

Global Competition Also Heating Up

While Nvidia maintains a dominant position, competitors around the world are also pushing for breakthroughs. For example, China’s DeepSeek has recently claimed it can create competitive chatbots while using far fewer chips than many U.S. rivals, adding to the growing international race for AI supremacy.

MLCommons’ report also included results from Advanced Micro Devices (AMD) and others, though Nvidia’s Blackwell system stood out in the training category.

Baidu Says Homegrown Tech Shields AI Ambitions from U.S. Chip Curbs

/in /tarafından

Chinese tech giant Baidu asserted on Wednesday that its artificial intelligence (AI) development remains largely insulated from recent U.S. semiconductor export restrictions, thanks to an expanding domestic supply of chips and software. The company also reported stronger-than-expected Q1 financial results, fueled by growth in its AI cloud segment.

“Domestically developed chips and increasingly efficient homegrown software will form a strong foundation for long-term innovation in China’s AI ecosystem,” said Shen Dou, Baidu’s Vice President, during a conference call with analysts.

The statement follows the latest U.S. curbs on advanced chips — including Nvidia’s H20, a product tailored for the Chinese market — which officially took effect last month. Baidu’s confidence mirrors that of rival Tencent, which recently cited existing chip stockpiles as a buffer against Washington’s tightening export controls.

Baidu’s first-quarter revenue rose 3% year-over-year to 32.45 billion yuan ($4.5 billion), surpassing analysts’ estimates of 30.9 billion yuan, according to LSEG. The company’s non-online marketing revenue, primarily driven by its AI cloud business, jumped 40% to 9.4 billion yuan, highlighting Baidu’s accelerating pivot away from its legacy ad-based search engine model.

While revenue from its online marketing segment fell 6% to 17.31 billion yuan — slightly below forecasts — Baidu posted a robust profit of 21.59 yuan per American Depositary Share, up from 14.91 yuan a year earlier.

Baidu has made aggressive moves in the generative AI space since becoming one of the first Chinese firms to launch a ChatGPT-style chatbot in early 2023. However, its flagship Ernie model now faces stiff competition from fast-rising domestic players like DeepSeek.

In response, Baidu scrapped subscription fees for premium AI chatbot services in April and launched enhanced models including Ernie X1 and Ernie 4.5, later upgrading both to “Turbo” versions. The company’s AI ambitions are powered by its self-developed P800 Kunlun chips, with a 30,000-chip cluster said to be capable of training DeepSeek-scale models.

Despite the upbeat earnings and AI momentum, Baidu’s U.S.-listed shares were slightly down 0.3% in Wednesday morning trading.

AI Leaders Urge U.S. to Boost Exports and Infrastructure to Stay Ahead of China

/in /tarafından

Top executives from OpenAI, Microsoft, and AMD warned U.S. lawmakers on Thursday that the country risks losing its lead in artificial intelligence to China unless it expands infrastructure, loosens AI chip export restrictions, and strengthens workforce training. Their testimony before the U.S. Senate Commerce Committee, chaired by Senator Ted Cruz, emphasized the urgent need for pro-growth AI policies to counter China’s rapid advancements.

The call to action follows China’s DeepSeek AI breakthrough last year and Huawei’s rollout of advanced AI chips, both of which have shaken Washington’s confidence in maintaining AI dominance.

“The number-one factor that will define whether the U.S. or China wins this race is whose technology is most broadly adopted in the rest of the world,” said Brad Smith, President of Microsoft. He added that Microsoft has banned internal use of DeepSeek due to data privacy and propaganda concerns.
“The lesson from Huawei and 5G is that whoever gets there first will be difficult to supplant.”

Key Takeaways from the Senate Hearing:

  • OpenAI CEO Sam Altman emphasized the need for massive infrastructure investment, including data centers and power generation, to fuel AI’s growth.

  • AMD CEO Lisa Su highlighted the importance of maintaining competitiveness in AI chip design while also ensuring export flexibility.

  • Smith called for broader AI education, R&D funding, and skilled labor development, including more electricians for AI facilities.

The tech industry is pushing back against Biden-era AI export rules that aimed to limit China’s access to powerful AI chips. In response, the Trump administration is preparing to rescind those curbs and replace them with a new framework — a move praised by Cruz, Altman, and Su during the session.

“The Biden administration’s misguided midnight AI diffusion rule on chips and model weights would have crippled American tech companies’ ability to sell AI to the world,” Cruz said.

China’s DeepSeek, based in Hangzhou, made waves by launching a powerful, cost-effective AI model competitive with OpenAI and Meta — a move that intensified pressure on U.S. lawmakers to act quickly.

Meanwhile, Huawei is preparing to mass-ship advanced AI chips to Chinese customers despite ongoing U.S. trade restrictions.

With national security, economic leadership, and technological supremacy at stake, AI executives stressed that global market penetration—not just technical capability—will determine who wins the AI race.