Yazılar

DeepSeek claims AI model trained for just $294,000, challenging U.S. rivals

/in /tarafından

Chinese AI developer DeepSeek has disclosed that its reasoning-focused R1 model cost just $294,000 to train—dramatically below the hundreds of millions reportedly spent by U.S. leaders such as OpenAI. The figure, revealed in a Nature article co-authored by founder Liang Wenfeng, is the company’s first public estimate of training costs and is likely to reignite debate over China’s position in the global AI race.

According to the paper, R1 was trained on a cluster of 512 Nvidia H800 chips over 80 hours. DeepSeek acknowledged for the first time that it also owns Nvidia A100 GPUs, which were used in preparatory phases before training shifted to the China-specific H800s. The H800 was designed to comply with U.S. export restrictions that bar Nvidia from selling its more powerful H100 and A100 chips to China.

The cost revelation is striking: OpenAI CEO Sam Altman has said foundational models cost “much more” than $100 million to train, though OpenAI has never published detailed figures. DeepSeek’s claim of drastically lower costs fueled January’s investor selloff in global tech stocks, amid fears it could disrupt the market dominance of Nvidia and other AI giants.

Skepticism remains. U.S. officials have suggested DeepSeek may have obtained H100 chips despite restrictions, while U.S. companies have questioned whether its development relied on model distillation—a technique where one AI model learns from another. DeepSeek has admitted using Meta’s open-source Llama models and said its training data may have included content generated by OpenAI systems, though it insists this was incidental.

DeepSeek defends distillation as an efficient way to cut costs and expand access to AI by reducing the enormous energy and resource demands of large-scale training. Analysts note this could accelerate the spread of competitive AI models outside the U.S., though questions about intellectual property and national security will remain central to the debate.

China investors stay bullish on Cambricon despite index reshuffle

/in /tarafından

Cambricon Technologies, often dubbed China’s Nvidia, faces more than 8 billion yuan ($1.1 billion) in passive outflows due to a quarterly rebalancing of the STAR50 Index, but analysts say investor confidence in the AI chipmaker remains intact.

The company’s stock, which more than doubled in August, exceeded the 10% cap for individual weightings in the tech-heavy index. Though Cambricon shares fell 14% last week on profit-taking and rebalancing fears, they have since rebounded 10%, hovering near record highs.

Valuations are eye-watering—Cambricon trades at 521 times earnings, compared with Nvidia’s multiple of 50—but Beijing’s push for tech self-sufficiency, the DeepSeek AI breakthrough, and large-scale investments by Alibaba, Tencent, and Baidu continue to fuel the rally.

“Maybe some investors will use it as a reason to take profit, but I don’t think that will affect the long-term trend,” said Shihao Li, analyst at CLSA. Gavekal’s Tilly Zhang added that optimism is growing that China’s AI sector has entered a “self-sustaining cycle of rising investment and higher profitability.”

Cambricon’s fundamentals have helped power the surge. First-half revenue jumped to 2.9 billion yuan ($407 million) from just 64.8 million yuan a year earlier, swinging to a 1 billion yuan profit. The company forecasts 5–7 billion yuan in operating revenue for 2025.

Still, risks remain. Some fund managers warn of a speculative bubble, while others argue that growth potential tied to China’s strategic need to replace foreign AI chips may justify lofty valuations.

Broader Chinese markets are riding the same wave. The CSI AI Index is up 60% this year, far outpacing the 15% gain in the CSI300, and the Shanghai Composite has hit levels not seen in a decade.

The spotlight now shifts to whether Cambricon can sustain profitability and meet surging demand for AI chips—critical to maintaining its role as the flagship of China’s AI boom.

OpenAI Explores U.S. Data Center Sites for Stargate AI Project Amid China Competition

/in /tarafından

OpenAI announced on Thursday that it is evaluating several U.S. states as potential locations for data centers supporting its Stargate AI venture. The initiative is positioned as a strategic move to maintain U.S. leadership over China in the AI race.

Chris Lehane, OpenAI’s chief global affairs officer, highlighted the competitive urgency. “With the emergence of DeepSeek, it’s clear this competition is serious. Whoever prevails will shape the future, whether it’s democratic and open AI or authoritarian and autocratic AI,” he said.

Stargate, unveiled by U.S. President Donald Trump last month, represents a private sector AI infrastructure investment of up to $500 billion. Funded by SoftBank, OpenAI, and Oracle, the project has already committed $100 billion for immediate deployment, with further investments to roll out over the next four years.

Sixteen U.S. states have expressed interest in hosting Stargate data centers, with Texas designated as the flagship location. The first data center, under construction in Abilene, Texas, is being developed by startup Crusoe and is expected to be partially operational later this year. Keith Heyde, leading site selection for Stargate, said, “We are looking at five to ten sites for our campus footprint.”

However, the emergence of China’s low-cost DeepSeek AI model has cast doubt on the assumption that large-scale, specialized data centers are essential for AI advancement. DeepSeek researchers claimed they trained their model on less sophisticated chips at a fraction of the cost required by American AI models.

This development sent shockwaves through global markets. Investors reacted by dumping tech stocks, particularly Nvidia, the leading AI chipmaker, wiping out $593 billion of its market value—the largest one-day loss ever recorded on Wall Street.