Chinese researchers backed by a Hangzhou-based hedge fund recently released a new version of a large language model (LLM) called DeepSeek-R1 that rivals the capabilities of the most advanced U.S.-built products but reportedly does so with fewer computing resources and at much lower cost.

High Flyer, the hedge fund that backs DeepSeek, said that the model nearly matches the performance of LLMs built by U.S. firms like OpenAI, Google and Meta, but does so using only about 2,000 older generation computer chips manufactured by U.S.-based industry leader Nvidia while costing only about $6 million worth of computing power to train.

By comparison, Meta’s AI system, Llama, uses about 16,000 chips, and reportedly costs Meta vastly more money to train.

Open-source model

The apparent advance in Chinese AI capabilities comes after years of efforts by the U.S. government to restrict China’s access to advanced semiconductors and the equipment used to manufacture them. Over the past two years, under President Joe Biden, the U.S. put multiple export control measures in place with the specific aim of throttling China’s progress on AI development.

DeepSeek appears to have innovated its way to some of its success, developing new and more efficient algorithms that allow the chips in the system to communicate with each other more effectively, thereby improving performance.

At least some of what DeepSeek R1’s developers did to improve its performance is visible to observers outside the company, because the model is open source, meaning that the algorithms it uses to answer queries are public.

Market reaction

The news about DeepSeek’s capabilities sparked a broad sell-off of technology stocks on U.S. markets on Monday, as investors began to question whether U.S. companies’ well-publicized plans to invest hundreds of billions of dollars in AI data centers and other infrastructure would preserve their dominance in the field. When the markets closed on Monday, the tech-heavy Nasdaq index was down by 3.1%, and Nvidia’s share price had plummeted by nearly 17%.

However, not all AI experts believe the markets’ reaction to the release of DeepSeek R1 is justified, or that the claims about the model’s development should be taken at face value.

Mel Morris, CEO of U.K.-based Corpora.ai, an AI research engine, told VOA that while DeepSeek is an impressive piece of technology, he believes the market reaction has been excessive and that more information is needed to accurately judge the impact DeepSeek will have on the AI market.

“There’s always an overreaction to things, and there is today, so let’s just step back and analyze what we’re seeing here,” Morris said. “Firstly, we have no real understanding of exactly what the cost was or the time scale involved in building this product. We just don’t know. … They claim that it’s significantly cheaper and more efficient, but we have no proof of that.”

Morris said that while DeepSeek’s performance may be comparable to that of OpenAI products, “I’ve not seen anything yet that convinces me that they’ve actually cracked the quantum step in the cost of operating these sorts of models.”

Doubts about origins

Lennart Heim, a data scientist with the RAND Corporation, told VOA that while it is plain that DeepSeek R1 benefits from innovative algorithms that boost its performance, he agreed that the general public actually knows relatively little about how the underlying technology was developed.

Heim said that it is unclear whether the $6 million training cost cited by High Flyer actually covers the whole of the company’s expenditures — including personnel, training data costs and other factors — or is just an estimate of what a final training “run” would have cost in terms of raw computing power. If the latter, Heim said, the figure is comparable to the costs incurred by better U.S. models.

He also questioned the assertion that DeepSeek was developed with only 2,000 chips. In a blog post written over the weekend, he noted that the company is believed to have existing operations with tens of thousands of Nvidia chips that could have been used to do the work necessary to develop a model that is capable of running on just 2,000.

“This extensive compute access was likely crucial for developing their efficiency techniques through trial and error and for serving their models to customers,” he wrote.

He also pointed out that the company’s decision to release version R1 of its LLM last week — on the heels of the inauguration of a new U.S. president — appeared political in nature. He said that it was “clearly intended to rattle the public’s confidence in the United States’ AI leadership during a pivotal moment in U.S. policy.”

Dean W. Ball, a research fellow at George Mason University’s Mercatus Center, was also cautious about declaring that DeepSeek R1 has somehow upended the AI landscape.

“I think Silicon Valley and Wall Street are overreacting to some extent,” he told VOA. “But at the end of the day, R1 means that the competition between the U.S. and China is likely to remain fierce, and that we need to take it seriously.”

Export control debate

The apparent success of DeepSeek has been used as evidence by some experts to suggest that the export controls put in place under the Biden administration may not have had the intended effects.

“At a minimum, this suggests that U.S. approaches to AI and export controls may not be as effective as proponents claim,” Paul Triolo, a partner with DGA-Albright Stonebridge Group, told VOA.

“The availability of very good but not cutting-edge GPUs — for example, that a company like DeepSeek can optimize for specific training and inference workloads — suggests that the focus of export controls on the most advanced hardware and models may be misplaced,” Triolo said. “That said, it remains unclear how DeepSeek will be able to keep pace with global leaders such as OpenAI, Google, Anthropic, Mistral, Meta and others that will continue to have access to the best hardware systems.”

Other experts, however, argued that export controls have simply not been in place long enough to show results.

Sam Bresnick, a research fellow at Georgetown’s University’s Center for Security and Emerging Technology told VOA that it would be “very premature” to call the measures a failure.

“The CEO of DeepSeek has gone on record saying the biggest constraint they face is access to high-level compute resources,” Bresnick said. “If [DeepSeek] had as much compute at their fingertips as Google, Microsoft, OpenAI, etc, there would be a significant boost in their performance. So … I don’t think that DeepSeek is the smoking gun that some people are claiming it is [to show that export controls] do not work.”

Bresnick noted that the toughest export controls were imposed in only 2023, meaning that their effects may just be starting to be felt. He said that the real test of their effectiveness will be whether U.S. firms are able to continue to outpace China in coming years.
…

The multibillion-dollar Stargate Project announced by U.S. President Donald Trump will focus on building data centers with the goal of turning the U.S. into a computing power empire, according to experts.

Some believe the significant boost in U.S. computational capabilities will widen the gap with China in artificial intelligence.

“And this is an industrial buildout that, at least right now, China really is not in a position to do because of the [semiconductor] export controls that the United States is placing,” said Dean W. Ball, a research fellow at George Mason University’s Mercatus Center. However, there are signs that China is catching up with U.S. companies in key AI metrics by relying on open-source software.

Click here for the full report in Mandarin.
…

New York — Wall Street is tumbling Monday on fears the big U.S. companies that have feasted on the artificial-intelligence frenzy are under threat from a competitor in China that can do similar things for much cheaper.

The S&P 500 was down 1.9% in early trading. Big Tech stocks that have been the market’s biggest stars took the heaviest losses, with Nvidia down 11.5%, and they dragged the Nasdaq composite down 3.2%. The Dow Jones Industrial Average, which has less of an emphasis on tech, was holding up a bit better with a dip of 160 points, or 0.4%, as of 9:35 a.m. Eastern time.

The shock to financial markets came from China, where a company called DeepSeek said it had developed a large language model that can compete with U.S. giants but at a fraction of the cost. DeepSeek’s app had already hit the top of Apple’s App Store chart by early Monday morning, and analysts said such a feat would be particularly impressive given how the U.S. government has restricted Chinese access to top AI chips.

Skepticism, though, remains about how much DeepSeek’s announcement will ultimately shake the AI supply chain, from the chip makers making semiconductors to the utilities hoping to electrify vast data centers running those chips.

“It remains to be seen if DeepSeek found a way to work around these chip restrictions rules and what chips they ultimately used as there will be many skeptics around this issue given the information is coming from China,” according to Dan Ives, an analyst with Wedbush Securities.

DeepSeek’s disruption nevertheless rocked stock markets worldwide.

In Amsterdam, Dutch chip company ASML slid 8.9%. In Tokyo, Japan’s Softbank Group Corp. lost 8.3% and is nearly back to where it was before spurting on an announcement that it was joining a partnership trumpeted by the White House that would invest up to $500 billion in AI infrastructure.

And on Wall Street, shares of Constellation Energy sank 16.9%. The company has said it would restart the shuttered Three Mile Island nuclear power plant to supply power for Microsoft’s data centers.

All the worries sent a gauge of nervousness among investors holding U.S. stocks toward its biggest jump since August. They also sent investors toward bonds, which can be safer investments than any stock. The rush sent the yield of the 10-year Treasury down to 4.53% from 4.62% late Friday.

It’s a sharp turnaround for the AI winners, which had soared in recent years on hopes that all the investment pouring into the industry would lead to a possible remaking of the global economy.

Nvidia’s stock had soared from less than $20 to more than $140 in less than two years before Monday’s drop, for example.

Other Big Tech companies had also joined in the frenzy, and their stock prices had benefited too. It was just on Friday that Meta Platforms CEO Mark Zuckerberg was saying he expects to invest up to $65 billion this year, while talking up a massive data center it would build in Manhattan.

In stock markets abroad, movements for indexes across Europe and Asia weren’t as forceful as for the big U.S. tech stocks. France’s CAC 40 fell 0.6%, and Germany’s DAX lost 0.8%.

In Asia, stocks edged 0.1% lower in Shanghai after a survey of manufacturers showed export orders in China dropping to a five-month low.

The Federal Reserve holds its latest policy meeting later this week. Traders don’t expect recent weak data to push the Fed to cut its main interest rate. They’re virtually certain the central bank will hold steady, according to data from CME Group.
…

Plastic waste accounts for 10 to 12 percent of all solid waste in Kenya, according to the United Nations Environmental Program. A Kenyan tech company is using plastic waste to print 3D models that help college students with their learning while reducing damage to the environment. Mohammed Yusuf reports from Nairobi.
…

ABOARD AIR FORCE ONE — U.S. President Donald Trump said on Saturday he was in talks with multiple people over buying TikTok and would likely have a decision on the popular app’s future in the next 30 days.

“I have spoken to many people about TikTok and there is great interest in TikTok,” Trump told reporters on Air Force One during a flight to Florida.

Earlier in the day, Reuters reported two people with knowledge of the discussions said Trump’s administration is working on a plan to save TikTok that involves tapping software company Oracle and a group of outside investors to effectively take control of the app’s operations.

Under the deal being negotiated by the White House, TikTok’s China-based owner, ByteDance, would retain a stake in the company, but data collection and software updates would be overseen by Oracle, which already provides the foundation of TikTok’s Web infrastructure, one of the sources told Reuters.

However, in his comments to reporters on the flight, Trump said he had not spoken to Oracle’s Larry Ellison about buying the app.

Asked if he was putting together a deal with Oracle and other investors to save TikTok, Trump said: “No, not with Oracle. Numerous people are talking to me, very substantial people, about buying it and I will make that decision probably over the next 30 days. Congress has given 90 days. If we can save TikTok, I think it would be a good thing.”

The sources did say the terms of any potential deal with Oracle were fluid and likely to change. One source said the full scope of the discussions was not yet set and could include the U.S. operations as well as other regions.

National Public Radio on Saturday reported the deal talks for TikTok’s global operations, citing two people with knowledge of the negotiations. Oracle had no immediate comment.

The deal being negotiated anticipates participation from ByteDance’s current U.S. investors, according to the sources. Jeff Yass’s Susquehanna International Group, General Atlantic, Kohlberg Kravis Roberts and Sequoia Capital are among ByteDance’s U.S. backers.

Representatives for TikTok, ByteDance investors General Atlantic, KKR, Sequoia and Susquehanna could not immediately be reached for comment.

Others vying to acquire TikTok, including the investor group led by billionaire Frank McCourt and another involving Jimmy Donaldson, better known as the YouTube star Mr. Beast, are not part of the Oracle negotiation, one of the sources said.

Oracle responsible

Under the terms of the deal, Oracle would be responsible for addressing national security issues. TikTok initially struck a deal with Oracle in 2022 to store U.S. users’ information to alleviate Washington’s worries about Chinese government interference.

TikTok’s management would remain in place, to operate the short video app, according to one of the sources.

The app, which is used by 170 million Americans, was taken offline temporarily for users shortly before a law that said it must be sold by ByteDance on national security grounds, or be banned, took effect on Jan. 19.

Trump, after taking office a day later, signed an executive order seeking to delay by 75 days the enforcement of the law that was put in place after U.S. officials warned that under ByteDance, there was a risk of Americans’ data being misused.

Officials from Oracle and the White House held a meeting on Friday about a potential deal, and another meeting has been scheduled for next week, NPR reported.

Oracle was interested in a TikTok stake “in the tens of billions,” but the rest of the deal is in flux, the NPR report cited the source as saying.

Trump has said he “would like the United States to have a 50% ownership position in a joint venture” in TikTok.

NPR cited another source as saying that appeasing Congress is seen as a key hurdle by the White House.

Free speech advocates have opposed TikTok’s ban under a law passed by the U.S. Congress and signed by former President Joe Biden.

The company has said U.S. officials have misstated its ties to China, arguing its content recommendation engine and user data are stored in the United States on cloud servers operated by Oracle while content moderation decisions that affect American users are also made in the U.S. 
…

HARRISBURG, PENNSYLVANIA — Looking for a quick fix for their fast-growing electricity diets, tech giants are increasingly looking to strike deals with power plant owners to plug in directly, avoiding a potentially longer and more expensive process of hooking into a fraying electric grid that serves everyone else. 

It’s raising questions over whether diverting power to higher-paying customers will leave enough for others and whether it’s fair to excuse big power users from paying for the grid. Federal regulators are trying to figure out what to do about it, and quickly. 

Front and center is the data center that Amazon’s cloud computing subsidiary, Amazon Web Services, is building next to the Susquehanna nuclear plant in eastern Pennsylvania. 

The arrangement between the plant’s owners and AWS — called a “behind the meter” connection — is the first to come before the Federal Energy Regulatory Commission. For now, FERC has rejected a deal that could eventually send 960 megawatts — about 40% of the plant’s capacity — to the data center. That’s enough to power more than 500,000 homes. 

That leaves the deal and others that likely would follow in limbo. It’s not clear when FERC, which blocked the deal on procedural grounds, will take up the matter again or how the change in presidential administrations might affect things. 

“The companies, they’re very frustrated because they have a business opportunity now that’s really big,” said Bill Green, the director of the MIT Energy Initiative. “And if they’re delayed five years in the queue, for example — I don’t know if it would be five years, but years anyway — they might completely miss the business opportunity.” 

Driving demand for energy-hungry data centers 

The rapid growth of cloud computing and artificial intelligence has fueled demand for data centers that need power to run servers, storage systems, networking equipment and cooling systems. 

That’s spurred proposals to bring nuclear power plants out of retirement, develop small modular nuclear reactors, and build utility-scale renewable installations or new natural gas plants. In December, California-based Oklo announced an agreement to provide 12 gigawatts to data center developer Switch from small nuclear reactors powered by nuclear waste. 

Federal officials say fast development of data centers is vital to the economy and national security, including to keep pace with China in the artificial intelligence race. 

For AWS, the deal with Susquehanna satisfies its need for reliable power that meets its internal requirements for sources that don’t emit planet-warming greenhouse gases, such as coal, oil or gas-fueled plants. 

Big Tech also wants to stand up their centers fast. But tech’s voracious appetite for energy comes at a time when the power supply is already strained by efforts to shift away from planet-warming fossil fuels. 

They can build data centers in a couple years, said Aaron Tinjum of the Data Center Coalition. But in some areas, getting connected to the congested electricity grid can take four years, and sometimes much more, he said. 

Plugging directly into a power plant would take years off their development timelines. 

What’s in it for power providers 

In theory, the AWS deal would let Susquehanna sell power for more than they get by selling into the grid. Talen Energy, Susquehanna’s majority owner, projected the deal would bring as much as $140 million in electricity sales in 2028, though it didn’t disclose exactly how much AWS will pay for the power. 

The profit potential is one that other nuclear plant operators are embracing after years of financial distress and frustration with how they are paid in the broader electricity markets. Many say they’ve been forced to compete in some markets flooded with cheap natural gas and state-subsidized solar and wind energy. 

Power plant owners also say the arrangement benefits the wider public, by bypassing the costly buildout of long power lines and leaving more transmission capacity on the grid for everyone else. 

FERC’s big decision 

A favorable ruling from FERC could open the door to many more huge data centers and other massive power users like hydrogen plants and bitcoin miners, analysts say. 

FERC’s 2-1 rejection in November was procedural. Recent comments by commissioners suggest they weren’t ready to decide how to regulate such a novel matter without more study. 

In the meantime, the agency is hearing arguments for and against the Susquehanna-AWS deal. 

Monitoring Analytics, the market watchdog in the mid-Atlantic grid, wrote in a filing to FERC that the impact would be “extreme” if the Susquehanna-AWS model were extended to all nuclear power plants in the territory. 

Energy prices would increase significantly and there’s no explanation for how rising demand for power will be met even before big power plants drop out of the supply mix, it said. 

Separately, two electric utility owners — which make money in deregulated states from building out the grid and delivering power — have protested that the Susquehanna-AWS arrangement amounts to freeloading off a grid that ordinary customers pay to build and maintain. Chicago-based Exelon and Columbus, Ohio-based American Electric Power say the Susquehanna-AWS arrangement would allow AWS to avoid $140 million a year that it would otherwise owe. 

Susquehanna’s owners say the data center won’t be on the grid and question why it should have to pay to maintain it. But critics contend that the power plant itself is benefiting from taxpayer subsidies and ratepayer-subsidized services — and shouldn’t be able to strike deals with private customers that could increase costs for others. 

FERC’s decision will have “massive repercussions for the entire country” because it will set a precedent for how FERC and grid operators will handle the waiting avalanche of similar requests from data center companies and nuclear plants, said Jackson Morris of the Natural Resources Defense Council. 

Stacey Burbure, a vice president for American Electric Power, told FERC at a hearing in November that it needs to move quickly. 

“The timing of this issue is before us,” she said, “and if we take our typical five years to get this perfect, it will be too late.” 
…