CNBC
"Ex-senior Federal Reserve advisor charged with economic espionage to benefit China"
Dan Mangan @_DanMangan
Published Fri, Jan 31 2025
Key Points
* A former senior advisor for the Federal Reserve was arrested on charges that he conspired to steal Fed trade secrets for the benefit of China.
* The data that advisor John Harold Rogers shared with his co-conspirators could allow China to manipulate the U.S. market “in a manner similar to insider trading,” the Department of Justice said.
* The information Rogers allegedly sought as part of the scheme included “sensitive information about the Federal Open Market Committee,” whose interest rate decisions have broad market effects.
A former senior advisor for the Federal Reserve was arrested Friday on charges that he conspired to steal Fed trade secrets for the benefit of China.
The data that the advisor John Harold Rogers allegedly shared with his co-conspirators could allow China to manipulate the U.S. market “in a manner similar to insider trading,” according to the U.S. Attorney’s Office in Washington, D.C.
“Gaining advance knowledge of U.S. economic policy, including advance knowledge of changes to the federal funds rate, could provide China with an advantage when selling or buying U.S. bonds or securities,” the office said.
Prosecutors noted that China holds about $816 billion of U.S. government debt.
Rogers’ two alleged co-conspirators were members of China’s intelligence and security apparatus who posed as graduate students at a university in that country, prosecutors said.
Those conspirators allegedly gave him gifts, arranged and paid for a beach vacation for Rogers, and also arranged and paid for his airfare, lodging and meals during his visits to China, where he worked as a part-time professor at Fudan University in Shanghai after retiring from the Fed.
Rogers, a 63-year-old Vienna, Virginia, resident who has a Ph.D in economics, was indicted in D.C. federal court on charges of conspiracy to commit economic espionage and making false statements.
He faces a maximum possible sentence of 15 years in prison if convicted of the espionage charge.
Rogers appeared Friday in court, where a magistrate judge at the request of prosecutors ordered him temporarily detained without bond pending an arraignment and detention hearing scheduled for Tuesday.
A Federal Reserve spokesman declined to comment to CNBC when asked about Rogers’ arrest.
Jonathan Gitlen, a lawyer for Rogers, in a statement to CNBC said, “Dr. Rogers denies the allegations as set forth in the indictment."
"Dr. Rogers will have a further comment at a later date.”
The case was announced on the same day that the White House said that President Donald Trump would impose tariffs on China, as well as on Canada and Mexico, on Saturday.
Rogers worked as a senior advisor in the international finance division of the Federal Reserve Board of Governors from 2010 until 2021, U.S. Attorney’s Office said.
As part of that job, he “was entrusted with confidential FRB information,” according to prosecutors.
Rogers since 2018 allegedly exploited his employment “by soliciting trade-secret information regarding proprietary economic data sets, deliberations about tariffs targeting China, briefing books for designated governors, and sensitive information about Federal Open Market Committee ... deliberations and forthcoming announcements,” the U.S. Attorney’s Office said.
The FOMC is responsible for setting the federal funds rate — the interest rate banks charge one another for short-term loans.
FOMC decisions on that rate can significantly affect U.S. financial markets.
The indictment accuses him of passing that information electronically from his personal email, in violation of Fed policy, or printing it out prior to traveling to China to meet with co-conspirators.
“Under the guise of teaching ‘classes,’ Rogers met with his co-conspirators in hotel rooms in China where he conveyed sensitive, trade-secret information that belonged to the FRB and the FOMC,” the U.S. Attorney’s Office said.
Rogers in 2023 was paid about $450,000 as a part-time professor at a Chinese university, the indictment notes.
The indictment alleges that in February 2020, in response to questioning by the Fed’s inspector general’s office — the central bank’s internal watchdog — “Rogers lied about his accessing and passage of sensitive information and his associations with his co-conspirators.”
FBI Assistant Director in Charge of the Washington office David Sundberg in a statement said, “The Chinese Communist Party has expanded its economic espionage campaign to target U.S. government financial policies and trade secrets in an effort to undermine the U.S. and become the sole superpower.”
Sundberg was told Thursday that he was being forced out the FBI, NBC News reported Friday, citing two senior law enforcement sources.
His departure is part of a purge of top FBI executives and bureau field office leadership by the new Trump administration.
https://www.cnbc.com/2025/01/31/ex-seni ... china.html
CHINA
Re: CHINA
Interesting Engineering
"Research exposes DeepSeek’s AI training cost is not $6M, it’s a staggering $1.3B - One of the most prominent claims in circulation is that DeepSeek V3 incurs a training cost of around $6 million."
Kapil Kajal
Updated: Jan 31, 2025
In its latest report, SemiAnalysis, an independent research company, has spotlighted DeepSeek, a rising player in the AI landscape.
The SemiAnalysis challenges some of the prevailing narratives surrounding DeepSeek’s costs and compares them to competing technologies in the market.
One of the most prominent claims in circulation is that DeepSeek V3 incurs a training cost of around $6 million.
However, the SemiAnalysis report deconstructs this figure, stating that it failed to account for several critical factors.
DeepSeek’s $6 million myth
The $6 million estimate primarily considers GPU pre-training expenses, neglecting the significant investments in research and development, infrastructure, and other essential costs accruing to the company.
The report highlights that DeepSeek’s total server capital expenditure (CapEx) amounts to an astonishing $1.3 billion.
Much of this financial commitment is directed toward operating and maintaining its extensive GPU clusters, the backbone of its computational power.
DeepSeek reportedly has access to approximately 50,000 Hopper GPUs, leading to some misconceptions in the industry.
SemiAnalysis clarifies that this does not equate to having 50,000 H100s, as some previously inferred.
Instead, the GPU inventory comprises a mix of models, including H800s, H100s, and the country-specific H20s produced by NVIDIA in response to U.S. export restrictions.
This nuanced understanding of their hardware inventory underscores the strategic decisions in sourcing and operational efficiency at DeepSeek.
A notable aspect of the report is its reflection on DeepSeek’s organizational structure.
Unlike some of the larger AI laboratories, DeepSeek operates its data centers and employs a streamlined model that aids in its agility and efficiency.
As the AI landscape grows increasingly competitive, this ability to adapt quickly becomes a vital asset.
Better reasoning capabilities
Performance-wise, the analysis indicates that DeepSeek’s R1 model demonstrates comparable reasoning capabilities to OpenAI’s o1.
However, it stops short of labeling DeepSeek as the undisputed frontrunner across every performance metric.
While DeepSeek’s pricing strategy has garnered attention and accolades, there’s an important caveat: Google’s Gemini Flash 2.0, which is similar in capability, proves even more economical when accessed through API services.
This positions DeepSeek at a crossroads where balancing performance and cost is key to its future success.
The Multi-Head Latent Attention (MLA) technology is a groundbreaking innovation highlighted in the report.
This cutting-edge approach significantly slashes inference costs by an impressive 93.3% through reduced usage of key-value (KV) caching, representing a major leap toward cost-effective AI solutions.
Experts suggest that innovations that emerged from DeepSeek are likely to be swiftly adopted by Western AI labs eager to remain competitive.
While there’s optimism about potential improvements and efficiency gains, SemiAnalysis warns of external challenges.
The report speculates that operational costs could plummet another fivefold by the end of the year, driven by DeepSeek’s ability to adapt compared to its larger, more bureaucratic counterparts quickly.
Nevertheless, scaling operations amid tightening U.S. export controls represents a significant hurdle DeepSeek must navigate carefully.
In conclusion, SemiAnalysis paints a complex picture of DeepSeek’s current standing within the AI realm.
The revelations regarding its cost structure, GPU utilization, and innovative capabilities position DeepSeek as a formidable player.
As the firm continues to evolve, the industry watches closely — eager to see how it will respond to emerging challenges and opportunities in an ever-changing landscape.
https://interestingengineering.com/cult ... st-billion
"Research exposes DeepSeek’s AI training cost is not $6M, it’s a staggering $1.3B - One of the most prominent claims in circulation is that DeepSeek V3 incurs a training cost of around $6 million."
Kapil Kajal
Updated: Jan 31, 2025
In its latest report, SemiAnalysis, an independent research company, has spotlighted DeepSeek, a rising player in the AI landscape.
The SemiAnalysis challenges some of the prevailing narratives surrounding DeepSeek’s costs and compares them to competing technologies in the market.
One of the most prominent claims in circulation is that DeepSeek V3 incurs a training cost of around $6 million.
However, the SemiAnalysis report deconstructs this figure, stating that it failed to account for several critical factors.
DeepSeek’s $6 million myth
The $6 million estimate primarily considers GPU pre-training expenses, neglecting the significant investments in research and development, infrastructure, and other essential costs accruing to the company.
The report highlights that DeepSeek’s total server capital expenditure (CapEx) amounts to an astonishing $1.3 billion.
Much of this financial commitment is directed toward operating and maintaining its extensive GPU clusters, the backbone of its computational power.
DeepSeek reportedly has access to approximately 50,000 Hopper GPUs, leading to some misconceptions in the industry.
SemiAnalysis clarifies that this does not equate to having 50,000 H100s, as some previously inferred.
Instead, the GPU inventory comprises a mix of models, including H800s, H100s, and the country-specific H20s produced by NVIDIA in response to U.S. export restrictions.
This nuanced understanding of their hardware inventory underscores the strategic decisions in sourcing and operational efficiency at DeepSeek.
A notable aspect of the report is its reflection on DeepSeek’s organizational structure.
Unlike some of the larger AI laboratories, DeepSeek operates its data centers and employs a streamlined model that aids in its agility and efficiency.
As the AI landscape grows increasingly competitive, this ability to adapt quickly becomes a vital asset.
Better reasoning capabilities
Performance-wise, the analysis indicates that DeepSeek’s R1 model demonstrates comparable reasoning capabilities to OpenAI’s o1.
However, it stops short of labeling DeepSeek as the undisputed frontrunner across every performance metric.
While DeepSeek’s pricing strategy has garnered attention and accolades, there’s an important caveat: Google’s Gemini Flash 2.0, which is similar in capability, proves even more economical when accessed through API services.
This positions DeepSeek at a crossroads where balancing performance and cost is key to its future success.
The Multi-Head Latent Attention (MLA) technology is a groundbreaking innovation highlighted in the report.
This cutting-edge approach significantly slashes inference costs by an impressive 93.3% through reduced usage of key-value (KV) caching, representing a major leap toward cost-effective AI solutions.
Experts suggest that innovations that emerged from DeepSeek are likely to be swiftly adopted by Western AI labs eager to remain competitive.
While there’s optimism about potential improvements and efficiency gains, SemiAnalysis warns of external challenges.
The report speculates that operational costs could plummet another fivefold by the end of the year, driven by DeepSeek’s ability to adapt compared to its larger, more bureaucratic counterparts quickly.
Nevertheless, scaling operations amid tightening U.S. export controls represents a significant hurdle DeepSeek must navigate carefully.
In conclusion, SemiAnalysis paints a complex picture of DeepSeek’s current standing within the AI realm.
The revelations regarding its cost structure, GPU utilization, and innovative capabilities position DeepSeek as a formidable player.
As the firm continues to evolve, the industry watches closely — eager to see how it will respond to emerging challenges and opportunities in an ever-changing landscape.
https://interestingengineering.com/cult ... st-billion
Re: CHINA
REUTERS
"Record high imports pressure US trade deficit"
By Lucia Mutikani
February 5, 2025
Summary
* Trade deficit increases 24.7% to $98.4 billion in December
* Trade gap at $918.4 billion in 2024, largest since 2021
* Imports jump 3.5% in December to record high $364.9 billion
* Exports decrease 2.6% to $266.5 billion
* Goods deficit widens with Canada, shrinks with China, Mexico
WASHINGTON, Feb 5 (Reuters) - The U.S. trade deficit widened sharply in December as imports surged to a record high against the backdrop of tariff threats, which might have prompted businesses to rush purchases of foreign-made goods like finished metals and computers.
The report from the Commerce Department on Wednesday showed the United States experienced significant deficits with several trade partners, including China, Mexico and Canada, which have been targeted by President Donald Trump's administration for broad or additional tariffs.
Trump on Monday suspended a 25% tariff on Mexican and Canadian goods until next month.
An additional 10% levy on goods from China went into effect on Tuesday.
Though the new administration has mostly explained the tariffs as related to controlling illegal immigration and movement of illicit drugs, the surge in the deficit could strengthen its argument for a protectionist trade policy.
"The strength of imports appears largely driven by businesses rushing orders ahead of potential tariffs, a trend unlikely to reverse any time soon given there is still the risk of 25% tariffs on Mexico and Canada next month," said Thomas Ryan, North America economist at Capital Economics.
"Even though survey data point to an imminent rebound in exports, this suggests the trade deficit will remain wide this quarter."
The trade gap increased 24.7% to $98.4 billion, the highest since March 2022, from a revised $78.9 billion in November, the Commerce Department's Bureau of Economic Analysis (BEA) said.
It was the second-largest deficit on record and the monthly increase was biggest since March 2015.
Economists polled by Reuters had forecast the trade deficit soaring to $96.6 billion from the previously reported $78.2 billion in November.
The trade deficit swelled 17.0% to $918.4 billion in 2024, the largest since 2021.
Imports increased 3.5% to an all-time high of $364.9 billion.
Goods imports soared 4.0% to $293.1 billion.
They were boosted by a $10.8 billion jump in industrial supplies and materials, mostly reflecting a $9.2 billion increase in finished metal shapes, mostly from Switzerland.
That raised doubts among some economists that front-loading of imports was the whole story behind the surge in the trade deficit.
"Switzerland is pretty far from the top of President Trump's tariff hit list," said Stephen Stanley, chief U.S. economist at Santander U.S. Capital Markets.
"I assume that this is just a fluky one-off, which means that there is a good chance that the trade gap recedes substantially in January, unless the import-ahead-of-tariffs dynamic kicked in vigorously last month."
A survey from the Institute for Supply Management (ISM) on Wednesday showed services businesses feared higher prices and input shortages from tariffs in January.
Some providers of professional, scientific and technical services reported "the threat of tariffs is causing prices to rise," adding that "the threat of unstable international markets is resulting in shortages for various materials."
Businesses in the real estate, rental and leasing sector said the "concern going forward is the cost of materials and project work, if any tariffs go into effect."
The ISM's nonmanufacturing PMI slipped to 52.8 last month from 54.0 in December.
Steve Miller, chair of the ISM Services Business Survey Committee, said "poor weather conditions were highlighted by many respondents as impacting business levels and production."
Stocks on Wall Street were trading higher.
The dollar slipped against a basket of currencies.
U.S. Treasury yields fell.
WEAK EXPORTS
The trade report showed capital goods imports increased $1.3 billion, lifted by computers as well as computer accessories.
But imports of civilian aircraft fell as did those of automotive vehicles, parts and engines.
Consumer goods increased $2.2 billion, driven by toys, games and sporting goods, cell phones and other household goods.
Exports fell 2.6% to $266.5 billion.
Goods exports fell 4.2%, the most since May 2020, to $170.2 billion.
They were pulled down by a $1.8 billion decline in consumer goods.
Exports of industrial supplies and materials, which include petroleum, dropped $1.8 billion.
Capital goods exports declined $1.4 billion while those of automotive vehicles, parts and engines fell $0.9 billion.
The goods trade deficit jumped 18.2% to a record $123.0 billion.
Adjusted for inflation, the goods deficit widened 15.4% to $111.9 billion.
The goods trade deficit with Canada increased $2.9 billion to $7.9 billion in December.
While the goods trade gap with China narrowed in December, it increased to $295.4 billion in 2024 from $279.1 billion in 2023.
The shortfall with Mexico contracted to $15.2 billion from $15.4 billion in November.
Services imports increased $1.0 billion to a record $71.8 billion, while exports rose $0.4 billion to an all-time high of $96.3 billion.
December's data was broadly in line with the assumptions the BEA made in its advance gross domestic product estimate for the fourth quarter published last week, which showed trade had a surprisingly neutral impact on GDP after being a drag for three straight quarters.
The economy grew at a 2.3% annualized rate, with most of the drag coming from inventories, after expanding at a 3.1% pace in the July-September quarter.
"There is currently little risk of a major revision to the growth pace," said Capital Economics' Ryan.
Reporting by Lucia Mutikani; Editing by Chizu Nomiyama and Andrea Ricci
https://www.reuters.com/world/us/us-tra ... 025-02-05/
"Record high imports pressure US trade deficit"
By Lucia Mutikani
February 5, 2025
Summary
* Trade deficit increases 24.7% to $98.4 billion in December
* Trade gap at $918.4 billion in 2024, largest since 2021
* Imports jump 3.5% in December to record high $364.9 billion
* Exports decrease 2.6% to $266.5 billion
* Goods deficit widens with Canada, shrinks with China, Mexico
WASHINGTON, Feb 5 (Reuters) - The U.S. trade deficit widened sharply in December as imports surged to a record high against the backdrop of tariff threats, which might have prompted businesses to rush purchases of foreign-made goods like finished metals and computers.
The report from the Commerce Department on Wednesday showed the United States experienced significant deficits with several trade partners, including China, Mexico and Canada, which have been targeted by President Donald Trump's administration for broad or additional tariffs.
Trump on Monday suspended a 25% tariff on Mexican and Canadian goods until next month.
An additional 10% levy on goods from China went into effect on Tuesday.
Though the new administration has mostly explained the tariffs as related to controlling illegal immigration and movement of illicit drugs, the surge in the deficit could strengthen its argument for a protectionist trade policy.
"The strength of imports appears largely driven by businesses rushing orders ahead of potential tariffs, a trend unlikely to reverse any time soon given there is still the risk of 25% tariffs on Mexico and Canada next month," said Thomas Ryan, North America economist at Capital Economics.
"Even though survey data point to an imminent rebound in exports, this suggests the trade deficit will remain wide this quarter."
The trade gap increased 24.7% to $98.4 billion, the highest since March 2022, from a revised $78.9 billion in November, the Commerce Department's Bureau of Economic Analysis (BEA) said.
It was the second-largest deficit on record and the monthly increase was biggest since March 2015.
Economists polled by Reuters had forecast the trade deficit soaring to $96.6 billion from the previously reported $78.2 billion in November.
The trade deficit swelled 17.0% to $918.4 billion in 2024, the largest since 2021.
Imports increased 3.5% to an all-time high of $364.9 billion.
Goods imports soared 4.0% to $293.1 billion.
They were boosted by a $10.8 billion jump in industrial supplies and materials, mostly reflecting a $9.2 billion increase in finished metal shapes, mostly from Switzerland.
That raised doubts among some economists that front-loading of imports was the whole story behind the surge in the trade deficit.
"Switzerland is pretty far from the top of President Trump's tariff hit list," said Stephen Stanley, chief U.S. economist at Santander U.S. Capital Markets.
"I assume that this is just a fluky one-off, which means that there is a good chance that the trade gap recedes substantially in January, unless the import-ahead-of-tariffs dynamic kicked in vigorously last month."
A survey from the Institute for Supply Management (ISM) on Wednesday showed services businesses feared higher prices and input shortages from tariffs in January.
Some providers of professional, scientific and technical services reported "the threat of tariffs is causing prices to rise," adding that "the threat of unstable international markets is resulting in shortages for various materials."
Businesses in the real estate, rental and leasing sector said the "concern going forward is the cost of materials and project work, if any tariffs go into effect."
The ISM's nonmanufacturing PMI slipped to 52.8 last month from 54.0 in December.
Steve Miller, chair of the ISM Services Business Survey Committee, said "poor weather conditions were highlighted by many respondents as impacting business levels and production."
Stocks on Wall Street were trading higher.
The dollar slipped against a basket of currencies.
U.S. Treasury yields fell.
WEAK EXPORTS
The trade report showed capital goods imports increased $1.3 billion, lifted by computers as well as computer accessories.
But imports of civilian aircraft fell as did those of automotive vehicles, parts and engines.
Consumer goods increased $2.2 billion, driven by toys, games and sporting goods, cell phones and other household goods.
Exports fell 2.6% to $266.5 billion.
Goods exports fell 4.2%, the most since May 2020, to $170.2 billion.
They were pulled down by a $1.8 billion decline in consumer goods.
Exports of industrial supplies and materials, which include petroleum, dropped $1.8 billion.
Capital goods exports declined $1.4 billion while those of automotive vehicles, parts and engines fell $0.9 billion.
The goods trade deficit jumped 18.2% to a record $123.0 billion.
Adjusted for inflation, the goods deficit widened 15.4% to $111.9 billion.
The goods trade deficit with Canada increased $2.9 billion to $7.9 billion in December.
While the goods trade gap with China narrowed in December, it increased to $295.4 billion in 2024 from $279.1 billion in 2023.
The shortfall with Mexico contracted to $15.2 billion from $15.4 billion in November.
Services imports increased $1.0 billion to a record $71.8 billion, while exports rose $0.4 billion to an all-time high of $96.3 billion.
December's data was broadly in line with the assumptions the BEA made in its advance gross domestic product estimate for the fourth quarter published last week, which showed trade had a surprisingly neutral impact on GDP after being a drag for three straight quarters.
The economy grew at a 2.3% annualized rate, with most of the drag coming from inventories, after expanding at a 3.1% pace in the July-September quarter.
"There is currently little risk of a major revision to the growth pace," said Capital Economics' Ryan.
Reporting by Lucia Mutikani; Editing by Chizu Nomiyama and Andrea Ricci
https://www.reuters.com/world/us/us-tra ... 025-02-05/
Re: CHINA
REUTERS
"Applied Materials gives dour quarterly revenue forecast as export curbs tighten; shares fall"
By Reuters
February 13, 2025
Feb 13 (Reuters) - Applied Materials forecast second-quarter revenue below market estimates on Thursday, expecting escalating geopolitical tensions to weigh on sales of its chipmaking equipment, sending its shares down more than 5% in extended trading.
Revenue from China, Applied's largest market by revenue share in the first quarter, is threatened by tighter U.S. restrictions on exports of chipmaking technology.
Such curbs will hurt 2025 revenue by about $400 million, Chief Executive Officer Gary Dickerson said on a post-earnings call.
Half of this impact will be seen in the second quarter, financial chief Brice Hill said on the call.
This represents about 1.4% of the $29.18 billion analysts expect Applied to record in fiscal 2025 revenue, according to data compiled by LSEG.
The U.S. government said in December that new controls will be placed on the export of semiconductor manufacturing equipment needed to produce advanced-node chips to China.
"The ability of U.S. companies to serve the China market is constrained and has been further limited by updated trade rules," CEO Dickerson said.
Applied Materials forecast second-quarter revenue of about $7.1 billion, plus or minus $400 million, compared with the estimates of $7.21 billion
Half of the impact from trade restrictions to 2025 revenue is expected in the services segment, under which the company provides upkeep and optimization for machinery, the executives said, as the company is unable to address some customers in China.
Sales to China accounted for about 31% of Applied Materials' total first-quarter sales, down from about 45% of total revenue in the year-ago period.
Earlier on Thursday, U.S. President Donald Trump also tasked his economics team with devising a plan to impose reciprocal tariffs on every country that imposes duties on U.S. imports, targeting China, Japan and South Korea — some of Applied's largest markets.
This has offset the positive impact of a rise in demand for advanced chips capable of processing the vast data employed by generative AI.
Applied expects second-quarter adjusted profit of $2.30 per share, plus or minus 18 cents, in line with estimates of $2.30 per share.
The company reported first-quarter revenue of $7.17 billion, beating estimates.
Reporting by Arsheeya Bajwa in Bengaluru; Editing by Alan Barona
https://www.reuters.com/technology/appl ... 025-02-13/
"Applied Materials gives dour quarterly revenue forecast as export curbs tighten; shares fall"
By Reuters
February 13, 2025
Feb 13 (Reuters) - Applied Materials forecast second-quarter revenue below market estimates on Thursday, expecting escalating geopolitical tensions to weigh on sales of its chipmaking equipment, sending its shares down more than 5% in extended trading.
Revenue from China, Applied's largest market by revenue share in the first quarter, is threatened by tighter U.S. restrictions on exports of chipmaking technology.
Such curbs will hurt 2025 revenue by about $400 million, Chief Executive Officer Gary Dickerson said on a post-earnings call.
Half of this impact will be seen in the second quarter, financial chief Brice Hill said on the call.
This represents about 1.4% of the $29.18 billion analysts expect Applied to record in fiscal 2025 revenue, according to data compiled by LSEG.
The U.S. government said in December that new controls will be placed on the export of semiconductor manufacturing equipment needed to produce advanced-node chips to China.
"The ability of U.S. companies to serve the China market is constrained and has been further limited by updated trade rules," CEO Dickerson said.
Applied Materials forecast second-quarter revenue of about $7.1 billion, plus or minus $400 million, compared with the estimates of $7.21 billion
Half of the impact from trade restrictions to 2025 revenue is expected in the services segment, under which the company provides upkeep and optimization for machinery, the executives said, as the company is unable to address some customers in China.
Sales to China accounted for about 31% of Applied Materials' total first-quarter sales, down from about 45% of total revenue in the year-ago period.
Earlier on Thursday, U.S. President Donald Trump also tasked his economics team with devising a plan to impose reciprocal tariffs on every country that imposes duties on U.S. imports, targeting China, Japan and South Korea — some of Applied's largest markets.
This has offset the positive impact of a rise in demand for advanced chips capable of processing the vast data employed by generative AI.
Applied expects second-quarter adjusted profit of $2.30 per share, plus or minus 18 cents, in line with estimates of $2.30 per share.
The company reported first-quarter revenue of $7.17 billion, beating estimates.
Reporting by Arsheeya Bajwa in Bengaluru; Editing by Alan Barona
https://www.reuters.com/technology/appl ... 025-02-13/
Re: CHINA
REUTERS
"FACTBOX China's AI firms take spotlight with deals, low-cost models"
By Reuters
February 14, 2025
BEIJING, Feb 14 (Reuters) - Alibaba's announcement this week that it will partner with Apple to support iPhones' artificial intelligence services offering in China has again thrust the spotlight on the country's AI industry on the heels of DeepSeek.
Following are some of the prominent AI services and their developers to watch in China.
DOUBAO
Developed by ByteDance, a latecomer to the large language model (LLM) market, the namesake chatbot is the most popular consumer AI app in China.
The company's latest LLM, Doubao-1.5-pro, was released in January and outperforms ChatGPT's GPT-4 in some metrics.
ByteDance is among the leaders in developing cost-effective LLMs, and its offering is the cheapest available in the market, even more affordable than DeepSeek, according to a Barclays report.
Doubao-1.5 was priced at only 3% to 4% of GPT-4's pricing.
ByteDance's low-cost advantage is attributed to its use of the "mixture of experts" (MoE) framework, which is common among various AI models in China, including those of DeepSeek.
QWEN
Developed by e-commerce giant Alibaba, Qwen is the company's flagship LLM and chatbot.
It was released in March 2023 and has been rated among the top-tier in global benchmarks.
In January, Alibaba released Qwen 2.5-Max and claimed its functionality surpassed that of highly-acclaimed DeepSeek-V3.
A strong advocate of open AI development, Alibaba announced in September that many of its AI models would be made open-source.
ERNIE
Chinese search giant Baidu was among the first in China to launch a ChatGPT equivalent, Ernie Bot.
Baidu claims its latest model, Ernie 4.0, matches the capabilities of GPT-4.
On February 14, Baidu announced a plan to make its next-generation LLM, Ernie 4.5, open-source from June 30, a major shift in strategy as competition heats up.
HUNYUAN
Tencent released its flagship model Hunyuan in September and launched its chatbot, Yuanbao, in May 2024.
The company's massive user base, acquired through its popular consumer apps like WeChat, gives Tencent a significant potential advantage in pushing out consumer applications in the future.
Tencent also offers open-source models and advocates for open-source development.
Another advantage of Hunyuan is its focus on multi-modality, which includes text, image, and video generation capabilities.
GLM
Zhipu AI developed GLM and claims its latest LLM, GLM4, outperforms GPT-4 in some metrics.
Initially born out of a Tsinghua University laboratory, Zhipu counts Alibaba, Tencent, and the state-owned fund Zhongguancun Science City Innovation Development among its backers.
In January, the U.S. Commerce Department placed Zhipu on an export control entity list.
KIMI
It was developed by Moonshot, which is backed by the likes of Alibaba, Tencent, and Hongshan.
Its chatbot Kimi is known for its long-context processing technology and the company claims the model supports two million input tokens, or units of words, for each query.
On January 20, just two days before DeepSeek released the R1 model, Moonshot unveiled its own reasoning model, Kimi 1.5.
MINIMAX
Minimax is a pioneer among Chinese firms in researching the MoE framework and it open-sourced its most advanced model, MiniMax-01, in January.
MiniMax was founded in December 2021, well before the AI craze sparked by the launch of OpenAI's ChatGPT.
It is backed by Alibaba and other prominent investors, including Hongshan, Gaorong Capital, and IDG.
In addition to LLMs, MiniMax has gained recognition for its social consumer apps, such as Takie and Xingye.
01.AI
Similar to DeepSeek, 01.AI has adopted an open-source approach and pioneered the MoE framework.
The company claims that its models are trained with fewer resources and have a cost-effective advantage, achieving the lowest cost level of LLM inference in the industry, 1/40 less than GPT-4's list price.
Its founder Kai-Fu Lee, a former head of Google China, said recently in a written interview with Reuters that 01.AI has formed a joint partnership with Alibaba to create a lab focused on continuing the development of LLM technologies.
The company is backed by companies like Alibaba and Sinovation Ventures.
BAICHUAN
Baichuan was founded in April 2023 by a team of former executives from the search engine Sogou, including founder and CEO Wang Xiaochuan, who previously served as the CEO of Sogou.
The company gained recognition in June 2023 by becoming the first in China to launch an open-source model.
Reporting by Liam Mo and Brenda Goh; Editing by Miyoung Kim and Sharon Singleton
https://www.reuters.com/technology/arti ... 025-02-14/
"FACTBOX China's AI firms take spotlight with deals, low-cost models"
By Reuters
February 14, 2025
BEIJING, Feb 14 (Reuters) - Alibaba's announcement this week that it will partner with Apple to support iPhones' artificial intelligence services offering in China has again thrust the spotlight on the country's AI industry on the heels of DeepSeek.
Following are some of the prominent AI services and their developers to watch in China.
DOUBAO
Developed by ByteDance, a latecomer to the large language model (LLM) market, the namesake chatbot is the most popular consumer AI app in China.
The company's latest LLM, Doubao-1.5-pro, was released in January and outperforms ChatGPT's GPT-4 in some metrics.
ByteDance is among the leaders in developing cost-effective LLMs, and its offering is the cheapest available in the market, even more affordable than DeepSeek, according to a Barclays report.
Doubao-1.5 was priced at only 3% to 4% of GPT-4's pricing.
ByteDance's low-cost advantage is attributed to its use of the "mixture of experts" (MoE) framework, which is common among various AI models in China, including those of DeepSeek.
QWEN
Developed by e-commerce giant Alibaba, Qwen is the company's flagship LLM and chatbot.
It was released in March 2023 and has been rated among the top-tier in global benchmarks.
In January, Alibaba released Qwen 2.5-Max and claimed its functionality surpassed that of highly-acclaimed DeepSeek-V3.
A strong advocate of open AI development, Alibaba announced in September that many of its AI models would be made open-source.
ERNIE
Chinese search giant Baidu was among the first in China to launch a ChatGPT equivalent, Ernie Bot.
Baidu claims its latest model, Ernie 4.0, matches the capabilities of GPT-4.
On February 14, Baidu announced a plan to make its next-generation LLM, Ernie 4.5, open-source from June 30, a major shift in strategy as competition heats up.
HUNYUAN
Tencent released its flagship model Hunyuan in September and launched its chatbot, Yuanbao, in May 2024.
The company's massive user base, acquired through its popular consumer apps like WeChat, gives Tencent a significant potential advantage in pushing out consumer applications in the future.
Tencent also offers open-source models and advocates for open-source development.
Another advantage of Hunyuan is its focus on multi-modality, which includes text, image, and video generation capabilities.
GLM
Zhipu AI developed GLM and claims its latest LLM, GLM4, outperforms GPT-4 in some metrics.
Initially born out of a Tsinghua University laboratory, Zhipu counts Alibaba, Tencent, and the state-owned fund Zhongguancun Science City Innovation Development among its backers.
In January, the U.S. Commerce Department placed Zhipu on an export control entity list.
KIMI
It was developed by Moonshot, which is backed by the likes of Alibaba, Tencent, and Hongshan.
Its chatbot Kimi is known for its long-context processing technology and the company claims the model supports two million input tokens, or units of words, for each query.
On January 20, just two days before DeepSeek released the R1 model, Moonshot unveiled its own reasoning model, Kimi 1.5.
MINIMAX
Minimax is a pioneer among Chinese firms in researching the MoE framework and it open-sourced its most advanced model, MiniMax-01, in January.
MiniMax was founded in December 2021, well before the AI craze sparked by the launch of OpenAI's ChatGPT.
It is backed by Alibaba and other prominent investors, including Hongshan, Gaorong Capital, and IDG.
In addition to LLMs, MiniMax has gained recognition for its social consumer apps, such as Takie and Xingye.
01.AI
Similar to DeepSeek, 01.AI has adopted an open-source approach and pioneered the MoE framework.
The company claims that its models are trained with fewer resources and have a cost-effective advantage, achieving the lowest cost level of LLM inference in the industry, 1/40 less than GPT-4's list price.
Its founder Kai-Fu Lee, a former head of Google China, said recently in a written interview with Reuters that 01.AI has formed a joint partnership with Alibaba to create a lab focused on continuing the development of LLM technologies.
The company is backed by companies like Alibaba and Sinovation Ventures.
BAICHUAN
Baichuan was founded in April 2023 by a team of former executives from the search engine Sogou, including founder and CEO Wang Xiaochuan, who previously served as the CEO of Sogou.
The company gained recognition in June 2023 by becoming the first in China to launch an open-source model.
Reporting by Liam Mo and Brenda Goh; Editing by Miyoung Kim and Sharon Singleton
https://www.reuters.com/technology/arti ... 025-02-14/
Re: CHINA
REUTERS
"Foreign holdings of US Treasuries fall in December"
By Reuters
February 18, 2025
Feb 18 (Reuters) - Foreign holdings of U.S. Treasuries fell in December as the two largest foreign holders of the government debt, Japan and China, both cut their portfolios, data from the Treasury Department showed on Tuesday.
Holdings of U.S. Treasuries fell to $8.513 trillion in December from $8.633 trillion in November.
Holdings had reached $8.679 trillion in September.
Japan's Treasuries holdings fell to $1.060 trillion from $1.087 trillion the previous month.
Japan remains the largest foreign holder of U.S. Treasury securities.
China, which is No. 2, cut its holdings to $759 billion from $768.6 billion in November.
Treasury yields surged in December on expectations of higher growth and a possible inflation resurgence as traders gauged the likely impact of tariffs and immigration reforms by the Donald Trump administration, following Trump’s victory in the November presidential election.
Federal Reserve policymakers also said that they see fewer interest rate cuts in 2025 at the U.S. central bank’s December meeting, citing inflation concerns.
Reporting by Karen Brettell; Editing by Leslie Adler
https://www.reuters.com/markets/us/fore ... 025-02-18/
"Foreign holdings of US Treasuries fall in December"
By Reuters
February 18, 2025
Feb 18 (Reuters) - Foreign holdings of U.S. Treasuries fell in December as the two largest foreign holders of the government debt, Japan and China, both cut their portfolios, data from the Treasury Department showed on Tuesday.
Holdings of U.S. Treasuries fell to $8.513 trillion in December from $8.633 trillion in November.
Holdings had reached $8.679 trillion in September.
Japan's Treasuries holdings fell to $1.060 trillion from $1.087 trillion the previous month.
Japan remains the largest foreign holder of U.S. Treasury securities.
China, which is No. 2, cut its holdings to $759 billion from $768.6 billion in November.
Treasury yields surged in December on expectations of higher growth and a possible inflation resurgence as traders gauged the likely impact of tariffs and immigration reforms by the Donald Trump administration, following Trump’s victory in the November presidential election.
Federal Reserve policymakers also said that they see fewer interest rate cuts in 2025 at the U.S. central bank’s December meeting, citing inflation concerns.
Reporting by Karen Brettell; Editing by Leslie Adler
https://www.reuters.com/markets/us/fore ... 025-02-18/
Re: CHINA
REUTERS
"Lenovo quarterly results smash estimates, says DeepSeek's AI tech will drive growth"
By Reuters
February 20, 2025
BEIJING, Feb 20 (Reuters) - Chinese tech company Lenovo reported hefty gains in quarterly revenue and net profit that beat expectations and said it would benefit from the adoption of DeepSeek's artifical intelligence model in its products.
Driven by robust demand for its AI-powered servers, Lenovo's third-quarter revenue surged by a fifth to $18.8 billion, ahead of an LSEG consensus estimate of $17.8 billion.
Net profit more than doubled from the previous year to $693 million, trouncing an average forecast of $367.7 million.
Though best known for its personal computers, the company has diversified in recent years, expanding into software and services.
It also has positioned itself to benefit from the booming demand for AI applications by expanding its AI server and software businesses.
Lenovo launched its first AI-powered PCs in China last May and followed that with a global rollout in September.
CEO Yang Yuanqing has projected that AI PCs will account for a quarter of Lenovo's shipments by 2025, potentially reaching 80% by 2027.
The company said this week it has integrated technology from startup DeepSeek - which has upended the AI sector with its low-cost model - into its devices, including PCs and tablets.
"DeepSeek has improved AI efficiency."
"The new models with a high inferencing efficiency and low computing power costs will democratise access to AI," Yang told Reuters on Thursday in a post-earnings interview.
That would drive demand and increase the ways in which AI can be used, he said, adding that the emergence of DeepSeek may also boost demand for GPU servers.
Lenovo's infrastructure solutions group, which includes servers, saw a 59% revenue increase in the December quarter compared with the previous year.
The solutions and services group, which offers cloud-based software for enterprise clients, posted $2.3 billion in revenue, up 12% from a year earlier.
Lenovo's Hong Kong-listed shares initially jumped after the results but were last trading down 6%.
They have surged 17% for the year to date.
Reporting by Liam Mo, Che Pan and Brenda Goh; Editing by Edwina Gibbs
https://www.reuters.com/technology/leno ... 025-02-20/
"Lenovo quarterly results smash estimates, says DeepSeek's AI tech will drive growth"
By Reuters
February 20, 2025
BEIJING, Feb 20 (Reuters) - Chinese tech company Lenovo reported hefty gains in quarterly revenue and net profit that beat expectations and said it would benefit from the adoption of DeepSeek's artifical intelligence model in its products.
Driven by robust demand for its AI-powered servers, Lenovo's third-quarter revenue surged by a fifth to $18.8 billion, ahead of an LSEG consensus estimate of $17.8 billion.
Net profit more than doubled from the previous year to $693 million, trouncing an average forecast of $367.7 million.
Though best known for its personal computers, the company has diversified in recent years, expanding into software and services.
It also has positioned itself to benefit from the booming demand for AI applications by expanding its AI server and software businesses.
Lenovo launched its first AI-powered PCs in China last May and followed that with a global rollout in September.
CEO Yang Yuanqing has projected that AI PCs will account for a quarter of Lenovo's shipments by 2025, potentially reaching 80% by 2027.
The company said this week it has integrated technology from startup DeepSeek - which has upended the AI sector with its low-cost model - into its devices, including PCs and tablets.
"DeepSeek has improved AI efficiency."
"The new models with a high inferencing efficiency and low computing power costs will democratise access to AI," Yang told Reuters on Thursday in a post-earnings interview.
That would drive demand and increase the ways in which AI can be used, he said, adding that the emergence of DeepSeek may also boost demand for GPU servers.
Lenovo's infrastructure solutions group, which includes servers, saw a 59% revenue increase in the December quarter compared with the previous year.
The solutions and services group, which offers cloud-based software for enterprise clients, posted $2.3 billion in revenue, up 12% from a year earlier.
Lenovo's Hong Kong-listed shares initially jumped after the results but were last trading down 6%.
They have surged 17% for the year to date.
Reporting by Liam Mo, Che Pan and Brenda Goh; Editing by Edwina Gibbs
https://www.reuters.com/technology/leno ... 025-02-20/
Re: CHINA
REUTERS
"DeepSeek rushes to launch new AI model as China goes all in"
By Eduardo Baptista, Julie Zhu and Fanny Potkin
February 25, 2025
Summary
* DeepSeek likely to release next-generation R2 model before May - sources
* Startup shuns typical Chinese tech giant culture, is known for flat hierarchy
* China embraces DeepSeek after initial regulatory concerns about its mass chip purchases
* Firm instructed to keep low-profile amid global concerns about its privacy practices
BEIJING/HONG KONG/SINGAPORE, Feb 25 (Reuters) - DeepSeek is looking to press home its advantage.
The Chinese startup triggered a $1 trillion-plus sell-off in global equities markets last month with a cut-price AI reasoning model that outperformed many Western competitors.
Now, the Hangzhou-based firm is accelerating the launch of the successor to January's R1 model, according to three people familiar with the company.
Deepseek had planned to release R2 in early May but now wants it out as early as possible, two of them said, without providing specifics.
The company says it hopes the new model will produce better coding and be able to reason in languages beyond English.
Details of the accelerated timeline for R2's release have not been previously reported.
DeepSeek did not respond to a request for comment for this story.
Rivals are still digesting the implications of R1, which was built with less-powerful Nvidia chips but is competitive with those developed at the costs of hundreds of billions of dollars by U.S. tech giants.
"The launch of DeepSeek's R2 model could be a pivotal moment in the AI industry," said Vijayasimha Alilughatta, chief operating officer of Indian tech services provider Zensar.
DeepSeek's success at creating cost-effective AI models "would likely spur companies worldwide to accelerate their own efforts ... breaking the stranglehold of the few dominant players in the field," he said.
R2 is likely to worry the U.S. government, which has identified leadership of AI as a national priority.
Its release may further galvanize Chinese authorities and companies, dozens of which say they have started integrating DeepSeek models into their products.
Little is known about DeepSeek, whose founder Liang Wenfeng became a billionaire through his quantitative hedge fund High-Flyer.
Liang, who was described by a former employer as "low-key and introverted," has not spoken to any media since July 2024.
Reuters interviewed a dozen former employees, as well as quant fund professionals knowledgeable about the operations of DeepSeek and its parent company High-Flyer.
It also reviewed state media articles, social-media posts from the companies and research papers dating back to 2019.
They told a story of a company that functioned more like a research lab than a for-profit enterprise and was unencumbered by the hierarchical traditions of China's high-pressure tech industry, even as it became responsible for what many investors see as the latest breakthrough in AI.
DIFFERENT PATH
Liang was born in 1985 in a rural village in the southern province of Guangdong.
He later obtained communication engineering degrees at the elite Zhejiang University.
One of his first jobs was running a research department at a smart imaging firm in Shanghai.
His then-boss, Zhou Chaoen, told state media on Feb. 9 that Liang had hired prize-winning algorithm engineers and operated with a "flat management style."
At DeepSeek and High-Flyer, Liang has similarly shunned the practices of Chinese tech giants known for rigid top-down management, low pay for young employees and "996" - working from 9 a.m. to 9 p.m. six days a week.
Liang opened his Beijing office within walking distance of Tsinghua University and Peking University, China's two most prestigious education institutions.
He regularly delved into technical details and was happy to work alongside Gen-Z interns and recent graduates that comprised the bulk of its workforce, according to two former employees.
They also described usually working eight-hour days in a collaborative atmosphere.
"Liang gave us control and treated us as experts."
"He constantly asked questions and learned alongside us," said 26-year-old researcher Benjamin Liu, who left the company in September.
"DeepSeek allowed me to take ownership of critical parts of the pipeline, which was very exciting."
Liang did not respond to questions sent via DeepSeek.
While Baidu and other Chinese tech giants were racing to build their consumer-facing versions of ChatGPT in 2023 and profit off of the global AI boom, Liang told Chinese media outlet Waves last year that he deliberately avoided spending heavily on app development, focusing instead on refining the AI model's quality.
Both DeepSeek and High-Flyer are known for paying generously, according to three people familiar with its compensation practices.
At High-Flyer, it is not uncommon for a senior data scientist to make 1.5 million yuan annually, while competitors rarely pay more than 800,000, said one of the people, a rival quant fund manager who knows Liang.
The largesse was funded by High-Flyer, which became one of China's most successful quant funds and, even after a government crackdown on the sector, still manages tens of billions of yuan, according to two people in the industry.
COMPUTING POWER
DeepSeek's success with a low-cost AI model is based on High-Flyer's decade-long and substantial investment in research and computing power, three people said.
The quant fund was an earlier pioneer in AI trading and a top executive said in 2020 that High-Flyer was going "all in" on AI by re-investing 70% of its revenue, mostly into AI research.
High-Flyer spent 1.2 billion yuan on two supercomputing AI clusters in 2020 and 2021.
The second cluster, Fire-Flyer II, was made up of around 10,000 Nvidia A100 chips, used for training AI models.
DeepSeek had not been established at that time, so the accumulation of computing power caught the attention of Chinese securities regulators, said a person with direct knowledge of officials' thinking.
"Regulators wanted to know why they need so many chips?" the person said.
"How they were going to use it?"
"What kind of impact would that have on the market?"
Authorities decided not to intervene, in a move that would prove crucial for DeepSeek's fortunes: the U.S. banned the export of A100 chips to China in 2022, at which point Fire-Flyer II was already in operation.
Beijing now celebrates DeepSeek, but has instructed it not to engage with the media without approval, according to a person familiar with Chinese official thinking.
Authorities had asked Liang to keep a low-profile because they were worried that too much hype in the media would draw unnecessary attention, the person said.
China's cabinet and commerce ministry, as well as China's securities regulator, did not respond to requests for comment.
As one of the few companies with a large A100 cluster, High-Flyer and DeepSeek were able to attract some of China's best research talent, two former employees said.
"The key advantage of vast (computing) resources is that it allows for large-scale experimentation," said Liu, the former employee.
Some Western AI entrepreneurs, like Scale AI CEO Alexandr Wang, have claimed that DeepSeek had as many as 50,000 higher-end Nvidia chips that are banned for export to China.
He has not produced evidence for the allegation or responded to Reuters' requests to provide proof.
DeepSeek has not responded to Wang's claims.
Two former employees attributed the company's success to Liang's focus on more cost-effective AI architecture.
The startup used techniques like Mixture-of-Experts (MoE) and multihead latent attention (MLA), which incur far lower computing costs, its research papers show.
The MoE technique divides an AI model into different areas of expertise and activates only those related to a query, as opposed to more common architectures that use the entire model.
MLA architecture allows a model to process different aspects of one piece of information simultaneously, helping it detect key details more effectively.
While competitors like France's Mistral have developed models based on MoE, DeepSeek was the first firm to depend heavily on this architecture while achieving parity with more expensively built models.
DeepSeek's pricing was 20 to 40 times cheaper than what OpenAI charged for equivalent models, analysts at Bernstein brokerage estimated in early February.
For now, Western and Chinese tech giants have signaled plans to continue heavy AI spending, but DeepSeek's success with R1 and its earlier V3 model has prompted some to alter strategies.
OpenAI cut prices this month, while Google's Gemini has introduced discounted tiers of access.
Since R1's launch, OpenAI has also released an O3-Mini model that relies on less computing power.
Adnan Masood of U.S. tech services provider UST told Reuters that his laboratory had run benchmarks that found R1 often used three times as many tokens, or units of data processed by the AI model, for reasoning as OpenAI's scaled-down model.
STATE EMBRACE
Even before R1 gripped global attention, there were signs that DeepSeek had caught Beijing's favor.
In January, state media reported that Liang attended a meeting with Chinese Premier Li Qiang in Beijing as the designated representative of the AI sector, ahead of the leaders of better-known firms.
The subsequent fanfare over the cost competitiveness of its models has buoyed Beijing's belief that it can out-innovate the U.S., with Chinese companies and government bodies embracing DeepSeek models at a pace that has not been offered to other firms.
At least 13 Chinese city governments and 10 state-owned energy companies say they have deployed DeepSeek into their systems, while tech giants Lenovo, Baidu and Tencent - owner of China's largest social media app WeChat - have integrated DeepSeek's models into their products.
Chinese leader Xi Jinping and Li "have signalled they endorse DeepSeek," said Alfred Wu, an expert on Chinese policymaking at Singapore's Lee Kuan Yew School of Public Policy.
"Now everyone just endorses it."
The Chinese embrace comes as governments from South Korea to Italy remove DeepSeek from national app stores, citing privacy concerns.
"If DeepSeek becomes the go-to AI model across Chinese state entities, Western regulators might see this as another reason to escalate restrictions on AI chips or software collaborations," said Stephen Wu, an AI expert and founder of hedge fund Carthage Capital.
Further limits on advanced AI chips are a challenge that Liang has acknowledged.
"Our problem has never been funding," he told Waves in July.
"It's the embargo on high-end chips."
Additional reporting by Samuel Shen, Gu Li, Larissa Liao, Aditya Soni and Shanghai Newsroom; Editing by Brenda Goh and Katerina Ang
https://www.reuters.com/technology/arti ... 025-02-25/
"DeepSeek rushes to launch new AI model as China goes all in"
By Eduardo Baptista, Julie Zhu and Fanny Potkin
February 25, 2025
Summary
* DeepSeek likely to release next-generation R2 model before May - sources
* Startup shuns typical Chinese tech giant culture, is known for flat hierarchy
* China embraces DeepSeek after initial regulatory concerns about its mass chip purchases
* Firm instructed to keep low-profile amid global concerns about its privacy practices
BEIJING/HONG KONG/SINGAPORE, Feb 25 (Reuters) - DeepSeek is looking to press home its advantage.
The Chinese startup triggered a $1 trillion-plus sell-off in global equities markets last month with a cut-price AI reasoning model that outperformed many Western competitors.
Now, the Hangzhou-based firm is accelerating the launch of the successor to January's R1 model, according to three people familiar with the company.
Deepseek had planned to release R2 in early May but now wants it out as early as possible, two of them said, without providing specifics.
The company says it hopes the new model will produce better coding and be able to reason in languages beyond English.
Details of the accelerated timeline for R2's release have not been previously reported.
DeepSeek did not respond to a request for comment for this story.
Rivals are still digesting the implications of R1, which was built with less-powerful Nvidia chips but is competitive with those developed at the costs of hundreds of billions of dollars by U.S. tech giants.
"The launch of DeepSeek's R2 model could be a pivotal moment in the AI industry," said Vijayasimha Alilughatta, chief operating officer of Indian tech services provider Zensar.
DeepSeek's success at creating cost-effective AI models "would likely spur companies worldwide to accelerate their own efforts ... breaking the stranglehold of the few dominant players in the field," he said.
R2 is likely to worry the U.S. government, which has identified leadership of AI as a national priority.
Its release may further galvanize Chinese authorities and companies, dozens of which say they have started integrating DeepSeek models into their products.
Little is known about DeepSeek, whose founder Liang Wenfeng became a billionaire through his quantitative hedge fund High-Flyer.
Liang, who was described by a former employer as "low-key and introverted," has not spoken to any media since July 2024.
Reuters interviewed a dozen former employees, as well as quant fund professionals knowledgeable about the operations of DeepSeek and its parent company High-Flyer.
It also reviewed state media articles, social-media posts from the companies and research papers dating back to 2019.
They told a story of a company that functioned more like a research lab than a for-profit enterprise and was unencumbered by the hierarchical traditions of China's high-pressure tech industry, even as it became responsible for what many investors see as the latest breakthrough in AI.
DIFFERENT PATH
Liang was born in 1985 in a rural village in the southern province of Guangdong.
He later obtained communication engineering degrees at the elite Zhejiang University.
One of his first jobs was running a research department at a smart imaging firm in Shanghai.
His then-boss, Zhou Chaoen, told state media on Feb. 9 that Liang had hired prize-winning algorithm engineers and operated with a "flat management style."
At DeepSeek and High-Flyer, Liang has similarly shunned the practices of Chinese tech giants known for rigid top-down management, low pay for young employees and "996" - working from 9 a.m. to 9 p.m. six days a week.
Liang opened his Beijing office within walking distance of Tsinghua University and Peking University, China's two most prestigious education institutions.
He regularly delved into technical details and was happy to work alongside Gen-Z interns and recent graduates that comprised the bulk of its workforce, according to two former employees.
They also described usually working eight-hour days in a collaborative atmosphere.
"Liang gave us control and treated us as experts."
"He constantly asked questions and learned alongside us," said 26-year-old researcher Benjamin Liu, who left the company in September.
"DeepSeek allowed me to take ownership of critical parts of the pipeline, which was very exciting."
Liang did not respond to questions sent via DeepSeek.
While Baidu and other Chinese tech giants were racing to build their consumer-facing versions of ChatGPT in 2023 and profit off of the global AI boom, Liang told Chinese media outlet Waves last year that he deliberately avoided spending heavily on app development, focusing instead on refining the AI model's quality.
Both DeepSeek and High-Flyer are known for paying generously, according to three people familiar with its compensation practices.
At High-Flyer, it is not uncommon for a senior data scientist to make 1.5 million yuan annually, while competitors rarely pay more than 800,000, said one of the people, a rival quant fund manager who knows Liang.
The largesse was funded by High-Flyer, which became one of China's most successful quant funds and, even after a government crackdown on the sector, still manages tens of billions of yuan, according to two people in the industry.
COMPUTING POWER
DeepSeek's success with a low-cost AI model is based on High-Flyer's decade-long and substantial investment in research and computing power, three people said.
The quant fund was an earlier pioneer in AI trading and a top executive said in 2020 that High-Flyer was going "all in" on AI by re-investing 70% of its revenue, mostly into AI research.
High-Flyer spent 1.2 billion yuan on two supercomputing AI clusters in 2020 and 2021.
The second cluster, Fire-Flyer II, was made up of around 10,000 Nvidia A100 chips, used for training AI models.
DeepSeek had not been established at that time, so the accumulation of computing power caught the attention of Chinese securities regulators, said a person with direct knowledge of officials' thinking.
"Regulators wanted to know why they need so many chips?" the person said.
"How they were going to use it?"
"What kind of impact would that have on the market?"
Authorities decided not to intervene, in a move that would prove crucial for DeepSeek's fortunes: the U.S. banned the export of A100 chips to China in 2022, at which point Fire-Flyer II was already in operation.
Beijing now celebrates DeepSeek, but has instructed it not to engage with the media without approval, according to a person familiar with Chinese official thinking.
Authorities had asked Liang to keep a low-profile because they were worried that too much hype in the media would draw unnecessary attention, the person said.
China's cabinet and commerce ministry, as well as China's securities regulator, did not respond to requests for comment.
As one of the few companies with a large A100 cluster, High-Flyer and DeepSeek were able to attract some of China's best research talent, two former employees said.
"The key advantage of vast (computing) resources is that it allows for large-scale experimentation," said Liu, the former employee.
Some Western AI entrepreneurs, like Scale AI CEO Alexandr Wang, have claimed that DeepSeek had as many as 50,000 higher-end Nvidia chips that are banned for export to China.
He has not produced evidence for the allegation or responded to Reuters' requests to provide proof.
DeepSeek has not responded to Wang's claims.
Two former employees attributed the company's success to Liang's focus on more cost-effective AI architecture.
The startup used techniques like Mixture-of-Experts (MoE) and multihead latent attention (MLA), which incur far lower computing costs, its research papers show.
The MoE technique divides an AI model into different areas of expertise and activates only those related to a query, as opposed to more common architectures that use the entire model.
MLA architecture allows a model to process different aspects of one piece of information simultaneously, helping it detect key details more effectively.
While competitors like France's Mistral have developed models based on MoE, DeepSeek was the first firm to depend heavily on this architecture while achieving parity with more expensively built models.
DeepSeek's pricing was 20 to 40 times cheaper than what OpenAI charged for equivalent models, analysts at Bernstein brokerage estimated in early February.
For now, Western and Chinese tech giants have signaled plans to continue heavy AI spending, but DeepSeek's success with R1 and its earlier V3 model has prompted some to alter strategies.
OpenAI cut prices this month, while Google's Gemini has introduced discounted tiers of access.
Since R1's launch, OpenAI has also released an O3-Mini model that relies on less computing power.
Adnan Masood of U.S. tech services provider UST told Reuters that his laboratory had run benchmarks that found R1 often used three times as many tokens, or units of data processed by the AI model, for reasoning as OpenAI's scaled-down model.
STATE EMBRACE
Even before R1 gripped global attention, there were signs that DeepSeek had caught Beijing's favor.
In January, state media reported that Liang attended a meeting with Chinese Premier Li Qiang in Beijing as the designated representative of the AI sector, ahead of the leaders of better-known firms.
The subsequent fanfare over the cost competitiveness of its models has buoyed Beijing's belief that it can out-innovate the U.S., with Chinese companies and government bodies embracing DeepSeek models at a pace that has not been offered to other firms.
At least 13 Chinese city governments and 10 state-owned energy companies say they have deployed DeepSeek into their systems, while tech giants Lenovo, Baidu and Tencent - owner of China's largest social media app WeChat - have integrated DeepSeek's models into their products.
Chinese leader Xi Jinping and Li "have signalled they endorse DeepSeek," said Alfred Wu, an expert on Chinese policymaking at Singapore's Lee Kuan Yew School of Public Policy.
"Now everyone just endorses it."
The Chinese embrace comes as governments from South Korea to Italy remove DeepSeek from national app stores, citing privacy concerns.
"If DeepSeek becomes the go-to AI model across Chinese state entities, Western regulators might see this as another reason to escalate restrictions on AI chips or software collaborations," said Stephen Wu, an AI expert and founder of hedge fund Carthage Capital.
Further limits on advanced AI chips are a challenge that Liang has acknowledged.
"Our problem has never been funding," he told Waves in July.
"It's the embargo on high-end chips."
Additional reporting by Samuel Shen, Gu Li, Larissa Liao, Aditya Soni and Shanghai Newsroom; Editing by Brenda Goh and Katerina Ang
https://www.reuters.com/technology/arti ... 025-02-25/
Re: CHINA
REUTERS
"China says US plan to toughen semiconductor curb will backfire"
By Reuters
February 25, 2025
BEIJING, Feb 25 (Reuters) - The U.S. plan of coercing other countries into going after China's semiconductor industry will backfire, China's foreign ministry said on Tuesday.
Such actions by the United States will hinder development of the global semiconductor industry, said Lin Jian, a foreign ministry spokesperson, in a regular press briefing, when asked about the White House's plan to toughen semiconductor restrictions on China.
Reporting by Colleen Howe and Xiuhao Chen; Editing by Tom Hogue
https://www.reuters.com/technology/chin ... 025-02-25/
"China says US plan to toughen semiconductor curb will backfire"
By Reuters
February 25, 2025
BEIJING, Feb 25 (Reuters) - The U.S. plan of coercing other countries into going after China's semiconductor industry will backfire, China's foreign ministry said on Tuesday.
Such actions by the United States will hinder development of the global semiconductor industry, said Lin Jian, a foreign ministry spokesperson, in a regular press briefing, when asked about the White House's plan to toughen semiconductor restrictions on China.
Reporting by Colleen Howe and Xiuhao Chen; Editing by Tom Hogue
https://www.reuters.com/technology/chin ... 025-02-25/
Re: CHINA
REUTERS
"Exclusive: Nvidia's H20 chip orders jump as Chinese firms adopt DeepSeek's AI models, sources say"
By Fanny Potkin and Che Pan
February 24, 2025
Summary
* Alibaba, ByteDance, Tencent have boosted H20 purchases, sources say
* Smaller firms now buying servers with AI computing systems, source says
* Trump administration considering new H20 export controls, sources have said
SINGAPORE/BEIJING, Feb 25 (Reuters) - Chinese companies are ramping up orders for Nvidia's H20 artificial intelligence chip due to booming demand for DeepSeek's low-cost AI models, six people familiar with the matter said.
The surge in orders, which is being reported for the first time by Reuters, underlines Nvidia's dominance of the market and could help alleviate concerns that DeepSeek might cause a slide in AI chip demand.
Tencent, Alibaba and ByteDance have "significantly increased" orders of the H20 - a chip specific to China due to U.S. export controls - since the Chinese AI startup burst into the global public consciousness last month, two of the people said.
In addition to their internal needs for advanced AI chips, the three tech giants provide cloud computing services through which other firms can access and use AI tools.
Smaller companies in sectors like healthcare and education are also purchasing AI servers equipped with DeepSeek models and Nvidia H20 chips, said a source at one of China's largest server makers.
Previously only deep-pocketed financial and telecoms firms bought servers with AI computing systems, the source added.
U.S. President Donald Trump's administration is looking at imposing restrictions on the sale of the H20 chip to China, Reuters has reported.
While the threat of further controls could be a factor in the jump in orders, the sources cited DeepSeek as the reason.
The sources did not provide details on the size of the orders.
They were not authorised to speak to media and declined to be identified.
Nvidia did not respond to queries on how much demand for the H20 it was seeing from China but said its products won "on merit in a competitive field".
The company is set to report quarterly earnings on Wednesday.
Tencent, ByteDance and Alibaba did not respond to requests for comment.
DeepSeek's large language models rival Western systems in performance at a fraction of the cost as they focus on "inference" or producing conclusions.
That optimises computational efficiency rather than relying solely on raw processing power.
"When DeepSeek launched, many misjudged that computing power demand might stagnate or decrease."
"In reality, more advanced AI models drive deeper integration into daily life, exponentially increasing inference-level compute need," said Nori Chiou, investment director at Singapore-based White Oak Capital Partners.
A DeepSeek-induced global rout in AI stocks that began January 24 saw Nvidia shares lose as much as a fifth of their value at one point but they have since regained most of that ground and are down just 3% for the year to date.
Though wider deployment of DeepSeek AI models is expected to help Chinese chipmakers such as Huawei better compete in the domestic market thanks to the models' focus on inference, Nvidia's H20 chip remains the industry standard in China.
Analysts estimate Nvidia shipped approximately 1 million H20 units in 2024, generating over $12 billion in revenue for the company.
The H20 is the primary chip Nvidia is legally permitted to sell in China and was launched after the latest round of U.S. export restrictions took effect in October 2023.
Washington has banned exports of Nvidia's most advanced chips to China since 2022, concerned that advanced technologies could be used by China to build up its military capabilities.
Numerous Chinese companies have announced plans to use DeepSeek's models.
Among them are Tencent, which has said it will beta test integrating the models into its highly popular WeChat messaging app, and automaker Great Wall which has integrated DeepSeek's model into its connected vehicle system.
Reporting by Fanny Potkin in Singapore, Che Pan in Beijing and Brenda Goh in Shanghai; Editing by Edwina Gibbs
https://www.reuters.com/technology/arti ... 025-02-25/
"Exclusive: Nvidia's H20 chip orders jump as Chinese firms adopt DeepSeek's AI models, sources say"
By Fanny Potkin and Che Pan
February 24, 2025
Summary
* Alibaba, ByteDance, Tencent have boosted H20 purchases, sources say
* Smaller firms now buying servers with AI computing systems, source says
* Trump administration considering new H20 export controls, sources have said
SINGAPORE/BEIJING, Feb 25 (Reuters) - Chinese companies are ramping up orders for Nvidia's H20 artificial intelligence chip due to booming demand for DeepSeek's low-cost AI models, six people familiar with the matter said.
The surge in orders, which is being reported for the first time by Reuters, underlines Nvidia's dominance of the market and could help alleviate concerns that DeepSeek might cause a slide in AI chip demand.
Tencent, Alibaba and ByteDance have "significantly increased" orders of the H20 - a chip specific to China due to U.S. export controls - since the Chinese AI startup burst into the global public consciousness last month, two of the people said.
In addition to their internal needs for advanced AI chips, the three tech giants provide cloud computing services through which other firms can access and use AI tools.
Smaller companies in sectors like healthcare and education are also purchasing AI servers equipped with DeepSeek models and Nvidia H20 chips, said a source at one of China's largest server makers.
Previously only deep-pocketed financial and telecoms firms bought servers with AI computing systems, the source added.
U.S. President Donald Trump's administration is looking at imposing restrictions on the sale of the H20 chip to China, Reuters has reported.
While the threat of further controls could be a factor in the jump in orders, the sources cited DeepSeek as the reason.
The sources did not provide details on the size of the orders.
They were not authorised to speak to media and declined to be identified.
Nvidia did not respond to queries on how much demand for the H20 it was seeing from China but said its products won "on merit in a competitive field".
The company is set to report quarterly earnings on Wednesday.
Tencent, ByteDance and Alibaba did not respond to requests for comment.
DeepSeek's large language models rival Western systems in performance at a fraction of the cost as they focus on "inference" or producing conclusions.
That optimises computational efficiency rather than relying solely on raw processing power.
"When DeepSeek launched, many misjudged that computing power demand might stagnate or decrease."
"In reality, more advanced AI models drive deeper integration into daily life, exponentially increasing inference-level compute need," said Nori Chiou, investment director at Singapore-based White Oak Capital Partners.
A DeepSeek-induced global rout in AI stocks that began January 24 saw Nvidia shares lose as much as a fifth of their value at one point but they have since regained most of that ground and are down just 3% for the year to date.
Though wider deployment of DeepSeek AI models is expected to help Chinese chipmakers such as Huawei better compete in the domestic market thanks to the models' focus on inference, Nvidia's H20 chip remains the industry standard in China.
Analysts estimate Nvidia shipped approximately 1 million H20 units in 2024, generating over $12 billion in revenue for the company.
The H20 is the primary chip Nvidia is legally permitted to sell in China and was launched after the latest round of U.S. export restrictions took effect in October 2023.
Washington has banned exports of Nvidia's most advanced chips to China since 2022, concerned that advanced technologies could be used by China to build up its military capabilities.
Numerous Chinese companies have announced plans to use DeepSeek's models.
Among them are Tencent, which has said it will beta test integrating the models into its highly popular WeChat messaging app, and automaker Great Wall which has integrated DeepSeek's model into its connected vehicle system.
Reporting by Fanny Potkin in Singapore, Che Pan in Beijing and Brenda Goh in Shanghai; Editing by Edwina Gibbs
https://www.reuters.com/technology/arti ... 025-02-25/