Microsoft helped build AI in China. Chinese AI helped build Microsoft.

As “AI race” rhetoric reinforces fear of the U.S. losing dominance on the world stage, Microsoft could face hard decisions surrounding the robust AI ecosystem it helped build in China.

Microsoft's Beijing office

Microsoft Research Asia, or MSRA, has been known as one of the most influential hubs of AI research in the world, advancing speech recognition, natural language and image processing, and other deep-learning work.

Photo: Microsoft

Click here for more coverage

Through decades of support, Microsoft was an instrumental force helping China become the AI powerhouse it is today.

Now, as the very thought of a U.S. company partnering in tech projects in China draws scrutiny from lawmakers , national security hawks, and human rights advocates, Microsoft could be forced to grapple with tough decisions surrounding the thriving AI ecosystem it fostered there.

Microsoft established its research lab in Beijing in 1998, when it was a pioneer paving the way for AI research and business collaborations between the U.S. and China. It was three years before China joined the World Trade Organization , a time when President Bill Clinton actively pushed for closer trade ties with the country, and when AI was mostly the stuff of sci-fi pipe dreams.

Since then, Microsoft Research Asia, or MSRA, has been known as one of the most influential hubs of AI research in the world, advancing speech recognition, natural language and image processing, and other deep-learning work, spreading its discoveries far and wide.

Elements of research conducted at MSR China have been used to build Microsoft’s advertising, chatbots, Bing search, Windows, Xbox, Azure Cloud, and other products used everywhere. People who honed their skills at MSRA have moved on to help create and grow some of China’s best-known tech firms driving AI in the country and across the globe, including Alibaba, Baidu, and Tencent, as well as more controversial AI providers, including facial-recognition company Megvii.

“Basically you can argue that Microsoft Research Asia was the sort of seed capital from which a lot of Chinese AI companies and researchers and the sector really developed,” said Paul Triolo, senior vice president focused on China at global strategic consultancy Albright Stonebridge Group. “It’s proven to be a catalyst for the industry."

Microsoft’s storied lab in Beijing has helped the company cultivate AI talent and better understand China’s massive market. In 2018, Microsoft said it had invested more than $1 billion in research and development in China over the previous decade.

Basically you can argue that Microsoft Research Asia was the sort of seed capital from which a lot of Chinese AI companies and researchers and the sector really developed.

There are more than 300 scientists and engineers working at its MSRA labs in Beijing and Shanghai, but Microsoft has a much larger research and development group consisting of 6,000 scientists and engineers in Beijing, Shanghai, Shenzhen, and Suzhou as well as in Taiwan and Japan. That group, which includes a team focused on Cloud and AI technologies, helps transform research from MSRA into Microsoft products and services.

To help it get there, Microsoft has nurtured close ties with universities as well as state and local governments in the country.

Before the Trump administration launched its China initiative in 2018, “MSRA’s China government links would have been viewed as a major positive,” Triolo said.

But now the “D-word” is on many lips: decoupling. “Fast forward to the current situation, and being favored by the Chinese government is clearly viewed in Washington as a major negative,” Triolo said. “But the links are pretty deep and complex, and cannot be unraveled that easily, nor is there a strong desire on either side to break these ties.”

Still, Microsoft has made some changes. When the U.S. Treasury Department included China’s drone tech giant DJI on a list of sanctioned companies in December 2021, Microsoft ended a partnership with the company, which has been previously unreported. It has also been rumored that Microsoft has stopped recruiting from some Chinese universities.

Diversity of thought

When Microsoft opened an AI Innovation Center in Shanghai in conjunction with the Xuhui District government and a state-owned enterprise group there in 2018, the mayor of the district, Fang Shizhong, praised other “successful” collaborations with Microsoft, such as a startup accelerator program called Microsoft ScaleUp Shanghai .

The center would “further expand the depth and breadth of cooperation between Microsoft and Xuhui District, and will play an active role in bringing together talents, accelerating development, and leading innovation,” he said.

MSRA researchers have also assisted local Chinese companies on projects. In 2017, Xianyuan Zhan , a former MSRA associate researcher who got his Ph.D. from Indiana’s Purdue University, helped energy company China Energy Group use AI to make production at its coal-fired power plants more efficient.

“There was a senior leader from the China Energy Group who reached [out] to my manager at MSRA,” Zhan, now a research assistant professor at Tsinghua University’s Institute for AI Industry Research in China, told Protocol via video call from Beijing.

“The person from the China Energy Group said, ‘We have a complex problem. Maybe AI can help,’” Zhan said.

SHANGHAI, CHINA - AUGUST 06: Aerial view of Alibaba Shanghai R&D Center, ZJ Science City, IBM and Microsoft at AIsland on August 6, 2020 in Shanghai, China. (Photo by Shen Chunchen/VCG via Getty Images) Alibaba Shanghai R&D Center, ZJ Science City, IBM and Microsoft at AIsland in Shanghai. Photo: Shen Chunchen/VCG via Getty Images

When Zhan’s manager left MSRA to work for China’s fintech, AI, and cloud services company JD Technology later that year, Zhan joined him and carried on the project there. Since then, they’ve published some of that research . A Microsoft spokesperson confirmed that Zhan started the energy project while at MSRA and said the company did not continue the work after Zhan left.

More recently in 2021, Microsoft launched an international research community dedicated to AI for use in 5G networks, joining existing university partners in China including Tsinghua University, Nanjing University, and Peking University.

“The interaction between the top universities in the U.S., the top universities in China, now it’s almost on a day-to-day basis, and we also feel very honored to actually play a big role to make this the norm today,” said Hsiao-Wuen Hon, managing director, corporate vice president, and chairman of Microsoft’s Asia-Pacific R&D Group, during a 2018 episode of the Microsoft Research Podcast .

The contributions of Microsoft’s researchers in China “have and continue to benefit the international academic research community, which is why I believe it is important to invest in AI research in China,” said Peter Lee, corporate vice president of Microsoft research and incubations, in an email sent to Protocol.

He pointed to their 2014 discovery of a seminal deep-learning technique commonly known as ResNet , which the company helped share with the rest of the world.

Lee chalked up the success of ResNet to diversity of thought. MSRA’s China researchers came upon the breakthrough because they were prone to using an approach to AI problem-solving called backpropagation, one typically not considered by AI researchers in the U.S., he said.

“If this engagement had not happened, ResNet might have been left undiscovered outside of China. Instead, researchers in the U.S. and Europe have been able to learn from and build on this research,” Lee said.

Since ResNet was introduced, researchers at companies including Apple , DeepMind , and Facebook have conducted work involving the technique.

Work out of MSRA continues to be incorporated into other projects that are bound to be spread far and wide. For instance, in September when Meta announced its new Make-A-Video AI system that automatically creates video clips using text prompts, the company shared details of the components that went into it, some of which originated at MSRA in Beijing, including training and testing data sets .

'They’re scaring people'

But Microsoft’s AI-related ties in China go well beyond research. The company actively recruits people in China with AI-related expertise to work on its products for that market. It also partners on AI collaborations with China-based companies. For instance, a Microsoft principal software engineer demonstrated a machine-learning project in August conducted in conjunction with a ByteDance software engineer based in Seattle.

Despite the lab’s historic role bridging AI research discoveries between China and the rest of the world, these days Microsoft is not eager to broadcast it. Following several months of requests for interviews, the company would only agree to provide written email responses to Protocol.

Part of the reason for its reticence might be that the company’s expanding business in China has become increasingly sensitive in the U.S. Even when Microsoft announced plans in September to grow operations in China to coincide with its 30th anniversary in the country, the company did not publicize the news in English-language media.

“In celebration of Microsoft’s 30-year presence in China, the local team shared the company is on track to reach 10,000 employees in China — a 10% growth target disclosed more than four years ago,” a Microsoft spokesperson told Protocol, noting that the company is “very close” to hitting that 10,000 employee goal.

“All of a sudden, AI has been dragged into the U.S.-China geopolitical competition in ways that companies like Microsoft and others in the sector were not really expecting,” Triolo said.

“If they have researchers, for example, that are collaborating with a particular Chinese individual who happens to be at an institute or a company that all of a sudden is now sensitive, then this becomes a big issue for the company in terms of reputational risk,” he said.

Microsoft is hardly alone in its reluctance to talk about its AI tech work in China. Heightened geopolitical tensions amid a U.S. government crackdown on research and tech partnerships with China — driven by national security , economic, and human rights concerns — have influenced many U.S. companies to keep mum about their AI business in China. The political rhetoric has formed a cloud of anti-China sentiment that has some Chinese AI businesspeople feeling fearful.

All of a sudden, AI has been dragged into the U.S.-China geopolitical competition in ways that companies like Microsoft and others in the sector were not really expecting.

“[U.S. government officials are] scaring people,” said a U.S. citizen of Chinese descent who works as the chief scientist of a U.S. business that sells AI software to companies in mainland China. The source asked not to be named in this story for fear of political retribution.

“It’s why we are taking very regular steps to make sure we’re compliant,” said the scientist.

Most of the top U.S. tech companies building AI today have operations in China. Amazon has AI development labs in Shenzhen and Shanghai and says it has a “ long-term commitment in China .” Although Google closed an AI center it opened in Shanghai in 2017 , the company has recruited people with AI expertise recently to join teams in China. Apple also does AI work in China and recently looked for people with machine-learning skills to work at its China Vision Lab in Beijing. Meanwhile, several China-based tech companies engaged in AI-related work have offices or labs in the U.S., including Alibaba, Baidu, ByteDance, and Tencent.

But the increased pressures around competition between the U.S. and China are taking a toll on businesses. The number of U.S. companies doing business in China that reported negative business impacts such as lost sales and supply sourcing disruptions in June reached 87%, up from 73% in 2018, according to the U.S.-China Business Council, a group that counts Amazon, Apple, eBay, Meta, Intel, IBM, and Microsoft among its members .

Next stop: Tianjin

Microsoft’s AI research team in China has been instrumental in the company’s AI advancements.

Ten years ago, a team of AI researchers crowded around Rick Rashid’s desk in his cramped office at Microsoft’s Redmond campus. It was time to test an automated speech recognition system Rashid had challenged the company’s researchers in China and Redmond three years prior to build.

“The teams in China picked up on this, working with our guys in Redmond, and really tried to integrate a system that made heavy use of this new neural network technology, deep neural networks,” said Rashid, a founder of Microsoft Research who would later lead the division.

Now the moment of truth had arrived. “We set it all up. We had the computer there, nice microphone. And we were talking about it. And I noticed that [while] we had the system running, that it was actually transcribing our conversation really, really well, and doing a great job of it. And that gave me a lot of confidence that when the time came to do a speech in China, that that would probably work,” Rashid told Protocol.

The excitement was palpable. Rashid finally knew that the groundbreaking system — designed to recognize speech, transcribe it to text, then translate it from English to Chinese in real time — worked well enough to show the world.

Rashid’s next stop was Tianjin, where the MSRA research team would hold an annual lavish event demonstrating its latest achievements.

“I got up onstage in Tianjin . And I gave my little section of the speech that was actually translated in real time in Chinese. It was just an amazing experience, because the audience was just stunned,” he said. “There were literally students in the audience that I could see crying, because I assumed they never expected they would see something like that happen."

Cool research projects are one thing. Even in the early days, Microsoft’s China lab produced a variety of research that ended up in products such as its DirectX multimedia game installer, said Nathan Myhrvold, another co-founder of Microsoft Research and today the founder of tech startup investment firm Intellectual Ventures.

Almost all major Microsoft products over the years until today all have the MSRA contribution.

“China did a lot of image processing work that wound up being in some of our games; they wound up being in the DirectX platform stuff, which was the graphics code. They contributed along with others to the vision system that we used in the game thing for a while,” Myhrvold told Protocol.

“Almost all major Microsoft products over the years until today all have the MSRA contribution,” Hon said on the 2018 Microsoft Research Podcast. “We also have several technologies inside Microsoft [that were] totally started [by] MSRA."

China’s Big Tech alma mater

But it wasn’t just applicable research MSRA was producing. It was AI talent.

“Instead of waiting for folks to come to the U.S., get their Ph.D. or whatever, and then hire them into MSR, we could just go directly there,” said Kevin Goldsmith, chief technology officer at open-source AI software company Anaconda, who joined Microsoft Research’s graphics group in 1994.

“And obviously China is a massive market. It was good to have folks who were there who could actually understand the market, work with the market better, help the company produce things for that market in a more locally aware way.”

MSRA has also become an alma mater for some of China’s top corporate AI engineers and executives. Baidu’s chief scientist, Jingdong Wang, who specializes in computer vision-related work, joined the company after 14 years at MSRA. Wei-Ying Ma, now a chair professor at Tsinghua University, became vice president at ByteDance’s AI Lab after serving as assistant managing director at MSRA, overseeing research groups focused on machine learning and natural-language computing.

Earlier this year, computer vision researchers mourned the untimely death of Jian Sun , a former research manager at MSRA who contributed to the important ResNet work there. Sun joined Megvii as its chief scientist and managing director of research in 2016. Three years later, the company was added to the U.S. entity list of sanctioned companies for enabling human rights violations against Muslim minority groups in China.

Perhaps MSRA’s biggest star is Kai-Fu Lee, founder of MSR’s China lab who went on to start Google’s operation in China in 2005. Today Lee is a prominent investor in AI companies in China and an evangelist of the country’s AI advancements.

Recovering lost revenue was also a key factor pushing Microsoft into China. The company hoped being there might help it curb rampant software piracy, said Usama Fayyad, executive director at the Institute for Experiential AI at Northeastern University, who worked as senior researcher and manager at Microsoft Research from 1996 to 2000.

“China had very little regard for intellectual property , and there was a lot of unlicensed Microsoft software being used in China and being distributed by entities in China,” he said.

“So, [Microsoft wanted] collaboration with the government and institutions there to help them realize there is value to this IP. We can cooperate and help you advance in research and computer science; at the same time, you need to also appreciate the value to us and help us figure out what is the right answer to get the right laws in place to stop a lot of the abuses in software piracy,” Fayyad said.

And, while national security and intelligence authorities in the U.S. often cite risk of intellectual property theft among key concerns driving the U.S. scrutiny on AI from China, security was top of mind for Microsoft from the start, Goldsmith said.

“If you went to China, you couldn’t bring company equipment. You’d get a separate piece of equipment to bring that you would immediately dispose of when you left the country. And that was all around concerns around intellectual property,” he said.

Fraying ties

As the U.S. widened its sanctions net to include more Chinese technology companies, Microsoft felt the effects.

The company was able to maintain a deal with Huawei after the U.S. restricted partnerships with the telco in 2019. However, the outcome was different when drone maker DJI was sanctioned by the U.S. in 2021. As a result, the companies ended a partnership allowing developers to build Windows applications to remotely pilot DJI drones , a Microsoft spokesperson told Protocol.

The Treasury Department sanctioned DJI and other tech companies for “actively [supporting] the biometric surveillance and tracking of ethnic and religious minorities in China, particularly the predominantly Muslim Uyghur minority in Xinjiang.”

Rumors have emerged that MSRA has severed ties with universities in China listed on the U.S. government’s entities list that are believed to conduct research with military applications. In August, the U.S. Commerce Department added some technology-focused schools with military ties to the list.

Microsoft declined to confirm or deny the reports, but a spokesperson said, “Microsoft fully complies with U.S. laws — including U.S. export controls and sanctions laws and regulations.”

What the future holds for Microsoft’s entrenched AI partnerships in China remains to be seen. “It is possible that the U.S. government could put pressure on Microsoft to not pursue certain types of AI research at MSRA, but this would be a major issue for Microsoft in the AI domain,” Triolo said.

He suggested Microsoft could move AI researchers in China to Redmond, “forming a kind of ‘away team’ in the U.S.” However, he said, “That would come with costs, and likely undermine morale and some of the relationships the firm has built in China via MSRA over the years.”


Judge Zia Faruqui is trying to teach you crypto, one ‘SNL’ reference at a time

His decisions on major cryptocurrency cases have quoted "The Big Lebowski," "SNL," and "Dr. Strangelove." That’s because he wants you — yes, you — to read them.

The ways Zia Faruqui (right) has weighed on cases that have come before him can give lawyers clues as to what legal frameworks will pass muster.

Photo: Carolyn Van Houten/The Washington Post via Getty Images

“Cryptocurrency and related software analytics tools are ‘The wave of the future, Dude. One hundred percent electronic.’”

That’s not a quote from "The Big Lebowski" — at least, not directly. It’s a quote from a Washington, D.C., district court memorandum opinion on the role cryptocurrency analytics tools can play in government investigations. The author is Magistrate Judge Zia Faruqui.

Keep Reading Show less
Veronica Irwin

Veronica Irwin (@vronirwin) is a San Francisco-based reporter at Protocol covering fintech. Previously she was at the San Francisco Examiner, covering tech from a hyper-local angle. Before that, her byline was featured in SF Weekly, The Nation, Techworker, Ms. Magazine and The Frisc.

The financial technology transformation is driving competition, creating consumer choice, and shaping the future of finance. Hear from seven fintech leaders who are reshaping the future of finance, and join the inaugural Financial Technology Association Fintech Summit to learn more .

Keep Reading Show less
The Financial Technology Association (FTA) represents industry leaders shaping the future of finance. We champion the power of technology-centered financial services and advocate for the modernization of financial regulation to support inclusion and responsible innovation.

AWS CEO: The cloud isn’t just about technology

As AWS preps for its annual re:Invent conference, Adam Selipsky talks product strategy, support for hybrid environments, and the value of the cloud in uncertain economic times.

Photo: Noah Berger/Getty Images for Amazon Web Services

AWS is gearing up for re:Invent, its annual cloud computing conference where announcements this year are expected to focus on its end-to-end data strategy and delivering new industry-specific services.

It will be the second re:Invent with CEO Adam Selipsky as leader of the industry’s largest cloud provider after his return last year to AWS from data visualization company Tableau Software.

Keep Reading Show less
Donna Goodison

Donna Goodison ( @dgoodison ) is Protocol's senior reporter focusing on enterprise infrastructure technology, from the 'Big 3' cloud computing providers to data centers. She previously covered the public cloud at CRN after 15 years as a business reporter for the Boston Herald. Based in Massachusetts, she also has worked as a Boston Globe freelancer, business reporter at the Boston Business Journal and real estate reporter at Banker & Tradesman after toiling at weekly newspapers.

Image: Protocol

We launched Protocol in February 2020 to cover the evolving power center of tech. It is with deep sadness that just under three years later, we are winding down the publication.

As of today, we will not publish any more stories. All of our newsletters, apart from our flagship, Source Code, will no longer be sent. Source Code will be published and sent for the next few weeks, but it will also close down in December.

Keep Reading Show less
Bennett Richardson

Bennett Richardson ( @bennettrich ) is the president of Protocol. Prior to joining Protocol in 2019, Bennett was executive director of global strategic partnerships at POLITICO, where he led strategic growth efforts including POLITICO's European expansion in Brussels and POLITICO's creative agency POLITICO Focus during his six years with the company. Prior to POLITICO, Bennett was co-founder and CMO of Hinge, the mobile dating company recently acquired by Match Group. Bennett began his career in digital and social brand marketing working with major brands across tech, energy, and health care at leading marketing and communications agencies including Edelman and GMMB. Bennett is originally from Portland, Maine, and received his bachelor's degree from Colgate University.


Why large enterprises struggle to find suitable platforms for MLops

As companies expand their use of AI beyond running just a few machine learning models, and as larger enterprises go from deploying hundreds of models to thousands and even millions of models, ML practitioners say that they have yet to find what they need from prepackaged MLops systems.

As companies expand their use of AI beyond running just a few machine learning models, ML practitioners say that they have yet to find what they need from prepackaged MLops systems.

Photo: artpartner-images via Getty Images

On any given day, Lily AI runs hundreds of machine learning models using computer vision and natural language processing that are customized for its retail and ecommerce clients to make website product recommendations, forecast demand, and plan merchandising. But this spring when the company was in the market for a machine learning operations platform to manage its expanding model roster, it wasn’t easy to find a suitable off-the-shelf system that could handle such a large number of models in deployment while also meeting other criteria.

Some MLops platforms are not well-suited for maintaining even more than 10 machine learning models when it comes to keeping track of data, navigating their user interfaces, or reporting capabilities, Matthew Nokleby, machine learning manager for Lily AI’s product intelligence team, told Protocol earlier this year. “The duct tape starts to show,” he said.

Keep Reading Show less
Kate Kaye

Kate Kaye is an award-winning multimedia reporter digging deep and telling print, digital and audio stories. She covers AI and data for Protocol. Her reporting on AI and tech ethics issues has been published in OneZero, Fast Company, MIT Technology Review, CityLab, Ad Age and Digiday and heard on NPR. Kate is the creator of RedTailMedia.org and is the author of "Campaign '08: A Turning Point for Digital Media," a book about how the 2008 presidential campaigns used digital media and data.

Latest Stories