Chinese and American scientists and businessmen inside of an old computer
Illustration: Mari Fouz; Photos: Getty Images

The great AI race that wasn’t

Protocol Enterprise

Hello, and welcome to Protocol Enterprise! Today: lessons learned from our big series unpacking the narratives that surround the development of AI in the U.S. and China, why Twilio flew too close to the sun, and Nvidia plots a way around new U.S. export controls.

The misconceptions underpinning the US-China AI race

When I joined Protocol a year ago, I began to realize just how much the notion that the U.S. and China are in an “AI race” was driving AI industry PR, media coverage, and U.S. policy on AI and China.

But the “AI race” narrative hinges on a slew of misconceptions and profit motives that could counteract the goals of the U.S. and deflect attention from federal AI and data privacy regulations. A series we published throughout the past week illuminates and scrutinizes the assumptions driving the so-called AI race narrative through six feature stories, compelling audio interviews, and a virtual event.

The misconceptions? Oh, the misconceptions:

AI can be “won.”

  • While reporters love asking who will “win” the AI race, many of my sources who are AI/ML engineers and practitioners told me it’s a faulty construct.
  • “I have no idea what the AI finish line would be,” said Rick Rashid, a founder of Microsoft Research who helped set up its legendary lab in Beijing, which is featured in a series story analyzing the past and future of Microsoft Research Asia in China.

AI is built within, and confined to, country borders.

  • As my story delving into open-source AI collaboration between the U.S. and China shows, the foundational technical components used to build AI products have been developed and evolved through open-source contributions made by people around the world.
  • Not only do many policymakers not realize this, but they may also not realize that some popular foundational AI frameworks, data sets, and machine learning models have either been developed in China or have been built using a combination of components from Chinese and American AI companies.

The U.S. can attract AI technologists from China while attacking the technology they build.

  • As former Google CEO and AI investor Eric Schmidt and others, including national security adviser Jake Sullivan, have framed their AI labor recruitment drive as a battle against China, they have helped sow suspicion of the very researchers they aim to attract.
  • As I detail in my profile of Schmidt, who stands to benefit financially from the anti-China AI crusade, he himself has implied that AI built in China is ethically dubious.
  • Chinese researchers in the U.S. are bearing the brunt of the suspicion, as evidenced by data from a recent survey of Chinese American computer science, engineering, and match researchers in the U.S., 61% of whom said they had thought about leaving the U.S.

AI from the U.S. embeds “democratic values” while AI from China is ethically flawed.

  • Ever since an influential final report from the Schmidt-led National Security Commission on Artificial Intelligence claimed that “The AI competition is also a values competition,” demanding that the federal government spend billions more on AI R&D to combat China’s AI advancements, the simplistic idea has spread.
  • “[A lot of people] have this notion that AI that’s developed in China somehow embeds a different system of ethics and values that's uniquely Chinese,” Rebecca Arcesati, an analyst at the Mercator Institute for China Studies, told me, calling the idea an “Orientalist trap.”
  • This concept — explored in a story published today — not only propels assumptions that fuel increased government funding for and procurement of AI software, but it could allow the U.S. to deflect data or AI regulations based on an incorrect assumption that China has no data or AI regulations of its own.

The series addresses these issues and more, and is intended to add nuance to an important conversation with serious implications. I hope you’ll check it out and let me know what you think!

— Kate Kaye (email | twitter)


As companies scale in the cloud, they face new data management challenges. Watch our webinar to hear Capital One and Snowflake experts discuss operationalizing data mesh, and learn how Capital One approached scaling its data ecosystem.

Watch on demand

Too big, too fast

After years of explosive growth, Twilio now finds itself in cost-cutting mode. Over the past several months, the communications API provider has slowed hiring, closed offices, and announced restructuring plans that would lay off 11% of its workforce.

Macroeconomic factors are partially to blame for this, but the company also made some internal missteps, chief operating officer Khozema Shipchandler told me last week.

During Twilio’s investor day on Thursday, company executives admitted they tried to expand the company too quickly, adding excessive head count, failing to integrate sales teams properly, and struggling with the trade-offs between profitability and growth.

For instance, as demand for its communications APIs drove more than 50% growth for Twilio, the company drastically scaled its sales team and operations. But the company overcorrected, creating a bloated sales force it would later need to cut.

“And I think the gist of it is, that it just got too big too fast, and that's what you've seen happening a lot in tech recently,” said Shipchandler.

Now Twilio is trying to cut some of the fat by reducing sales and marketing spending and shifting future hiring to lower-cost geographies. All of this restructuring is estimated to reduce operating expenses by more than $300 million and put the company on track to reach profitability next year.

Moving forward, Twilio is pulling its annual revenue growth target and taking a much more measured approach to spending and hiring. But Shipchandler isn’t worried: “[M]ake no mistake, we’re still a very high grower; we intend to be for some time,” he said.

Read more about Twilio’s plans.

— Aisha Counts (email | twitter)

Whack a chip

In response to the Biden administration’s sweeping controls on chip tech exported to China, Nvidia has begun to sell a modified version of one of its advanced AI chips that it says meets the new standard set by the Commerce Department.

Nvidia’s China-specific chip is called the A800, and it went into production during the company’s fiscal third quarter, a spokesperson said in an email. The new chip is a substitute for the company’s flagship A100 processor that has been blocked for export to China and Russia.

The company says that it passes the “U.S. government’s clear test for reduced export control, and cannot be programmed to exceed it.” According to the Commerce Department, “Items with lower level capabilities may still be exported/reexported to or transferred within China but are still subject to U.S. restrictions based on their end uses, end users, or other restrictions and we encourage companies to conduct robust due diligence and invest in effective compliance when transacting with PRC entities,” it said in a statement.

Nvidia’s decision to launch a new AI chip product is an early example of how the semiconductor industry is adjusting its strategy to comply with export controls that now require licenses to export advanced AI chips and the tools that manufacture them. Protocol first reported on the administration’s objectives and plan to implement the controls in August.

Based on public disclosures to date, Nvidia looks to suffer the most immediate, direct impact from the new rules of the large chip designers. Rival chipmakers Intel and AMD have not yet disclosed material revenue damage from the China-related controls.

Nvidia said earlier this year the new rules on exports to China outlined in a notification letter would cost the company as much as $400 million in lost quarterly revenue.

The chip manufacturing tool makers based in the U.S. have disclosed the most significant impact to revenue. The Fremont-based Lam Research told investors it expected as much as a $2.5 billion hit to the top line in 2023, for example.

— Max A. Cherney (email | twitter)

Around the enterprise

Salesforce is planning to lay off thousands of employees by the Thanksgiving break, and has already notified hundreds of people in its sales department that they were affected by the cuts, Protocol’s Joe Williams reported.

Zoom launched new email and calendar software products in hopes of competing with Microsoft Office and Google Workspace as the company searches for growth beyond its video-conferencing platform.

Microsoft released patches for a pair of Exchange vulnerabilities more than five weeks after the disclosure of the flaws, which include the “ProxyNotShell” remote code execution vulnerability.


As companies scale in the cloud, they face new data management challenges. Watch our webinar to hear Capital One and Snowflake experts discuss operationalizing data mesh, and learn how Capital One approached scaling its data ecosystem.

Watch on demand

Thanks for reading — see you tomorrow!

Recent Issues

The AI talent gap