Trending Now
We have updated our Privacy Policy and Terms of Use for Eurasia Group and its affiliates, including GZERO Media, to clarify the types of data we collect, how we collect it, how we use data and with whom we share data. By using our website you consent to our Terms and Conditions and Privacy Policy, including the transfer of your personal data to the United States from your country of residence, and our use of cookies described in our Cookie Policy.
{{ subpage.title }}
Biden tightens China’s access to chips one last time
Throughout Joe Biden’s presidency, the Commerce Department has gradually tightened its chokehold on China’s access to semiconductors needed to access, train, and build artificial intelligence. On Dec. 2, Commerce Secretary Gina Raimondo announced what she told reporters amounted to the “strongest controls ever” meant to restrict China’s access to AI for military applications. Today, China responded with its own new restrictions, sending a strong signal to the incoming US president.
The new US controls announced Monday, the third order in as many years, apply to 24 types of semiconductor manufacturing equipment, three types of software tools, and high-bandwidth memory, or HBM, an interface often used in producing AI chips. The department also added 140 Chinese companies to its Entity List, which requires regulatory approval should a US company wish to sell to a member of the list. “By adding key semiconductor fabrication facilities, equipment manufacturers, and investment companies to the Entity List, we are directly impeding the PRC’s military modernization, WMD programs, and ability to repress human rights,” said Matthew Axelrod, assistant secretary for export enforcement at the Commerce Department.
In response, on Dec. 3, China banned shipments of certain materials using gallium, germanium, and antimony to the US, as well as super-hard materials such as diamonds. These items can be used both for military and semiconductor applications. “China firmly opposes the US overstretching the concept of national security, abuse of export control measures, and illegal unilateral sanctions and long-arm jurisdiction against Chinese companies,” said Lin Jian, a Chinese Foreign Ministry spokesperson.
Jacob Feldgoise, an analyst at Georgetown University’s Center for Security and Emerging Technology, said the new US order plugged holes in the previous year’s rules. It requires a license for many more exported tools, focuses on high-bandwidth memory “because HBM is used by nearly all of the most capable AI chips” and strengthens the US’s grasp beyond its borders. “Notably, this set of controls is newly extraterritorial: It will impose licensing requirements on certain foreign-produced tools so long as they contain US technology,” Feldgoise said.
Xiaomeng Lu, director of Eurasia Group's geo-technology practice, noted that the US excluded the Chinese semiconductor company ChangXin Memory Technologies from the Entity List to appease the Japanese government. CXMT has been buying materials from Japanese suppliers to make its memory chips. “With the Trump administration on its way, they are expected to take a more unilateral approach and will be less likely to make concessions per requests of allies,” she said.
Jeremy Mark, a nonresident senior fellow at the Atlantic Council's GeoEconomics Center, said it’s difficult to judge how significant these new rules are because of the looming change of guard in the White House. Had they come ahead of the transition to a Kamala Harris administration, “they would continue making life complicated for Chinese semiconductor companies and US companies that rely on the China market for a significant portion of their sales.” However, Mark said that Donald Trump could strengthen or weaken export controls when he takes office, so it’s “impossible to say” what the legacy of this final move will be.
For Biden, it marks the end of an era of success: While his restrictions on China could have been tighter or less porous, he leaves office with China still searching for AI breakthroughs. The US, at least under Biden’s watch, is still on top.
But China’s next-day retaliation shows that it is ready to play hardball ahead of the incoming Trump administration. Beijing understands that diplomacy alone might not do the trick, and that to succeed in getting America to the bargaining table it needs to safeguard its own crucial resources. “This is a step up in China’s reaction to US technology sanctions,” Lu said. “China is very frustrated with the lack of communication channels with the incoming administration. They are trying to send a shot across the bow to get attention from the Trump team.”
A logo of the autonomous driving technology startup Pony.ai is seen on a screen during an event in Beijing, China, in 2021.
Hard Numbers: Pony time, Book deals, ByteDance sues an intern, Japan’s investment, Your death clock is ticking
13: Pony AI, a Chinese robotaxi company debuted on the Nasdaq stock exchange, the latest Chinese tech company to enter the US public markets. The company issued an initial public offering at $13 per share on Nov. 27 about two years after China started a high-profile crackdown on its companies listing on US markets. It raised $260 million during its IPO, with Bloomberg remarking that it signaled “strong investor interest” in the company.
8,000: A venture-backed startup named Spines plans to publish 8,000 books next year, charging authors $1,200–$5,000 for the production process, including AI-assisted proofreading, design, and distribution. While Spines says it can offer opportunities to would-be writers and save them weeks of labor, traditional publishing houses have criticized the startup for trying to make money off of these writers with technology that makes many in the industry uncomfortable.
1.1 million: ByteDance, the Chinese company that owns TikTok, is suing a former intern in a Beijing court for $1.1 million, alleging that the intern deliberately sabotaged its generative AI training model by manipulating and modifying its code. The company said, however, that rumors that it lost millions of dollars and thousands of powerful graphics processors were exaggerated.
9.9 billion: The Japanese government is earmarking an extra $9.9 billion for its semiconductor and artificial intelligence ambitions. Some of that money will likely go to Rapidus, the homegrown chipmaking initiative that’s been heavily funded by the Japanese government, which aims to achieve mass production by 2027.
1,200: Want to know when you’ll die? Death Clock, an AI-powered longevity app trained on 1,200 life expectancy studies with 53 million participants, promises to tell users exactly when they’re going to perish. The app costs $40 a year and suggests lifestyle changes to users so they can delay their ticking countdown.
Intel CEO Pat Gelsinger delivers a speech at the COMPUTEX forum in Taipei, Taiwan, on June 4, 2024.
Intel is ready to move forward — without its CEO
Now, time has run out for the man in charge. Pat Gelsinger, CEO of Intel, resigned after being forced out of the company on Dec. 2. Gelsinger spent nearly four years at the helm of the company, and he oversaw a 30% revenue decline between 2021 and 2023, a trend capped off by cutting 15,000 jobs. Executives David Zinsner and Michelle Johnston Holthaus will helm the company while the board of directors searches for a new CEO.
Intel is an integrated chipmaker, meaning it designs and manufactures its own chips. That’s a different model than Nvidia, a chip designer that sends its units to a contract manufacturer such as Taiwan Semiconductor Manufacturing Company to be made. But over the years it has lost ground on both the design and manufacturing fronts, and investors have grown impatient. The company’s stock has fallen nearly 50% since the beginning of 2024, even as the Biden administration has agreed to give Intel $7.86 billion to build new facilities in Arizona, New Mexico, Ohio, and Oregon.
Whoever assumes Intel’s top job will oversee a legend of American technology that has failed to meet the AI moment. That person must improve the technology and the manufacturing process to have a chance at positioning Intel at the forefront of AI. No small task — probably one of the biggest jobs in American industry.
Illustration of the Amazon logo next to a phone with the flagship artificial intelligence model codenamed OLYMPUS, on it.
Amazon is set to announce its newest AI model
The multimodal model, codenamed Olympus, can reportedly search video archives using text commands. It’s unclear whether it can generate video content like OpenAI’s Sora or Meta’s Movie Gen, text-to-video models that are still not broadly released to the public.
But the new model is a sign not only of Amazon’s internal ambitions but also its potentially decreasing reliance on a key investment: Anthropic, the maker of the chatbot Claude.
On Nov. 22, Amazon announced it’s investing another $4 billion into Anthropic, doubling its total investment to $8 billion. In exchange, Anthropic would commit to using Amazon’s Tranium series of chips, Amazon’s “moonshot” attempt to rival Nvidia, the world’s leading AI chip designer.
Amazon already has an enterprise chatbot called Q, as well as AI business solutions for companies through its Amazon Web Services cloud offerings. Olympus could be announced as soon as the annual AWS re:Invent conference being held this week in Las Vegas, Nevada. Matt Garman, who took over as CEO of AWS in May, will address conference-goers on Tuesday and disclose “real, needle-moving changes” on AI.
If Olympus is indeed a business-to-business offering from AWS, then perhaps Anthropic’s Claude will continue being Amazon’s consumer-facing bet while it focuses on the more lucrative work of selling to other companies.
Sam Altman is seen on a mobile device screen next to an OpenAI logo in this illustration.
Can OpenAI reach 1 billion users?
How will it woo them? The startup is set to develop AI “agents” that can complete tasks for users rather than simply chat with them and launch its own search engine while further integrating ChatGPT with Apple products.
OpenAI, which Microsoft backs to the tune of $13 billion, wants to secure its financial future. (Microsoft has been building up its own internal AI capabilities and now considers OpenAI a “competitor.”) One way for OpenAI to grow is by adjusting its subscription revenue model. The company is reportedly considering expanding into advertising as a potential revenue model and hiring ad execs from top tech companies. The AI search engine Perplexity has already integrated ads into its business.
But it is also considering lowering its long-term costs by building data centers across the United States, something cofounder and CEO Sam Altman reportedly discussed with President Joe Biden at the White House in September. Chris Lehane, head of global policy at OpenAI, told the Financial Times that the company needs “chips, data and energy” to meet its expansion goals. Altman has previously expressed interest in raising trillions of dollars for a chip startup, though that hasn’t yet amounted to anything. Altman has, however, invested in Oklo, a nuclear power startup, that could power energy-intensive data centers.
Infrastructure investments could be key to a sustainable future as it grows — the company is reportedly losing billions a year training and deploying its models. But as Silicon Valley startups often go, profitability — or breaking even — could come long after achieving a user base in the billions.
President-elect Donald Trump points his finger at the Palm Beach County Convention Center on Nov. 6, 2024.
Trump wants a White House AI czar
If appointed, this person would be the White House official tasked with coordinating the federal government’s use of the emerging technology and its policies toward it. And while the role will not go to Elon Musk, the billionaire tech CEO who has been named to run a government efficiency commission for Trump, he will have input as to who gets the job.
The Trump administration has promised a deregulatory attitude toward artificial intelligence, including undoing President Joe Biden’s 2023 executive order on AI.
That order not only tasked the federal departments and agencies with evaluating how to regulate the technology given their statutory authority but also how to use it to further their own goals. Under Biden, each agency was tasked with naming a chief AI officer. If Trump is to keep those positions, the White House AI czar would likely coordinate with these officials across the executive branch.Will AI companies ever be profitable?
The leading AI startups and Big Tech incumbents are striving to make massive technological strides, but they still have an open question to answer: Can they actually make money?
What’s the business model for AI?
The Big Tech outlook: The largest technology companies — such as Amazon, Apple, Google, Meta, and Microsoft — are all-in on artificial intelligence. Many of them have their own large language models, such as Google’s Gemini 1.5 Pro and Meta’s open-source Llama 3. For these companies, betting on AI is one of many bets: Google is still a search engine and an advertising company, Meta is still a social media giant, but they’re betting that they can make strides in developing this breakthrough technology. Depending on how you look at it, these AI plays are either loss leaders or long-term investments to keep these companies on the cutting edge and boost their existing product lines.
And what the Silicon Valley giants can’t do themselves, they’re happy to outsource to specialized startups. For example, Amazon invested $8 billion into Anthropic, and Microsoft poured $13 billion into OpenAI. While Microsoft develops its own AI models, it has also integrated OpenAI into its Copilot assistant across its enterprise suite of products.
The startups: But without other products to cross-subsidize their ambitions, the smaller pure-play AI startups are left hunting for rock-solid business models.
OpenAI, which makes ChatGPT, lets users test out the chatbot for free but sells subscriptions for their most advanced tools and higher usage rates for $20 a month. The company also sells enterprise subscriptions to companies — reportedly about $60 per user each month for companies with 150 employees or more, which comes out to $108,000 per year for a company of that size. And it’s found some success with more than 1 million paid users of the enterprise version of ChatGPT — about $720 million in revenue. Other AI startups, such as Anthropic (which makes Claude), the search engine Perplexity, the image generator Midjourney, and the music generator Suno, have similar freemium models with bigger checks coming from business-to-business sales.
“The real money will be in business-to-business AI solutions provided they’re carefully deployed securely — something that the likes of Salesforce and Microsoft are promising,” said Gadjo Sevilla, senior analyst at the market research firm eMarketer. “This is easy for companies with large captive user bases since AI features will be an incremental cost to existing services and are also scalable across enterprises.”
The open-sourcers: While most AI companies have proprietary (or closed-source) models, a few have opted for open-source development, whereby they publish their code for free for developers to use and adapt it. Stability AI, which makes the open-source image generator Stable Diffusion, lets people use its model for free but charges companies that make $1 million or more in annual revenue for commercial licenses and support. That’s a monetization strategy that Meta could pursue in the future for its currently free and open-source Llama models.
The government option: AI companies have a third source of revenue beyond consumers and businesses: governments. OpenAI has secured contracts with US government agencies and public institutions as varied as NASA, the National Gallery of Art, the IRS, and Los Alamos National Laboratory, according to FedScoop. Microsoft, meanwhile, has AI deals with both the US and UK governments. And specialized firms like Palantir and Anduril have capitalized on US defense contracts with their AI technology for battlefields.
Running at a loss
OpenAI is currently valued at $157 billion, but the company behind the ChatGPT chatbot is still losing money. In September, the New York Times reported that OpenAI expects to make $3.7 billion in 2024, but it’s set to spend $5 billion in the process — a net loss of $1.3 billion.
The company’s internal projections estimate that revenues will hit $11.6 billion in 2025, but it will need to keep its costs — on training its models, running its services, and paying employees — stable to turn a profit. Meanwhile, Anthropic is reportedly burning through $2.7 billion this year. These companies’ top costs are computing infrastructure such as servers and chips, staffing with top talent, and the cost of offering free services to casual users.
To become profitable, these companies must lower costs, raise prices, or develop in-house capabilities like chips and data centers to reduce reliance on paying other firms.
Playing the long game
Perhaps AI startups need to think like the tech giants and play the long game. After all, these many billions of dollars in funding should give them some runway.
Sevilla said that OpenAI is headed in the right direction. “OpenAI shifting from nonprofit think tank to a for-profit AI innovator now behooves the company to generate sustainable profits,” he said, referring to the company’s recent change in ownership structure. “It’s challenging Google in search and browsers, it's trying to make inroads into education, and there's a good chance it will develop its own hardware to reduce reliance on Nvidia. Any of these areas can generate profits, but it could take time.”
Dev Saxena, director of Eurasia Group’s geo-technology practice, said the real value lies in building platforms that other companies will use to develop their own AI applications — “the same way that the internet unleashed so much entrepreneurialism and innovation."
In other words: The winners of the AI race might not be the companies with the most advanced AI but those who build the infrastructure and platforms other businesses need — and those who find a way to make money doing it.
The logo of Huawei's global flagship store is displayed in the Huangpu district of Shanghai, China.
The US is thwarting Huawei’s chip ambitions
The US government under President Joe Biden has imposed significant export controls not only on US-made chips but also on semiconductor manufacturing equipment necessary for Huawei to mass produce its own chip designs. US rules have largely cut Huawei off from the most powerful machines made by Dutch lithography company ASML, which essentially makes stencils to imprint miniature designs on chips for mass manufacturing, and TSMC, the world’s largest contract chipmaker. (The US Commerce Department is investigating how Huawei chips recently ended up on TSMC assembly lines.) Instead, Huawei relies on the Chinese chip manufacturer SMIC, which uses less powerful models of ASML machines.
But despite Huawei’s ambitions, Reuters reports that the company has been struggling with these restrictions to make effective chips at scale. For the Ascend 910C, the yield rate — the percentage that comes off manufacturing lines fully functional — is reportedly only 20%, while experts say a 70% yield rate is needed to be commercially viable. China’s top chip designer will need to make a breakthrough with limited resources to make good on its public promises to compete with Nvidia.