Analysis

Chinese political leaders are keeping their eye firmly on the new Trump Administration and how China can respond to the US’s new policy directions. In particular, Trump’s attention on territorial grabs in Panama and Greenland, for example, spiked the attention of both local residents and the Chinese government, which is keen to find a bargaining chip for preventing the US from interfering in case of an invasion of Taiwan. 

Offering to accept US takeovers of Greenland in exchange for China’s takeover of Taiwan is precisely the kind of transactional approach that the CCP will likely take with Trump, and which make Taiwanese officials nervous that they may be traded away in a game of superpower dealmaking. Beijing will likely also try to include a Russia-Ukraine land swap agreement as part of a broader bargaining, meaning that upcoming high-level US-China negotiations are likely to be dominated by discussions of Taiwan and territory, with China keen on amending the US’s current (and purposefully) ambiguous stance towards Taiwan. 

The Trump Administration’s slew of new executive orders and proposed policy changes may have China in its sights, but the messy rollout of some policies will likely give the CCP a potential angle to exploit. As the fight with Denmark over Greenland highlights, the CCP will be most happy if the US quarrels with its allies, especially if this leads to the US’s withdrawal of NATO or other multilateral arrangements–this is a longstanding desire of both China and its closest ally, Russia. In the context of new 10% tariffs against Chinese (along with Canadian and Mexican) imports to the US, China will therefore be looking for alternative forms of leverage in future negotiations with the Trump Administration. 

One clear response from the first few weeks of the new administration is that the EU is shifting its stance to focus more on what collective security without the US might look like. For businesses, one important fallout from this is that the EU is also looking to wean itself off any reliance on the US economy. As the Trump Administration continues to push corporate America to align with Republican party values, EU governments and companies may therefore look to insulate themselves by separating their US-based entities from their European counterparts, much as many companies already do with their China-based operations. Multinational companies should therefore watch conversations about economic nationalism in the US, EU, and China closely, as they may portend future corporate fragmentation that will be prudent to protect against unwanted political influence.

Analysis

In response to DeepSeek’s shock announcement of its new models, which were purportedly trained on Nvidia chips that were compatible with Biden-era export controls (see “Tech Futures” below), John Moolenaar (R-MI) called for stricter export controls to include AI models, chip components, and other tech measures necessary for AI innovation. Other Republicans called for an end to US-China collaboration on AI, although this seems to miss the point that many AI developers in the US are themselves Chinese. The challenge for this approach is that the horse is already out of the barn: either companies like DeepSeek are already using slower chips and still managing to out-perform US rivals, or they are quietly using gray market,  “banned” chips and able to circumvent export controls in the first place. 

Trump’s promises of sweeping tariffs appear to respond to multiple policy ambitions, from Colombia’s refusal to accept returned migrants from the US with threats of 25% tariffs on Colombian imports, to threatening Taiwan with tariffs on chip imports. As US chip manufacturers gradually shift to producing chips on US soil, demand for chips looks likely to increase beyond the US’s current domestic capabilities, which means that the threat of tariffs will likely dominate conversations about AI capabilities and the US’s competitive advantage. Importantly, as most semiconductor assembly occurs overseas, tariffs that target chips may push import-reliant manufacturers to relocate abroad to avoid tariffs and increase prices for US consumers. This is an opposite effect of the tariff’s intentions. Tariffs, chips, and AI are therefore tightly intertwined, and businesses should pay attention to all three to keep ahead of trends.   

DeepSeek’s success is a predictable consequence of the Biden Administration’s export controls on semiconductors and AI chips to China. Whereas China previously depended on US technology, the export controls forced China to learn how to innovate-–or evade the controls–and it appears to have done so with astounding speed and quality. This does not mean that export controls should be stopped altogether, as Anthropic’s CEO convincingly argued. Rather, export controls are still effective in slowing down China’s access to the necessary chips for AI innovation. That said, export controls should be considered as one tool in an arsenal of economic, political, and technological levers against China and other competitors. The challenge for the current administration is that this arsenal requires a multi-pronged and long-term strategy with fewer opportunities for flashy short-term wins; that may not be a winning strategy for a Trump Administration that will need to quickly demonstrate its tough-on-China approach.

Analysis

With the launch of its R1 model on 10 January 2025, the Chinese-developed, generative AI software DeepSeek (深度求索) has forced the AI industry to rethink its business fundamentals. The Chinese software is claimed to have been developed using approximately 2000 of Nvidia’s older, H800 chips and on a budget of just under $6M, with performance metrics that rival US-based, industry-leading models. Compared to Silicon Valley giants’ AI softwares–like OpenAI’s ChatGPT, Meta’s LLama, and Google’s Gemini–that operate on Nvidia’s most advance, H100 chips and required hundreds of billions of dollars to develop, DeepSeek’s cost-efficient model has suddenly made advanced AI technology possible for businesses with less capital to spend. On the news that DeepSeek was developed using the less powerful Nvidia H800 chips, the company’s stock fell almost $600B in a single day, although they have recovered nearly half of that value in recent days. 

While there is still great uncertainty regarding how DeepSeek will affect the future development of AI software, it is clear that it offers a way forward separate from the leading US softwares. In an interview with DeepSeek founder Liang Wenfeng, he makes clear that his goal is to make DeepSeek the industry standard, keeping the company focused on research and innovation rather than application development. In this way, he hopes other companies will develop applications using DeepSeek, in turn making the software the industry standard. Mr. Liang has also become an overnight hero in China for his successful disruption of the US-dominated industry. Such nationalistic pride is also reflected in Chinese news outlets, which have suggested that DeepSeek has been subject to multiple waves of US-led, IP theft attempts, despite DeepSeek’s model being entirely open source.

DeepSeek still retains certain “Chinese characteristics,” insomuch as some topics remain off-limits. For instance, questions about the 1989 Tiananmen Square Massacre or well-documented, state-sponsored detention and forced labor of China’s Uyghur population will return no results (although similar questions, for example, about the 1970 Kent State shootings in the US, are dutifully answered). Some experts worry that, if new programs are built using DeepSeek’s open-source code, that the model’s inherent censorship will become baked into future AI models. Despite this, the fact that the company is not developing applications themselves exempts them from the greater portion of Chinese censorship laws and allows them to build models that others will then have to adapt for their specific purposes. Regardless, US AI companies will now have to compete with cheaper Chinese models and develop new tactics for marketing their more expensive but largely equivalent products.

Analysis

DeepSeek’s R1 model caused a firestorm in Silicon Valley and in global markets this past week. Most of the hype centered around DeepSeek’s model being produced much cheaper than its competitors, with slower chips that comply with the US’s export controls, and because the model came from Chinese developers whose progress was supposedly behind their American counterparts. While the first two points remain debatable, DeepSeek has certainly shifted global conversations in favor of China’s capacity for technological innovation. The US is no longer the clear front runner in AI innovation, at least as far as media reports are concerned. 

DeepSeek’s announcement triggered a series of dramatic shifts in the US. OpenAI announced a partnership with US National Laboratories and Trump called in Nvidia’s CEO for questioning about how its chips had been used in DeepSeek’s models. Politicians and investors raised questions about OpenAI’s open source approach, particularly after the company called for more investment in physical infrastructure to support AI development amid claims that DeepSeek accelerated innovation because of its access to open source models. These criticisms similarly came under fire by proponents of DeepSeek, who point out that the company also produced open source models. One interesting development is that several US companies, including Microsoft, have made DeepSeek’s R1 model available on their platforms as a nod to the company’s use of open source. 

DeepSeek, however, is neither typical of Chinese innovation, nor is it a state-backed tech giant like Tencent or Alibaba. Indeed, DeepSeek’s announcement will likely ruffle feathers at Chinese enterprises just as much as it has in US companies. There are approximately ten other AI labs in China, for example, producing similar models with varying capabilities and advantages, not in the least Alibaba’s Qwen 2.5, which was announced immediately after DeepSeek’s R1. Each of them, including DeepSeek, complies with Chinese government regulations about towing the party line, which adds an additional risk to the widespread use of its models (as we detail below in “Espionage Alert”).  

In short, DeepSeek’s R1 model made a splash, but it is not a signal that China has already won the AI race, nor is it a sign that China’s technological innovation environment is superior. Rather, it demonstrates that if the US wants to continue its leadership position in AI innovation, it needs to continue innovating, attracting global talent, and directing large-scale investment towards AI and related fields.

Analysis

Concerns of espionage, data exposure, and national security plagued TikTok’s recent ban in the US. Yet this past week, four Chinese apps appeared among Apple’s “most downloaded” in its app store: DeepSeek (深度求索), Temu, Shein (希音), the Bytedance-owned tutoring app Gauthlite (精精 唐), with social media site Red Note (小红书) just dropping out of the list of most-downloaded apps after a surge in interest with TikTok’s US ban. With DeepSeek’s surge to the most-downloaded app for iPhones, in particular, questions about the pervasiveness of Chinese apps in the US market are starting to raise security flags for both US users and for US national security. If TikTok was banned on suspicion of espionage, why are other Chinese apps permitted to operate in the US without similar scrutiny? 

There are several options for Congress, the courts, and the Trump Administration. First, a blanket ban on Chinese apps–that is, apps that send user data to servers in China–could help to justify bans on the grounds of ensuring that US data remains in the US. Second, tighter restrictions on the export of US data by any apps–Chinese or otherwise–can shield the US government from accusations of targeting China specifically, which some have claimed as unfair targeting or a new tech “red scare.” The US should (and does) target China’s aggressive and adversarial means to try to undermine US technology, infrastructure, and data, not in the least because China’s complaints about app bans are rich considering US apps are banned in China. But it is smart to do so from a comprehensive position that understands the collective risk to US data from foreign apps and data harvesting technologies, which China will likely continue to exploit even if individual Chinese-made apps are banned.

While China is a key proponent of data harvesting (and with it, IP theft and other malicious practices), the real issue is that US users are far behind where they should be in terms of being smart about where their data goes, who has access to it, and how malign actors can exploit that data for financial or intelligence advantages. LLMs and other AGI models, like DeepSeek, therefore present a potentially far more widespread and invasive data-mining risk to US users–and by extension national security–than TikTok. As the EU has noted, data security is crucially important to users and governments, with LLMs presenting a new frontier in how this data is collected and leveraged for corporate gain.

Banning one platform does not make the US secure. But as we note below about TikTok, the forced sales of Chinese apps is a trend that is likely to expand to other apps and platforms and will have business implications for those engaging with these apps. But it also points to a growing sentiment of corporate or tech nationalism that equates companies and products with the government of their country of origin. While this is not necessarily a bad thing, particularly in the sense that it foregrounds much-needed conversations about Americans’ data security, it does suggest a reshaping of global corporate practices in the years ahead. Businesses watching the tech space, or encouraging experimentation with new apps for work efficiencies, should therefore pay close attention to emerging government responses and arising security implications.

Analysis

Oracle and Microsoft have emerged as the frontrunners of the TikTok “bidding war,” with President Trump hoping to wrap up negotiations within 30 days. They are currently in advanced talks with the White House and other partners to work out a solution for owning TikTok’s global operations. Under the current terms, which are subject to change, ByteDance will cede control over the app’s underlying algorithms, data collection, and software updates to Oracle, which already serves as the backbone for TikTok's web-based infrastructure. Trump denied rumors circulated suggesting he and Larry Ellison, the chairman of Oracle, were in contact regarding the sale. While Microsoft is participating in the discussions, its specific role remains somewhat ambiguous. The collaboration may be a strategic move to combine Oracle’s expertise in data management with Microsoft’s technological capabilities.

In addition to Oracle and Microsoft, several other parties have placed bids, with varying levels of follow-up. Kevin O’Leary and Frank McCourt have placed “The People’s Bid,” and popular YouTuber and influencer MrBeast (a.k.a. Jimmy Donaldson) has also placed a bid, both reportedly around $20 billion. Perplexity AI has also officially submitted a bid with a plan to merge with TikTok’s US operations. While Elon Musk, Steve Mnuchin (the former US Treasury Secretary), and others have floated the idea of organizing a purchase, they have not officially placed bids. 

Looking ahead, the sale of TikTok highlights increased scrutiny and regulatory challenges that other Chinese companies will likely face in the US. This case will set a legal precedent, potentially leading to similar divestment demands. Chinese companies will need to navigate complex legal frameworks and may reconsider their investment strategies in the US market, ironically something that American companies have increasingly struggled with in the Chinese market.