The Year of the Technology Superpower.
This year, evolutions in software development, export controls, and geopolitical competition will test the definition of an artificial intelligence “superpower” in new ways.
As a parting initiative, the Biden administration is mulling more rules to curb China’s access to advanced technologies. This time, by regulating the flow of semiconductors and frontier models to other countries, including some US allies. It would be a huge geopolitical flex by Washington; but could it backfire?
Though details of the proposed “Export Control Framework for Artificial Intelligence Diffusion” aren’t public yet, they have started leaking out.
In a series of scathing articles published on Oracle’s website, the US software giant’s executive vice president, Ken Glueck, outlined the key elements of the proposed policy, which he argued would undermine America’s leadership in artificial intelligence.
So what? According to Glueck, the US government wants to impose a global licensing regime on access to cutting-edge semiconductors and AI models. An elite club of twenty countries would get privileged access, while other countries—even including some NATO allies—would be subject to chip quotas.
US cloud providers, meanwhile, would be able to get permission to export chips to their own facilities in third countries that don’t fall into the elite twenty club through a new “validated end user” program—provided they follow strict security protocols.
Glueck said the new export control framework would also attempt to control access to LLMs themselves, with an exception for open-source systems, although it’s unclear how this might work.
What’s going on? A conflict over how best to contain the potential national security risks from adversaries with access to frontier artificial intelligence systems.
Since 2022, the Biden administration has pursued what it calls a “small yard, high fence” strategy, attempting to limit China’s military advancement by putting tough restrictions on a small number of sensitive technologies.
Big yard, rickety fence. In reality, as former US National Security Council official Geoffrey Gertz noted in the New York Times last week, the yard has been steadily widening. The fence also arguably isn’t as high as advertised, as both China and private sector companies have found workarounds to previous rounds of restrictions.
News reports back in October, that TSMC-manufactured chips had ended up in shipments of Huawei’s “multi-chiplet” 910B AI processor, may have reinforced the sense of urgency in Washington over shutting down fence-jumping that allows China to gain access to advanced compute via third countries.
Inference: An easily weaponizable scheme, for controlling allies and adversaries alike, feels tailor made for a Trump administration with an appetite for brinkmanship, seeking leverage in trade deals, and technological ascendancy.
A superpower play? Yes, the US is the uniquely-placed hegemon, solely capable of pulling off such a move. Security hawks worry that if China or other “adversary” countries gain access to the most advanced artificial intelligence models, it could enable new weapons systems, military and cyber capabilities that would harm America’s national security interests, and present a credible challenge to its position.
For some hawks, the fact that manufacturing of advanced chips is heavily reliant on US technology—and thus subject to US export controls—is a powerful lever that should be pulled early and often as Washington wages its tech competition with China.
Or a flawed one? Skeptics worry that the US’ technology superpower status is more precarious than it may seem at first glance. Not only are export controls hard to enforce, but they can end up backfiring by creating strong financial and political incentives for companies—and countries—to find workarounds or even try to ditch US technology altogether.
China is already attempting to “design out” US chips and other technologies to reduce the risk of key industries getting further caught by the export control dragnet. European and other G7 allies concerned about “digital sovereignty” are also likely to take a dim view of US attempts to impose a global licensing regime on advanced compute and AI model weights.
Even for the twenty countries that will maintain relatively unfettered access to advanced chips, the idea that the US could presumably one day revoke that privileged status will be concerning.
Biden’s recent decision to block the acquisition of US Steel by a Japanese competitor, and the previous Trump administration’s use of national security justifications as political leverage in trade deals will only reinforce skepticism about relying on US cloud providers.
Third countries are unlikely to relish having to rely on US tech giants to access cutting-edge AI chips. Skeptics, including Glueck and former Trump national security advisor Robert O’Brien, argue the attempt to extend global controls on chips (and potentially even model weights) could give these other countries a strong incentive to ditch US technology altogether in favor of less capable—but still good enough—alternatives.
That would be a gift to China. As the Carnegie Endowment’s Evan Feigenbaum notes in a must-read piece on the US’s strategic position in Asia for Forbes this week, American “technonationalism” could complicate Washington’s efforts to bring India into the US technology fold, for example.
In sum, a new rule that extends US chip controls (and possibly controls on model weights of “frontier” systems) to third countries, creating a tiered system of privilege in which only the US’ closest allies get unfettered access to the best chips would be;
one final flexing of the Biden administration’s geopolitical muscles;
a hard act for the Trump administration to follow in terms of presenting a tougher on China posture; and
punishing for nations that are outside of China’s orbit too.
How are companies responding? They are hedging. US chip makers have attempted to insulate themselves from previous rounds of controls by creating alternate versions of their GPUs with reduced processing speeds, which they can still sell into the Chinese market.
Meanwhile, Chinese manufacturers like Huawei have also rushed to fill some of the vacuum left in the market for the most cutting edge chips in China, by selling chips like the Ascend 910C, which approximate the performance of Nvidia’s cutting-edge H100s, for example.
A choke-point, but for how long? The CEO of DeepSeek, one of China’s leading foundational model companies, has acknowledged that access to advanced chips is the main delimiting factor for progress in the domestic market, not the availability of capital.
A new rule that limits China’s ability to access cutting-edge compute via countries in the Middle East, South Asia, and other nations would widen this gulf significantly in the short run, but could undermine US technology dominance in the long run by making everyone think twice about relying on US components.
Moving at the speed of hype. In 2024, Inferences covered the explosive growth in the technology sector, driven by a period of “artificial exuberance”, as we called it. We reported on the changing geopolitical dynamics created by new breakthroughs, Nobel Prizes, components, controls and commercial deals. The big question remains: can the confidence and hype hold, until the value proposition is real?
This will be the year in which experimental budgets run out for many organisations, and boards demand results. The geopolitical context for this enterprise race can only become more important.
The upshot? The forthcoming US controls signal that diplomacy by limitation of technological resources will be a mainstay in the year to come, and raise many questions about how the US will leverage its status as the global technology superpower.
The details are still pending, but with the Trump administration set to take the reins, the appeal of using a brand new vector of control to influence political relations with allies and adversaries alike is undeniable.
As we look to the year ahead, we’d like to say thanks for joining us as a reader, and to the many of you who’ve reached out to engage us in discussion of the topics that really matter.
What we’re reading:
Evan A. Feigenbaum on the “whirlwind” of strategic opportunities for Asia presented by the new Trump administration.
New plans from the UK Government to make the most of Horizon Europe—the world’s largest R&D collaboration programme.
TBI puts proper emphasis on the twin challenges of the software and energy transitions in the Global South.
What we’re looking ahead to:
6 - 7 February: The Inaugural Conference of the International Association for Safe and Ethical AI, Paris, France.
10 - 11 February 2025: AI Action Summit in Paris, France.
11 - 13 February 2025: World Governments Summit 2025, Dubai, United Arab Emirates.
12 February 2025: Chief AI Officer Summit UK, London.
April 2025 (expected): G7 Digital Ministerial, Canada.
2 - 4 June 2025: AI+ Expo and Ash Carter Exchange in Washington, DC.
9 - 11 July 2025: AI for Good Global Summit.