Because we operate at the intersection of AI, national security, and cyber, I often get asked the question, "Are we winning the AI race with China?"
asserted
we → operate → China
At the highest level, we have the best chip technology, the best models and, more importantly, we have the best economic system and the talent to continue to win.
asserted
we → have → system
But at a deeper level, we still have to agree on what winning looks like.
asserted
winning → have → what
Fundamentally, is it building the best technology, or is it building the technology that becomes ubiquitous?
asserted
that → build → technology
The tech world is littered with examples of superior technologies that lost to inferior ones that took over the market and became the standard.
asserted
that → litter → market
Decades ago, when the stakes were much lower, VHS beat Betamax.
asserted
VHS → beat → Betamax
More recently, and much more relevant to our national security, Huawei beat its Western Hemisphere competitors to become the global leader in telecommunications gear.
asserted
Huawei → beat → gear
This handed China a major opportunity to gather information, intelligence, and leverage around the world.
asserted
This → hand → world
When the dust settles and the picture becomes clearer, what will matter most is whose technology has become the global standard.
asserted
technology → settle → ?
Which AI stacks are people around the world using to become informed, to automate, to increase productivity, to analyze, and decide?
asserted
people → use → productivity
Within the global AI adoption race, the contest is complex and better described as a triathlon, with all three legs being run simultaneously.
asserted
legs → describe → triathlon
Experts estimate we have at least a two-year lead on chips thanks to our fundamental advantage in lithography and incredible advances by Nvidia, its Taiwanese partner TSMC, and many others.
asserted
we → estimate → Nvidia
And on the models themselves, while the gap is narrowing, our frontier labs are leading by perhaps two to three generations — so eight to 12 months.
asserted
labs → narrow → generations
That may feel short, but it is a lifetime in frontier AI.
uncertain
it → feel → AI
Take the example of AI in cybersecurity, which has rightly been much in the news.
asserted
which → take → news
Booz Allen’s Cyber Weapon Index codifies how good these models are at conducting a cyberattack.
asserted
models → codify → cyberattack
Two models, Mythos from Anthropic and Astra from OpenAI, outscored all others by a wide margin.
asserted
models → outscore → margin
This is good news because both companies are speaking publicly about the need to behave responsibly and in partnership with the U.S. government in releasing these technologies to the world.
asserted
companies → speak → world
But several Chinese models have shown initial capability and are growing in expertise — getting better in every generation — likely with fewer guardrails than their American counterparts.
asserted
models → show → counterparts
The second leg of the AI adoption triathlon is based on cost.
asserted
leg → base → cost
Roughly speaking, the difference between the best American models and the best Chinese models is 5-10x the cost per token.
asserted
difference → speak → token
And while the Chinese models cannot fully replicate the capabilities of our frontier labs, they are often good enough for many tasks.
asserted
they → replicate → tasks
As a result, they are being used widely, especially by cost-sensitive customers.
asserted
they → use → customers
Think large global companies trying to manage IT budgets, cash-strapped startups in Silicon Valley, and developing-country governments with limited resources.
According to OpenRouter, a model marketplace where users can access a range of models, roughly 50% of tokens used in the last year were consumed on Chinese models.
uncertain
% → think → models
And we know of many U.S. startups that are using Chinese models — sometimes not realizing or disclosing their actual provenance — as code assistants or as the substrate for their applications.
asserted
that → know → applications
This matters because Booz Allen’s research demonstrates that Chinese models introduce more vulnerabilities when they code for an American application than when they do for a Chinese one.
asserted
they → matter → one
Over time, the accumulation of these tiny vulnerabilities could undermine the entire software supply chain on which the U.S. economic future rests.
uncertain
future → undermine → which
Companies, governments, and individuals are less likely to use a technology they can’t control or believe is operating against their interests.
asserted
they → use → interests
Here America — based on our values, way of life, free-market system, and history — has a right to win.
asserted
America → base → right
When American ingenuity created the internet, for example, the world eagerly adopted it, in no small part because the decentralized, transparent system of rules made it easy to use, to understand, and ultimately to trust.
asserted
it → create → rules
By contrast, the internet as it exists behind China’s Great Firewall, with its more rigid state controls and pervasive surveillance, would not be a good fit for most democracies around the world.
asserted
it → exist → world
Despite the underlying American advantage, the trust race is much closer than it should be.
asserted
it → underlie → advantage
Both countries and their respective AI ecosystems are eroding necessary trust unnecessarily.
asserted
countries → erode → trust
China, for example, trains its models to refuse to answer questions that go against Communist Party dogma, and to even refuse to perform certain tasks that the model perceives as being contrary to CCP interests.
asserted
model → train → interests
In America, for many reasons including disinformation and the lack of a clear framework, polls suggest our fellow citizens are turning negative on broad issues from the construction of data centers to the speed of AI advances.
asserted
citizens → include → advances
America must win this adoption triathlon by simultaneously pushing on all three fronts.
asserted
America → win → fronts
We must continue our leadership on the actual technology stack, invest in lower-cost alternatives to the frontier models, and rebuild trust.
asserted
We → continue → trust
The President’s America’s AI Action Plan provides a roadmap.
asserted
Plan → provide → roadmap
A few ideas to strengthen it would include:
- Frame the AI stack as a new element of our country’s critical infrastructure.
asserted
ideas → strengthen → infrastructure
Leverage lessons from other areas, from banking to the defense industrial base to our energy industry, where a combination of voluntary and mandatory rules helps simultaneously protect and strengthen these industries.
asserted
combination → leverage → industries
…and 7 more, not listed.
The Aporia
