The Pentagon Is Learning to Think Like a Tech Company
From the NSA’s pursuit of multiple AI models to Ukraine’s battlefield data and the Pentagon’s fight over Chinese semiconductors, a new American national-security architecture is beginning to emerge.
By Ashley Greer | MetaFleurs News
WASHINGTON D.C.— Something interesting is happening inside the American national-security establishment.
For years, one of the easiest criticisms to make of the Pentagon and intelligence community was that they moved too slowly. Silicon Valley developed technology in months while government procurement could take years. Commercial companies accumulated enormous datasets while information inside government remained separated by classification levels, agencies and incompatible systems.
But increasingly, the U.S. military and intelligence community appear to recognize the problem.
And instead of choosing a single technology and betting everything on it, they are beginning to build something more ambitious: an architecture capable of continuously absorbing new technology.
Recent developments involving the National Security Agency, Ukraine’s battlefield artificial-intelligence systems and a Chinese semiconductor manufacturer offer three very different windows into that transformation.
The NSA Doesn't Want One AI. It Wants Access to All of Them.
At the Intelligence and National Security Summit outside Washington this week, NSA Deputy Director Tim Kosiba made a deceptively important statement.
“We want access to all the models,” Kosiba said.
The NSA is talking with leading commercial AI developers as the government works to bring increasingly sophisticated frontier models into national-security environments. A June national-security memorandum goes further, directing defense and intelligence agencies to develop “deep, proactive partnerships” with industry and obtain advanced technology from diverse suppliers.
The objective is not to determine whether Anthropic, OpenAI, Google or some future company possesses the one model the government should use. It is to prevent America's national-security apparatus from becoming technologically dependent on any one of them.
This is what technological competition increasingly requires.
An AI model that is dominant today may be surpassed six months from now. A system particularly good at cyber operations may not be the best at imagery analysis, logistics or intelligence synthesis.
The government's competitive advantage therefore may not come from possessing the single best AI.
It may come from possessing the infrastructure capable of rapidly incorporating whichever AI is best.
That turns interoperability from an obscure technical problem into a national-security capability.
Kosiba has framed the larger technological competition explicitly. Speaking earlier this month about China, he argued that although the United States cannot match China's population, America can maintain an advantage through innovation and integration.
Integration may be the more important word.
Ukraine Has Something Almost as Valuable as the Weapons
Another development this week illustrates why.
Britain and Ukraine announced a new defense-AI partnership giving the United Kingdom access to Ukraine's Avengers AI Labs battlefield database.
Thousands of daylight cameras and infrared sensors operating across Ukraine have captured millions of observations of tanks, artillery, air-defense systems, infantry, drones and other battlefield objects. That information is being used to train artificial-intelligence systems.
According to defense intelligence firm Janes, a Ukrainian automated target-detection system trained through Avengers Labs is already processing more than 100,000 UAV video streams each month. Britain is examining potential applications of the Ukrainian data for decision support and target identification.
That means Ukraine has accumulated something extraordinarily difficult to manufacture in a laboratory:
real-world training data from modern warfare.
Weapons can be reverse-engineered. Drones can be purchased. Software can be copied. Processors eventually become commercially available.
A vast, continuously updated and accurately labeled dataset showing what military equipment actually looks like under battlefield conditions is considerably harder to reproduce.
Ukraine therefore isn't merely a recipient of Western military technology. It is becoming a contributor to the technological infrastructure of Western defense.
The partnership demonstrates another important characteristic of the emerging security architecture. The system does not stop at America's borders.
The United States and its allies increasingly possess different pieces of the same technological puzzle: American computing and AI companies, European defense manufacturers, Ukrainian battlefield experience, allied intelligence networks and enormous amounts of sensor data.
The strategic challenge is connecting them.
The Pentagon's Power Now Extends Deep Into Technology Markets
A third development shows the harder side of this strategy.
On Friday, Chinese memory-chip manufacturer ChangXin Memory Technologies, or CXMT, sued the Pentagon over its designation as a “Chinese military company.”
CXMT says it manufactures DRAM chips for civilian and commercial customers and denies affiliation with China's military. The Pentagon's designation can produce government-contracting restrictions as well as broader commercial and reputational consequences. CXMT argues that the determination was unsupported and violated its due-process rights.
The courts will determine whether the Pentagon properly applied the law in CXMT's case.
But the dispute illustrates a larger reality that Washington should not pretend does not exist.
Semiconductors are national-security infrastructure.
Memory chips sit inside servers, AI systems, computers and countless other devices. Telling the differnce between a civilian technology supply chain and a military technology supply chain has become increasingly difficult to maintain when the same underlying computing infrastructure powers both.
That creates uncomfortable decisions for the Pentagon.
It also makes those decisions unavoidable.
If American military superiority increasingly depends upon computing, data and artificial intelligence, then understanding who manufactures the components beneath those systems—and where strategic dependencies exist—is part of national defense.
Due process and judicial review remain important precisely because these designations are powerful. Scrutiny of the Pentagon's decisions should strengthen the system rather than prevent the government from making national-security judgments in the first place.
The Anthropic Fight Is Part of the Same Transition
That makes the Pentagon's recent confrontation with Anthropic particularly revealing.
The dispute has often been presented as a battle between government power and an AI company attempting to maintain ethical limits on its technology.
There is another way to understand it.
The military increasingly depends upon technology it did not invent, running on infrastructure it does not completely own and developed by corporations whose commercial interests do not always coincide with national-security requirements.
That is an inherently unstable arrangement.
The government's response should not be to abandon commercial AI. Commercial innovation is moving far too quickly, and the NSA's pursuit of multiple frontier models demonstrates that officials understand this.
Instead, the government has to develop a system in which no individual company becomes an irreplaceable point of failure.
The June national-security memorandum makes that objective unusually explicit: AI systems relied upon by American warfighters should not be vulnerable to being disabled, degraded or materially altered by a commercial entity without government knowledge and approval. It simultaneously calls for multiple vendors, interoperability and rapid adoption of advanced commercial AI.
The New Arsenal Is a Network
For most of American history, military power could be photographed relatively easily.
Ships. Aircraft. Missiles. Tanks. Factories.
But increasingly, the decisive capability is hidden underneath them.
It is the network connecting sensors to data, data to artificial intelligence, artificial intelligence to analysts, analysts to commanders and commanders to weapons and other instruments of national power.
That may explain why interoperability was one of the most persistent themes at this week's Intelligence and National Security Summit.
The United States does not necessarily need to invent every component of the future military itself.
It needs something potentially more difficult: the ability to identify the best technology being produced by American industry and its allies, test it, secure it, connect it to existing systems and put it into the hands of American personnel faster than an adversary can do the same.
The NSA's desire for access to multiple AI models, Britain's access to Ukrainian battlefield data and the Pentagon's scrutiny of Chinese semiconductor companies can look like unrelated developments.
They aren't.
They are pieces of a new defense architecture in which integration itself becomes a weapon.
And if the United States can build that architecture successfully, the sprawling combination of American military institutions, intelligence agencies, technology companies and allies that once looked cumbersome may turn out to possess an enormous advantage.
Not because America has one perfect system.
Because it can connect thousands of imperfect ones.