Why the Pentagon is turning to AI startups

The Last Supper

In the autumn of 1993, senior executives from America's big defence companies gathered for a highly confidential dinner at the Pentagon hosted by then-Secretary of Defence Les Aspin. Later dubbed 'the Last Supper,' it set in motion a wave of mergers and acquisitions, fundamentally reshaping the US military-industrial sector.

The Cold War had ended, and the US needed to cut its defence budget significantly, given the much-vaunted "peace dividend," redirecting money towards infrastructure projects and healthcare. From 1991-96, US defence spending fell by more than 15%, which put a strain on the defence firms, who relied on government contracts.

Seated next to Aspin at the dinner was Norman Augustine, boss of Martin Marietta, who began by asking his host why they were meeting. "You'll know in 15 minutes," said Aspin. "And you probably won't like what you hear." Aspin and his deputy William Perry then led a sobering discussion. Martin Marietta would later merge with Lockheed to form Lockheed Martin, a firm worth around $140bn at the end of 2024.

Perry's data was stark and pointed to a drastic reduction in the number of defence firms that the US government could sustain due to budget cuts. He explained that in key sectors—such as fighter jets—only three companies could be supported, while in others—such as tanks—only one would survive. 

It became clear that the government would not step in to save struggling firms. Rather, the market would determine the winners. Augustine later summarised the message as: "You have to merge your companies, or you're out of the market."

Alongside Martin Marietta's merger with Lockheed, Boeing bought McDonnell Douglas (creating a company today worth $135bn), while Northrop bought Grumman to create Northrop Grumman (valued at almost $80bn at the end of 2024). Numerous smaller defence firms were absorbed into larger entities. By the end of the decade, the number of big defence contractors had gone from 15 to five.

The US government had subtly encouraged consolidation, with the Department of Justice nodding them through on national security, but the shift came at a cost. Bureaucratic compliance was prioritised over innovation, risk-taking, and technological breakthroughs. Arguments between engineers and bean-counters were increasingly won by the latter.

With fixed profit margins, there were fewer incentives to seek cost-effective solutions or technological advances. The industry became less dynamic and adaptable. Today, the US government is still the largest client for AI startups, particularly those with military applications. Alongside traditional defence giants, these emerging companies compete for lucrative Pentagon contracts.

The five biggest contractors—Lockheed Martin, Raytheon, Northrop Grumman, General Dynamics, and Boeing—still dominate an industry built around long-term financial stability. Unlike the volatile commercial sector, military contracts provide funding over several years, which keeps the firms solvent. With the increasing importance of AI, will the tech firms upend the industry much like the Last Supper?

Follow the money

In April 2023, a Stanford University study revealed a sharp increase in US federal spending on AI throughout 2022. By June 2023, the House Appropriations Committee was prioritising legislation to integrate AI into an expanding array of military and government programmes. By November 2023, both the Department of Defence (DoD) and the State Department shifted from experimentation to implementation.

US federal spending on AI went from $261mn in 2022 to $675mn in 2023, while the potential value of AI-related contracts surged to more than $4.5bn. The number of DoD contracts related to AI jumped from 254 to 657. Worth around $4.3bn, they represented 95% of all federal AI spending. By contrast, agencies such as NASA and the Department of Health spent just a fraction of that on AI.

This surge in funding has reshaped the defence AI market. The number of companies securing contracts worth $10mn or more went from four to 205. The market remains highly fragmented, however. Most of these AI technology firms must operate within narrow contractual scopes tied to a single funding agency or department.

The Pentagon has launched a series of ambitious initiatives, including the Multiple Award Contract for Artificial Intelligence, projected to be the largest AI-related government contract worth $15bn over the next decade. It aims to develop an advanced data analytics platform capable of processing vast amounts of military and intelligence data with unparalleled speed and precision.

Another major project is Multiplier, focusing on mass-producing low-cost autonomous drones powered by AI. Designed to enhance operational efficiency, the initiative seeks to streamline command and guidance operations while fortifying drone networks against electromagnetic attacks.

Swift yet hindered

Given the scale and complexity of these initiatives, together with the breakneck pace of AI innovation, the traditional US defence contracting system is proving inadequate and bureaucratic, so the DoD is increasingly turning to AI startups that offer cutting-edge solutions, bypassing the slow-moving mechanisms of legacy contractors.

With their ability to deliver advanced technological solutions quickly and easily, AI startups are gradually securing a larger share of government contracts as the US seeks to maintain dominance in the autonomous weapons market. 

According to a Defence Advanced Research Projects Agency (DARPA) report, 70% of the agency's research now revolves around AI and machine learning. Matt Turk, DARPA's deputy director, said startups and the private sector were the primary drivers of AI innovation in defence applications.

Yet despite the tech firms' agility, federal acquisition regulations—shaped by legislative mandates and executive orders—still hinder the rapid adoption of AI. Designed to support national priorities (such as promoting US-made products), they nevertheless slow the process, which puts a strain on private-sector engagement and delays the military's ability to integrate cutting-edge AI technology.

As a result, total spending on startups remains relatively small. From a $411bn defence budget in the last fiscal year, startups accounted for just 1% of contracts. The US military is now exploring new contracting models to foster a more competitive and innovation-friendly AI ecosystem. 

One such initiative is the Interoperable Government Data Warehouses and Applications Challenge, launched in July 2023. It seeks to streamline AI procurement for the US military while expanding AI deployment across different branches of the armed forces.

Logistical challenges 

With its ability to learn, adapt, and perform complex tasks almost instantly, AI could revolutionise military operations, but developing and deploying these systems is far from easy. AI models require extensive training on vast datasets to operate with precision. Fine-tuning and optimisation is then achieved through multiple test phases before they can be integrated into real-world military environments.

To maximise AI's effectiveness, military forces must establish large-scale computing infrastructure within operational zones. This allows real-time analysis of data collected from ground sensors, drones, and satellites. The closer the data processing centres are to the battlefield, the more responsive and effective military AI systems become. 

Reduced reliance on cloud-based networks (which are vulnerable to electronic warfare and cyberattacks) is crucial for ensuring uninterrupted and secure operations, yet balancing the need for high-speed, large-scale data processing with network security and confidentiality remains a formidable challenge.

As AI integration in military operations expands, so do cybersecurity risks. One of the primary challenges, according to the National Security Technology Accelerator (NSTXL), is securing communication between peripheral AI-powered systems and central command networks to prevent cyber intrusions and system breaches.

This risk is compounded by the US military's reliance on adaptive AI, which necessitates regular updates to counter evolving threats. Cybersecurity experts warn that AI data theft and the decryption of classified military communications pose critical security vulnerabilities, requiring the urgent development of advanced encryption protocols to safeguard sensitive information.

Critical vulnerabilities

A recent Pentagon report revealed that while defence chiefs expressed confidence in the security of their systems, rigorous testing exposed critical vulnerabilities. Lawmakers expressed dismay over the extent to which modern weapons systems could be compromised by cyberattacks, especially for a digitally-reliant military.

Yet as weapon systems grow in complexity and sophistication, so too do their exploitable weaknesses, illustrated by Russia's cyberattacks against Ukraine, crippling the Ukrainian army's command and control networks and targeting critical infrastructure like military communications and energy grids. 

AI has the power to improve the accuracy of missile systems and support real-time decision-making, but its dependence on data integrity introduces a significant vulnerability. For instance, if AI-guided missiles or drones are fed manipulated or compromised data, weapons could be turned on their wielder or on civilians.

Once AI technology is deployed on the battlefield, soldiers need to learn how to use it without becoming dependent on it. They also need to trust its security. Exploited or breached, AI-led systems could be used to disrupt operations or steal classified intelligence.

Ethically, AI in warfare raises some dilemmas. While it had the potential to reduce civilian casualties by improving target accuracy, it also introduces lethal autonomous weapons systems (LAWS), whereby machines—not humans—determine the use of deadly force. Many argue that these systems need better oversight.

As AI continues to redefine the battlefield, it will continue to determine a combatant's fortunes in war, and states will continue to scramble their technological resources to stay one step ahead, companies will continue to scramble for contracts, and politicians will continue to scramble to create the laws and regulation needed to shape the future.