From its first introduction in 1980, the M1 Abrams tank was a symbol of American innovation and capability. It had clear competition from Soviet tanks, it had a clear role in American military doctrine, and it had a compelling pipeline of innovations to help it do its job better than ever. The tank integrated several new technologies: multi fuel turbine engines, composite armor, computerized fire control. Additionally, the tank’s cabin protected against nuclear fallout, biological contaminants, and common chemical warfare agents. When first deployed in 1991 during the Gulf War, 23 of the 1,956 deployed tanks—roughly 1%—were lost, all due to either friendly fire or intentionally destroyed to prevent enemy capture after getting stuck. The M1 Abrams is a classic example of what the defense tech industry considers an “exquisite system.” Capable of outclassing any other tank in the world, from its release through the modern era, the M1 Abrams was a symbol of American technical prowess, earning the title of best main battle tank in the world in Operation Desert Storm. It represents the pinnacle of Department of Defense (DoD) technocracy, which in theory works through meticulous planning about the future of war, considering the one best way to solve particular problems, testing several ideas for technical readiness and cost effectiveness, and then driving defense contractors to deliver that capability for the lowest possible price. 

For at least a decade after its release, the M1 Abrams was considered a premier land battle vehicle. Fast forward to Ukraine in 2025. Over 20 of the 31 M1 Abrams tanks sent to Ukraine are believed to be destroyed, disabled, or captured. Most main battle tanks fielded have proven extremely susceptible to cheap drones piloted remotely that can exploit weaknesses in their armor. As a result, the Ukrainians have deemphasized the role of M1 Abrams tanks on the front line. They are too slow and difficult to protect against cheap, omnipresent drone strikes. The M1 Abrams solved a critical problem at a critical time, but the economics and scale of drone warfare eclipsed its role on the battlefield in a way that nobody had predicted.

Bradley Fighting Vehicle

Next, consider the Bradley Fighting Vehicle (BFV), the heavily publicized example of defense procurement gone awry. Its lackluster reputation was a result of a tortured development process that became the subject of a book and HBO movie, The Pentagon Wars, satirizing the Department of Defense’s politics, perverse incentives, and alleged fraud of the development and acquisitions process. The Bradley was built by a committee (or rather committees), who continually changed their mind about what the tank should do over literal decades. After 22 years in development, the Bradley was finally released to little fanfare and much skepticism. 

When introduced in 1980, the Bradleys role was to move infantry and provide fire support, known as forceful, rapid reconnaissance. (Note that while it may look like a tank to a civilian, servicemembers debate if it merits that designation. My favorite description of it is a battle taxi.) This role was important in US military doctrine, but the Bradley itself was not a technological marvel. It was a unique blend of mobility, armor, and firepower that came with severe tradeoffs: the jack of all trades, but master of none. The Bradley would prove surprisingly effective in the Persian Gulf War. Bradleys destroyed more Iraqi armored vehicles than M1 Abrams and reportedly earned a few tank kills at close range. Bradley losses were comparable to the M1 Abrams, losing 20 of the 2,2200 for a comparable 1% loss rate. Similarly, 17 of those losses were due to friendly fire incidents. It took them 22 years of changing requirements and iterating, but apparently they settled on something effective.

What was even more unexpected is just how much more effective the Bradley would prove in 2025, 45 years after its release. The Bradley has become a favorite of Ukrainian soldiers for its survivability, mobility, and lethality. While still susceptible to drone strikes, its smaller size and increased mobility make it a more challenging target for drone operators. According to the Wall Street Journal, the protracted war has still claimed over 60% of the 300 Bradleys pulled from storage and shared with Ukraine. But, unlike the M1 Abrams, soldiers stand by the Bradley’s combat effectiveness against the new kinds of threats. Terrible requirements process, decent outcome, and a surprising increase in its utility that also was not predicted.

Together, these examples illustrate just how unpredictable the effectiveness of the Department of Defense’s procurement efforts can be. Every military has had to deal with uncertainty, the fog of war. The military has their own framework for how to make decisions under uncertainty on the battlefield: the OODA Loop. OODA stands for Observe, Orient, Decide, Act, at which point the cycle starts again. This framework is useful for understanding how different groups propose fixing the Department of Defense’s problems.

To understand how different groups approach fixing the Department of Defense's OODA Loop problems, we must first examine how US defense technology strategies have evolved through distinct historical phases, each with its own approach to managing uncertainty. There have been roughly three major evolutions of US defense technology. The first evolution took place during World War II when victory was achieved with good-enough-but-not-the-best technology that was deployed quickly, at scale, and rapidly iterated upon to improve performance. The attitude was a reflection of American industrial power and entrepreneurialism, built on relatively fast loops to resolve urgent issues across two unique battlefields in Europe and Asia. 

The next evolution was brought about by the atomic age and the increased technical sophistication of the US military post-World War 2. The Manhattan Project served as an inflection point in how science and technology would be the decisive factor in warfare. Nuclear weapons were so powerful that, for a time, the military questioned if there would ever be conventional armies again. In fact, nuclear weapons were so powerful they became unthinkable to use, as evidenced by the US deciding not to use nuclear weapons during the Korean War. Conventional militaries would still be relevant, and, for them to serve as effective deterrents, they too would need the most advanced weapons science and technology could create. These projects tended to emphasize superior technical capabilities over sheer quantity of weapons. Radar, satellites, jet engines, cryptography, and communication systems evolved quickly over this period, all backed by significant research and development (R&D) initiatives that attempted to balance superior products available today and funding longer-running R&D projects that would enable superiority in the next decade of competition. During this period, there was enough budget to dedicate to fast and slow OODA loops. 

The third evolution came with the end of the Cold War and “The Last Supper”. With the dissolution of the Soviet Union, the US stood alone in the size and capability of its military. Nuclear deterrents were in place, its conventional military outclassed everyone, and the world seemed poise for peace. The US believed it had sufficient deterrent, and therefore it could reduce defense spending significantly—the so-called Peace Dividend. The Last Supper refers to a dinner at the Pentagon on July 21, 1993 between then-U.S. Secretary of Defense Les Aspin, Deputy Secretary of Defense William Perry, and executives at defense contractors. The defense contractors were told to consolidate or go out of business. The industry consolidated from 51 contractors to just 5 today. 

With limited budget, feedback loops would slow to a crawl for all but the most strategic of programs. Through the Global War on Terror, the Pentagon would shift more spending towards protecting soldiers against improvised explosive devices, drones, and other surgical strike capabilities, while debating if it would really need to advance technologies for conventional warfare and maintain or expand its size. 32 years later, as Russia attempts to conquer Ukraine and China prepares for a conquest of Taiwan, US defense tech strategy faces a fourth evolution.

One solution just calls for faster cycles: reducing the time between observing a problem, orienting towards what a solution might look like, deciding on one plan among different options, then acting on it. The longer it takes to go through the loop, the more likely the situation has changed when it is time to decide and act. See the 22 years of development for the Bradley—the more they debated the right requirements, the more the situation would change, and the more they would have to revisit the requirements and revise them, like a dog chasing its own tail. Nathan Mintz, a serial founder of defense tech startups including Epirus and now CX2, writes on his Substack that any defense tech company’s “ultimate goal” is “delivering cutting edge systems to the customer as quickly, affordably and reliably as possible”. Note that quickly comes first. Eric Lofgren, who wrote extensively about how the Pentagon procures new technical capabilities on his site Acquisition Talk and is now a Professional Staff Member with the Senate Armed Services Committee, wrote, “It currently takes two years to approve a requirement, two years to find funding, and two years to get a company on contract. That is not relevant. Relevance is delivering capabilities faster than our adversaries and at the pace a conflict could require.” 

A second solution calls for better decision making, in this case better training for how defense tech planners orient to the problems at hand. Lofgren believes the DoD must “make room for opportunities in program requirements”. He means that requirements should not be overly prescriptive of what technologies must be used but emphasize what outcomes are necessary. By requiring outcomes rather than technologies, the teams building the systems can find the best solutions and avoid being stuck implementing a technology they do not believe is the best. In other cases, Lofgren believes planners simply miss core requirements, citing an example of forgetting to include a requirement that headsets for soldiers should not interfere with their regular combat actions. Better training could lead to better requirements, making for a smoother process and greater certainty of a useful piece of technology.

A third solution is what the rest of the economy does: competition and specialization. In this case, competition means competition inside the DoD. At the start of World War 2, the military had little industrial expertise and leaned significantly on industrialists and innovators to define the platforms that would best meet their needs. Companies worked directly with different branches on their unique platforms. Each service wrote its own requirements and spent its own budget on the vehicles that best met its needs. For example, the Navy needed planes that could launch from aircraft carriers but the Air Force did not. The result was a lot of money spent on similar technologies at the beginning of the Cold War. In contrast to the start of World War 2, the department no longer lacked industrial expertise—in fact, the military often believed they had more expertise than the contractors building components for bleeding edge applications like jet engines.The department came to see outsourcing to industry across each branch as duplicative and, in the interest of efficiency, pushed for centralized control of platforms that would serve all branches with as many common components as possible. 

That ethos carries through today. The DoD acts as a monopsony, a market situation where there is only one buyer of a product. If there is only one buyer of a product, the buyer can effectively dictate the features and price of the product to their suppliers, who risk losing their only source of revenue if they disagree. Monopsony gives one customer extreme power over potential suppliers––essentially creating the inversion of monopoly dynamics, with similar negative effects on product quality. The Lockheed Martin F-35 Lightning II is a recent example of this kind of planning. The F-35 was intended to be a fighter jet that could serve the needs of the US Air Force (conventional takeoff and landing from a runway), Navy (takeoff and landing from a carrier), and Marine Corps (short takeoff/vertical landing) with three variants of the same design that would share 80% of the same components. Its development took 25 years, the cost for the entire program was billions over its budget, and the variants only ended up sharing 20% of the same parts. By breaking up this centralized planning and letting each service set their own requirements, the feedback loop could produce better requirements that are faster to implement. Shyam Sankar, CTO of Palantir, believes fixing this issue would deliver the biggest results. On October 31, 2024, Sankar published his “18 Theses for Defense Reformation.” Number 1 is that “Monopsony is the root of what ails us.”

While the third solution focuses on creating internal competition within the DoD's existing structures, the fourth approach represents a more radical departure—letting industry itself, particularly startups with Silicon Valley DNA, drive both requirements and implementation. In this case, the Department of Defense or industry could observe the problem and independently orient, decide, and act, to varying degrees. This model is something of a return to defense philosophy at the start of World War 2, that leveraged technologists and industrialists to make appropriate tradeoffs for defense tech. These aspiring defense contractors, almost universally tech startups, believe they can outcompete the current defense contractors and the DoD planners in getting the right requirements and accelerating the time to deploy. The startups believe this model can unlock a dimension the Department of Defense has historically struggled with: software. Of course, software powers the exquisite systems of today, and a code update alone cannot make a main battle tank less vulnerable to a drone. 

Rather, their products are architecturally different, starting with a software-defined platform that makes integrating new pieces of hardware much easier and faster. Traditional defense systems are like custom homes where every component—from the foundation to the roof—is specifically designed for that particular building. If you want to add a new room or upgrade the kitchen, you need to redesign major structural elements, often requiring extensive demolition and rebuilding. Each system (like the M1 Abrams or F-35) has its own unique technical foundation, with software and hardware so tightly intertwined that updating either requires complex, expensive modifications to the entire system.

In contrast, startups propose software-defined platforms that function more like modular homes built on standardized foundations with consistent interfaces. The software acts as a universal translator and integration layer, allowing new hardware components to be 'plugged in' without redesigning the entire system. When a new threat emerges or technology advances, they can swap out individual components—sensors, weapons, communication devices—without rebuilding the entire platform. This approach prioritizes adaptability over optimization for a single purpose, trading some specialized performance for the ability to rapidly evolve as battlefield conditions change.

The consumer example for this kind of product is a Tesla electric vehicle. Each vehicle uses the same core technologies, so it uses the same software platform to integrate the components together. The system built to monitor the battery in a Model S can be reused in a Model 3. The software system that translates pressure on the accelerator to power to the motors can be shared, albeit tuned to specific batteries, motors, and driving dynamics of a new vehicle. There is always some work to be done for integrating each new platform, but it is still extremely efficient with engineering hours. Just as Tesla has no model years but continually refines their products for cost and new market requirements, defense contractors could continually upgrade their products to new battlefield requirements and technology shifts.

Additionally, new suppliers can take new kinds of risks. One growing concern is that the Department of Defense has made expensive weapons at the expense of making lots of weapons. For example, in a war game published by the think tank Center for Strategic and International Studies that simulated a non-nuclear conflict between China and the US over Taiwan, the US typically runs out of missiles after 8 days. Following that 8 day period is a race to restock missiles, when the Chinese industrial output dwarfs the US output. Scale issues plague other parts of the US. The US Navy has expressed concern that, given the shrinking workforce of shipbuilders, its suppliers can only produce about 20% of the ships that it believes it needs and that Congress has budgeted. The US Navy believes Chinese shipbuilding capacity is roughly 232 times that of the United States. With data points like this, it is not difficult to imagine a world where China easily eclipses the US military in both weapons capability and output. 

Anduril, the pure defense startup with the most revenue and highest valuation, wants to address scale from the requirement stage. In their mission document, Rebooting the Arsenal of Democracy, Anduril argues that the software-powered approach could be designed to scale from the beginning, adding a constraint that the supply chain for the parts be robust and likely already broadly used in industry instead of being custom-made for military specifications. Other primes like Palantir share this vision, though Anduril has laid out the most concrete vision of how they would develop a comparable solution with their Arsenal-1 Manufacturing Facility. The defense tech startups believe they can master emerging technologies faster than traditional defense contractors, notably AI-powered autonomous systems. 

These ambitious visions for reforming defense technology development are beginning to move from theory to practice as startups directly challenge traditional primes in the competitive arena, even within the constraints of the current procurement system. In April 2024, Anduril and General Atomics, the traditional defense contractor known for their drones, were selected for the next phase of the US Air Force’s collaborative combat aircraft (CCA) program, beating proposals from Lockheed Martin, Boeing, and Northrop Grumman. What is less obvious is how the traditional primes may survive if the Department of Defense adopts proposed acquisition reforms. 

The new defense contractors are explicitly fighting to avoid becoming like today's primes. They reject the cost-plus pricing model—where contractors are paid for their costs plus a guaranteed profit margin—which they argue incentivizes delayed deliveries, budget overruns, and risk aversion rather than innovation. Startups like Anduril and Palantir often develop capabilities with their own capital first, then seek contracts, inverting the traditional model where requirements precede development. They prefer fixed-price contracts that reward efficiency and on-time delivery. This fundamental business model dispute reflects their belief that the current primes aren't just participants in a broken feedback system but are themselves products of it, shaped by decades of perverse incentives that a technical fix alone cannot resolve.

In many ways, the primes are defined by the incentives under which they operate. Since their forced consolidation, the DoD works to ensure they all stay financially healthy to prevent further consolidation, easing the competitive pressures. While some were simply business units of industrial giants, many of them were once high-flying startups themselves who had to adapt to DoD contracting rules to grow their revenue and valuation. One view for why the modern startups are pushing so hard for reformation is to avoid this exact fate, servants of a system nobody thinks is working well. These startups' founders have explicitly stated they have no interest in becoming the next generation of traditional defense contractors. Their mission is not merely to win contracts within the existing system but to fundamentally transform how defense technology is developed, priced, and delivered.

Finally, for all the complexity, delays, and cost overruns, one critical point remains: the current prime defense contractors have delivered good products. Consider the aforementioned example of the F-35 fighter plane. Despite its decades in development, cost overruns, and failure to meet the initial goals, pilots and military analysts testify to how well the delivered plane fits into modern US air doctrine, leading with stealth; computerized automation to map, identify, and track threats; versatile targeting systems; and effective coordination with other units. Allied countries are voting with their dollars too, as the list of countries presently or intending to operate F-35s has grown to 17 even with an $80 million price tag. And the startups are, well, just starting up: they are delivering important capabilities such as drone vehicles and sensor platforms for detecting threats but have not yet fielded offensive weapon systems. Anduril’s Arsenal-1 manufacturing plant is an impressive plan that could solve a critical issue, but Anduril has yet to turn their modular design philosophy––where missiles are made with commercially available parts into a product–– let alone a product that can compete with the most performant missiles the military uses today. Just as the primes need to show they can update to meet the demands of today, startups also have a lot to prove.

These solutions are not mutually exclusive, nor do they exhaust all possible improvements to defense procurement. The M1 Abrams and Bradley Fighting Vehicle stories illustrate an uncomfortable truth: predicting which systems will prove most valuable decades after their development remains largely impossible. This unpredictability suggests that what the Department of Defense needs is not just faster OODA loops or better requirements, but a procurement ecosystem diverse and adaptable enough to respond to unforeseen threats and opportunities.

The competition between traditional primes and new entrants may ultimately prove more valuable than any single approach. Traditional contractors bring decades of experience delivering complex military systems that work in harsh environments, while startups bring software expertise and business models that prioritize speed and adaptation. The future of American military innovation likely depends not on choosing a single winner in this contest of ideas, but on intelligently blending these approaches to create a more resilient defense industrial base. That future is impossible, though, if the Department of Defense demands that the new defense contractors operate exactly as the previous ones did—something has to give.

As geopolitical tensions rise and technological change accelerates, improving this system is not merely an academic or bureaucratic concern. The security of the free world is at stake. For primes and startups alike, Shyam Sankar has given a unifying rallying cry: 'Thesis 5: The only requirement is winning.'"