The Old Guard's Quiet War

The Point

In the late 1950s, the Bethlehem Steel works at Sparrows Point, on the Patapsco River just southeast of Baltimore, was the largest steel plant in the world. It stretched some four miles along the waterfront, employed more than 30,000 people, and rolled the plate for ships that fought in both world wars. Steelmaking there ended in 2012. For a decade the site’s most visible tenants were warehouses.

On October 6, President Donald Trump stood at 1660 Shipyard Road and announced that Sparrows Point would build warships again, or at least pieces of them. The company behind the project is Anduril Industries, a nine-year-old, venture-backed defense firm named for a sword in The Lord of the Rings. Its new facility, Arsenal-2, is planned as a two-million-square-foot “software-defined shipyard” on 187 acres, with a 1,200-foot graving dock. “Sparrows Point helped build the fleet that won World War II,” Anduril’s CEO, Brian Schimpf, said in the state of Maryland’s announcement.

The headline number was $6.6 billion, but it is worth being precise about what that figure contains. Anduril says it will put up $3.7 billion of its own capital to build and staff the yard. The remaining $2.9 billion is a ceiling, not a check: a Navy contract worth up to that amount, tied to what Anduril actually delivers. The company has framed the arrangement as one in which Anduril, “not the taxpayer,” carries most of the execution risk.

What Arsenal-2 will make is also narrower than the ceremony suggested. Its first products are components for Virginia-class attack submarines, torpedo tubes and spherical air flasks, which will be shipped to the only two yards that build American nuclear submarines: General Dynamics’ Electric Boat in Connecticut and Huntington Ingalls’ Newport News Shipbuilding in Virginia. Initial operations are expected in 2030. An Anduril official told The War Zone that there is not yet a timeline for when its parts will be certified for installation on a submarine.

In other words, the newest entrant in American naval shipbuilding has arrived not as a rival to the old guard but as its supplier. That is the more interesting story. Washington has not decided to replace the legacy defense industry. It has decided that the legacy industry, on its own, can no longer build what the country needs, and it has begun paying outsiders to fill the gap.

The Arsenal, Remembered Selectively

Anduril did not choose the word arsenal by accident. In a radio address on December 29, 1940, Franklin Roosevelt told Americans what their country had to become.

“We must be the great arsenal of democracy.”

Anduril has borrowed the phrase for years. Its founder’s manifesto was titled Rebooting the Arsenal of Democracy, and its first factory plan, Rebuilding the Arsenal of Democracy. The borrowing carries an implied argument: that the country once knew how to arm itself quickly, and that the defense establishment has since forgotten.

There is truth in that story. Roosevelt said as much that night, telling listeners that makers of watches, cash registers, sewing machines, and lawn mowers were already turning out fuses, shells, and pistols. When the War Production Board was created in January 1942, the government turned to many more firms that had never built a weapon. Chrysler ran the government-owned Detroit Tank Arsenal. Ford built B-24 bombers at Willow Run. Typewriter companies retooled for small arms: Remington Rand produced nearly 900,000 M1911A1 pistols, more than Colt itself, while Smith-Corona made M1903A3 rifles and IBM made M1 carbines. The Navy, in its own postwar history, recorded that contracts were negotiated directly with firms rather than advertised for bids, and that advance payments were made when contractors needed cash to start work.

But the romantic version omits two inconvenient facts. The first is that the mobilization did not lack specialists. The B-24 that Ford mass-produced was designed by Consolidated Aircraft; Boeing, Douglas, and Lockheed were already established military suppliers, and the Army’s arsenals and the Navy’s yards had been building weapons for generations. Detroit’s genius was volume, not invention.

The second fact is more awkward for today’s reformers. The contract form now treated as the emblem of defense-industry complacency, cost-plus, was the engine of the wartime build-out. By the Navy’s accounting, almost three-quarters of the value of its wartime base-construction contracts used cost-plus-a-fixed-fee terms, adopted precisely because fixed prices would have been too slow. The factories themselves were often public property: the Defense Plant Corporation spent more than $9 billion on some 2,300 projects between 1940 and 1945. Nor was the effort free of oversight; a Senate committee led by Harry Truman spent the war investigating waste and profiteering.

The lesson of the original arsenal, then, is not that rules were absent. It is that the government, facing an existential deadline, was willing to spend its own money, absorb its own risk, and bend its own procedures to buy at scale from whoever could produce. What changed afterward was not the existence of rules but who learned to live inside them.

Fifty-One to Five

The modern defense industry was shaped less by the Second World War than by the end of the Cold War. As budgets fell in the 1990s, the Pentagon encouraged its suppliers to merge, and they did.

51→5

By the Defense Department’s own count, the five survivors were Lockheed Martin, Raytheon (now RTX), General Dynamics, Northrop Grumman, and Boeing.

Consolidation produced something closer to a regulated utility than a market. Weapons programs ran for decades; the government was the only customer; and the firms that survived were the ones that had mastered the government’s paperwork. It is tempting to say that cost-plus contracts let them profit regardless of performance, but the record is more mixed. Boeing’s charges on the KC-46 tanker, a fixed-price development program, had passed $7 billion by 2023, more than the Air Force’s $4.9 billion contract ceiling. The deeper problem was not that the primes were coddled. It was that the system rewarded the skills of a compliance organization over those of a factory.

Meanwhile, American commercial manufacturing went elsewhere, and in one case it took an extraordinary amount of know-how with it. In Apple in China (2025), the former Financial Times correspondent Patrick McGee argues that Apple did not merely use China’s factories; it helped build them. Drawing on internal documents, he reports that by 2015 Apple was investing roughly $55 billion a year in China, much of it in machinery and in training the engineers of its suppliers. Apple itself has said it trained more than 28 million workers there since 2008, more people than the entire labor force of California.

Apple trained assembly workers and process engineers, not submarine welders, so its lesson is not about phones. It is about the layer that sits between a design and a finished product: the process engineers, toolmakers, and second-tier suppliers that let a factory retool quickly and hold tolerances at volume. American defense firms kept the designs and the contracts and learned to manage programs. American technology firms, chasing cost and speed, helped build much of that middle layer abroad. Shipyards depend on their own version of that layer, and China, through decades of state industrial policy rather than anything Silicon Valley did, built one at home.

The result is visible in the tonnage. By the Center for Strategic and International Studies’ count, Chinese yards now produce more than half of the world’s commercial shipbuilding tonnage; American yards produce about 0.1 percent. In 2024 a single state-owned firm, China State Shipbuilding Corporation, built more commercial tonnage than the entire American industry has built since the end of the Second World War, and its largest yards build warships alongside container ships.

When the Pentagon now looks across the Pacific, then, it sees a navy drawing on an industry that operates on commercial timelines. When it looks at home, it sees Virginia-class submarines that, according to the Congressional Research Service, have never been delivered at the planned rate of two a year and have been built at about 1.1 to 1.2 a year since 2022.

The Side Door

The barrier facing a software or hardware startup was never a single regulation. It was the accumulated weight of the Federal Acquisition Regulation, its defense supplement, and the habits of a contracting workforce trained to buy bespoke systems. Founders worried most about two things: accounting rules built for cost-reimbursed contracts, and the government’s appetite for the technical data and software rights behind what it bought.

Some of those fears were older than the companies that held them. Congress told agencies in the Federal Acquisition Streamlining Act of 1994 to prefer commercial products, and a 1996 follow-on law exempted commercial-item contracts from the government’s Cost Accounting Standards. The problem was that the Pentagon often declined to treat what it was buying as commercial. Palantir tested that habit in court. In 2016 it sued the Army over a contract for an intelligence-analysis system, and a federal judge found that the service had broken the 1994 law by never seriously asking whether a commercial product could do the job. The Army went on to pick Palantir and Raytheon to build the system, and an appeals court upheld the ruling in 2018.

The more important door, though, was a legal instrument called an “other transaction.” The concept dates to the 1958 law that created NASA, which let the agency enter agreements that were neither contracts nor grants. The Defense Department received its own version much later: for research in 1989, for prototypes in 1994, and, in the decisive change, for follow-on production in 2016, according to the Pentagon’s account. Other transactions sit outside most procurement statutes and the FAR. Companies can be selected on the strength of a short solution brief rather than a thousand-page proposal, through organizations such as the Defense Innovation Unit or industry consortia.

The production authority is what turned experiments into businesses. Under what is now 10 U.S.C. §4022, a successful prototype can move into production without a new competition, provided the original prototype award was competitive and the project was completed successfully. Before 2016, a startup that won a prototype deal had to re-compete for production under traditional rules, often against incumbents far better equipped for that contest. The Defense Department itself has acknowledged that this cliff discouraged nontraditional companies from bidding at all.

The side door may soon matter less, because the front door is being rebuilt. In April 2025, an executive order launched what the administration calls the “Revolutionary FAR Overhaul,” and the FAR Council has since begun turning it into permanent regulation, proposing rules covering twenty parts of the FAR in June 2026 and sixteen more in September, with a goal of finishing by the end of this year. The overhaul is meant to strip the FAR back to what statutes require, and the rules for commercial purchases show what that means in practice. Nearly every commercial solicitation today carries two omnibus items: a lengthy set of representations and certifications, and a “clause of clauses” that lists dozens of requirements drawn from statutes and executive orders, with the contracting officer checking off the ones that apply. The September proposal would delete both, so that each clause has to be prescribed and included on its own. A companion rule on contractor responsibility cuts in a direction newcomers may find even more useful. It would restore poor past performance as grounds for turning a bidder away, while the FAR already bars contracting officers from rejecting a firm solely because it has no relevant record. That is a standard a company without a history can meet more easily than an incumbent with a bad one. If the overhaul works, the distinction between the startups’ express lane and everyone else’s procedure could narrow considerably.

The Software Layer and the High Ground

Anduril is the youngest of the challengers, and Arsenal-2 shows it entering the industrial base from below, as a supplier to the yards it might one day rival. Two older Silicon Valley companies show where that path can lead. Palantir has become something close to the operating system of the Army. SpaceX has done what no newcomer was supposed to be able to do: take a core military mission away from the incumbents outright.

Palantir’s arc is the cleanest measure of how much has changed in a decade. The company that sued the Army in 2016 for refusing to consider its product is now the Army’s default. In July 2025 the service folded 75 separate contracts, 15 on which Palantir was the prime and 60 more on which it held a share, into a single enterprise agreement that lets the Army and other Defense Department components buy Palantir software for up to ten years and up to $10 billion. Like the Navy’s commitment at Sparrows Point, that figure is a ceiling rather than a check, and the Army presented the deal as a cost measure: volume discounts, no reseller markups. But a ten-year enterprise agreement is also how a buyer signals that it does not expect to switch.

The same pattern runs through the Maven Smart System, the AI targeting and command-and-control platform that grew out of the Pentagon’s 2017 Project Maven. In May 2024 the department gave Palantir a five-year, $480 million Maven contract without competition, having concluded that no other company could meet the requirement. A year later it raised the ceiling by $795 million, to nearly $1.3 billion through 2029. In March, Deputy Defense Secretary Steve Feinberg directed that Maven become a formal program of record by the end of the fiscal year, with every Maven contract moved onto the Army’s Palantir agreement. A program of record has its own line in the budget. As an analyst at the Project on Government Oversight put it, that is about as permanent as a Pentagon program gets.

The revenue has followed. Palantir reported $809 million in U.S. government revenue for the second quarter of 2026, up 90 percent from a year earlier. Annualized, that is still small beside Lockheed Martin, which booked $71 billion in sales in 2024. But Palantir’s position does not depend on size. Software that sits between every sensor and every shooter does not have to be the largest contract to be the hardest one to replace.

SpaceX took the opposite route. Rather than the software layer, it went after the most physical bottleneck in national security: getting into orbit. For most of the previous decade, military launch belonged to United Launch Alliance, the joint venture Boeing and Lockheed Martin formed in 2006, itself a product of the consolidation era. In 2014 SpaceX sued the Air Force over a 36-core block buy awarded to ULA before SpaceX’s rocket could be certified to compete. Like Palantir, it used the courts to pry open a market; the case settled in 2015 with the Air Force agreeing to compete future launches.

A decade later, the competition is lopsided. In April 2025 the Space Force divided the high-end lane of its Phase 3 launch program for fiscal years 2025 through 2029: SpaceX was expected to fly 28 missions, about 60 percent, for roughly $5.9 billion; ULA got 19; Blue Origin, seven. Those are firm-fixed-price contracts, the opposite of the cost-plus terms that built the wartime arsenal, and the newcomer won most of them.

Launch was the beginning. The National Reconnaissance Office has been filling low orbit with what it calls the largest government constellation in history. The satellites are widely believed to be Starshield, a military variant of Starlink, built by SpaceX with Northrop Grumman, under a $1.8 billion contract first reported by Reuters. Then came Golden Dome, the administration’s missile-defense shield. In April the Space Force named SpaceX among twelve companies sharing up to $3.2 billion to prototype space-based interceptors. In the last week of May, the Space Force awarded SpaceX $2.29 billion for a military data network in orbit and, two days later, $4.16 billion for a constellation to track aircraft and cruise missiles from space, the kind of sensing layer Golden Dome depends on. Both were other transactions: the side door described above, now carrying awards larger than many traditional programs. By August, The Wall Street Journal counted SpaceX’s Golden Dome contracts at more than $8 billion. In June, the company went public at a valuation of about $1.77 trillion, the largest offering in history.

This is where the challengers’ story begins to rhyme with the one they set out to disrupt. The Pentagon’s lament about the 1990s was that it let fifty-one primes become five and lost its leverage along the way. In launch, it has gone from depending on one provider to depending mostly on another. The cost of that is not hypothetical. In 2022, Ukraine asked SpaceX to extend Starlink coverage to Sevastopol for a drone strike on Russia’s Black Sea Fleet. Musk refused, saying he did not want SpaceX complicit in a major act of war. Whatever one makes of his judgment, the episode showed a private executive making a call with battlefield consequences that, under the old system, would have belonged to a government. The Space Force seems to know the risk: announcing the $4.16 billion award, it promised further awards to drive “a vendor-diverse expansion.”

Loud and Quiet

The two sides of this contest do not speak the same language, or at the same volume. The newcomers write manifestos. Palmer Luckey’s Rebooting the Arsenal of Democracy argues that a Tesla carries more artificial intelligence than any American military vehicle. Anduril’s president, Chris Brose, made the case to Congress in December 2024 testimony before the House select committee on China. In 2022, the company’s chief revenue officer told a defense conference that for new firms working under traditional FAR contracts, there was “no fair competition.”

The incumbents rarely answer in kind. Their influence runs through program offices, trade associations, and appropriations committees, where a well-placed clause matters more than a press release. Yet the vocabulary of grievance is not unique to the startups. When Boeing withdrew from the competition to build the Air Force’s next intercontinental ballistic missile in 2019, it said the acquisition approach did not offer a level playing field for fair competition, leaving Northrop Grumman as the sole bidder, as the Pentagon later noted. Every contractor believes the rules favor someone else.

The strongest version of the incumbents’ case is not about fairness but about time. Venture capital can subsidize a company through years of losses, but it cannot certify a torpedo tube. Parts destined for a nuclear submarine must pass the Navy’s SUBSAFE quality regime, and Anduril has said its new Orange County, California, plant exists partly to work through that qualification before Arsenal-2 opens. A drone that fails in testing is a setback. A hull penetration that fails at depth is a catastrophe. Electric Boat and Newport News have spent generations learning that difference, and no amount of software changes the metallurgy.

The newcomers’ record is not uniformly triumphant, either. Anduril recently lost a Navy competition for medium-sized drone boats, and Navy officials have acknowledged navigation and software problems across the autonomous-vessel sector. Its first Arsenal factory, in Ohio, produced its first aircraft in July but remains a work in progress.

What has unmistakably shifted is proximity to power. Last month, Defense Secretary Pete Hegseth chartered Project Meridian, an effort to map the future of warfare, and, per a Pentagon memo, named Luckey among its leaders, alongside Elon Musk and Newt Gingrich. Both Luckey and Musk run companies that hold, and seek, large federal contracts; Musk’s has won more than $8 billion in Golden Dome work alone. The legacy primes spent decades cultivating the bipartisan machinery of Congress. Their new competitors have gone straight to the executive branch, and less than a week after Meridian was announced, the president himself came to Sparrows Point to make Anduril’s announcement.

Who Owns the Factory Floor

It would be easy to read Sparrows Point as the moment the old guard lost. That reading is premature. Lockheed Martin, General Dynamics, and their peers still hold the integration expertise, the certified workforces, and the program relationships that no startup can buy in a funding round. Arsenal-2 will not deliver its first certified part for years, and when it does, that part will be bolted into a submarine at a General Dynamics or Huntington Ingalls yard.

But the arrangement still marks a break. For a generation, the Pentagon treated its prime contractors as the industrial base. Now it is paying a company that did not exist a decade ago to build capacity beneath them. In one respect, that is closer to Roosevelt’s arsenal than anything the department has tried since: a government in a hurry, deciding that it will buy production from whoever can deliver it. In another, it is the arsenal turned inside out. In the 1940s, Washington financed much of the plant with public money and bought on cost-plus terms, which covered most of the Navy’s wartime base construction. At Sparrows Point, at least as Anduril describes the deal, private capital builds the yard and the government pays only for parts delivered. The urgency is Roosevelt’s. The balance sheet belongs to venture capital, whose patience is measured in fund cycles rather than wars, and whose returns depend on the government continuing to buy.

Palantir and SpaceX suggest how the story could end, and it is not with the abolition of an old guard but with the arrival of a new one. A decade ago both companies were suing the military to be allowed to compete. Today one runs the software beneath the Army’s targeting and the other flies most of its highest-value launches and is building the sensors for its missile shield. They are indispensable, concentrated, and politically connected, which is a fair description of the primes in 2000. The Pentagon’s real test is not whether it can bring outsiders in. It is whether it can keep asking more than one of them.

Whether the newcomers can actually deliver is the question the next decade will answer. The old guard is not in a panic, exactly. It is discovering something more unsettling: that for the first time since the 1990s, it is not the only one the government is willing to ask.

Sources

Government and primary documents

1. Anduril, “Anduril, U.S. Navy Invest $6.6 Billion to Expand America’s Submarine Industrial Base,” Oct. 6, 2026.
2. Maryland Department of Commerce, Arsenal-2 announcement, Oct. 6, 2026.
3. U.S. Department of Defense, State of Competition within the Defense Industrial Base, Feb. 2022.
4. Department of Defense, Project Meridian commissioning memo, Sept. 30, 2026.
5. Congressional Research Service, Navy Virginia-Class Submarine Program and AUKUS (RL32418).
6. Congressional Research Service, DOD Use of Other Transaction Authority (R45521), 2019.
7. FAR Council, Revolutionary FAR Overhaul, Acquisition.gov.
8. FAR Council, proposed rule, FAR Case 2026-003 (Parts 8, 12, 13, 15, 38, 44, 51, and 52), Federal Register doc. 2026-19162, 91 FR 59406, Sept. 18, 2026.
9. FAR Council, proposed rule, FAR Case 2026-011 (Parts 9, 27, 47, and 52), Federal Register doc. 2026-19159, 91 FR 59584, Sept. 18, 2026.
10. Federal Acquisition Regulation, 9.104-1, General standards.
11. Franklin D. Roosevelt, Fireside Chat on the “Arsenal of Democracy,” Dec. 29, 1940.
12. Franklin D. Roosevelt, Executive Order 9024, establishing the War Production Board, Jan. 16, 1942.
13. U.S. Navy Bureau of Yards and Docks, Building the Navy’s Bases in World War II, 1947.
14. House Select Committee on the CCP, hearing with Chris Brose, Dec. 16, 2024.
15. Palmer Luckey, Rebooting the Arsenal of Democracy (Anduril).
16. Palantir USG, Inc. v. United States (Fed. Cl. 2016), aff’d (Fed. Cir. 2018).
17. U.S. Army, “U.S. Army Awards Enterprise Service Agreement to Enhance Military Readiness and Drive Operational Efficiency,” July 31, 2025.
18. U.S. Space Force, National Security Space Launch Phase 3 Lane 2 contract awards, Apr. 4, 2025.
19. Palantir Technologies, Q2 2026 results, Form 8-K, Exhibit 99.1, Aug. 3, 2026.
20. Lockheed Martin, Fourth quarter and full year 2024 results, Form 8-K, Exhibit 99.1, Jan. 28, 2025.

Research

1. Center for Strategic and International Studies, Ship Wars: Confronting China’s Dual-Use Shipbuilding Empire, Mar. 11, 2025.

Books

1. Patrick McGee, Apple in China: The Capture of the World’s Greatest Company (Scribner, 2025).
2. Mark Reutter, Making Steel: Sparrows Point and the Rise and Ruin of American Industrial Might (review).

Reporting

1. Howard Altman, “Anduril Dives Into Submarine Construction With New Baltimore Plant,” The War Zone, Oct. 6, 2026.
2. “Anduril CRO: ‘There is no fair competition’ for non-traditional companies in defense,” Breaking Defense, Feb. 2022.
3. “Boeing hit with fresh $245 million KC-46A charge,” Breaking Defense, Apr. 2023.
4. “Judge rules in favor of Palantir in lawsuit against US Army,” Defense News, Oct. 31, 2016.
5. Arnold & Porter on the Federal Circuit’s Palantir ruling, Sept. 2018.
6. Citybiz on Anduril’s Sparrows Point talks, summarizing The Wall Street Journal, Aug. 2026.
7. PBS, “Brother, Can You Spare a Billion? The Story of Jesse Jones” (Defense Plant Corporation).
8. “American Arsenal,” GUNS Magazine.
9. “New contract expands Maven AI’s users ‘from hundreds to thousands’ worldwide,” Breaking Defense, May 2024.
10. “‘Growing demand’ sparks DOD to raise Palantir’s Maven contract to more than $1B,” DefenseScoop, May 23, 2025.
11. On the Feinberg directive making Maven a program of record, DefenseScoop, Apr. 3, 2026.
12. “DOD components face ‘aggressive’ timeline for Maven Smart System transition,” DefenseScoop, Apr. 15, 2026.
13. “SpaceX Files Suit Against Air Force,” Via Satellite, Apr. 28, 2014.
14. “SpaceX Ends Lawsuit, Settles With Air Force for Future Launch Bids,” Defense One, Jan. 2015.
15. “Space Force awards first next-gen critical launch contracts to Blue Origin, SpaceX, ULA,” Breaking Defense, Apr. 2025.
16. First NRO proliferated-architecture launch, Spaceflight Now, May 2024.
17. NROL-153 Starshield launch, Spaceflight Now, Jan. 2025.
18. “SpaceX, Anduril Among Companies to Win Golden Dome Contracts,” Bloomberg, Apr. 24, 2026.
19. On the SB-AMTI and Space Data Network awards, Inside Unmanned Systems, May 2026.
20. “SpaceX wins $4.16B Space Force contract to detect airborne moving targets,” Breaking Defense, May 29, 2026.
21. 24/7 Wall St. on SpaceX’s Golden Dome awards, summarizing The Wall Street Journal, Aug. 2026.
22. SpaceX Nasdaq listing, SpaceQ, June 12, 2026.
23. “Musk says he refused Kyiv request for Starlink use in attack on Russia,” Reuters, Sept. 2023.
24. Wiley Rein on the second round of FAR overhaul proposed rules, Sept. 2026.