Latest / The Fiscal Policy Podcast with Fexingo: Government Budgets, Stimulus, and Public Spending / How the Government Spends 4 Billion on Obsolete Mainframes
Transcript
- Lucas: Lucas here. Over the past few episodes, we've talked a lot about the government spending money on things like unused buildings and redundant software licenses. But there's a category of spending that's even more stubborn — the billions of dollars the federal government spends every year keeping mainframe computers running. Machines that are, in some cases, older than the people writing the checks to maintain them. Luna: Mainframes. Like, the giant beige boxes from the 1970s? Lucas: Exactly. We're talking about IBM zSeries mainframes — some of them still running code written in COBOL in the 1980s. The Government Accountability Office estimated that federal agencies spend roughly $4 billion a year on mainframe-related expenses. That's not including the cost of migrating off them, which would be enormous. But the real question is: why are we still paying for technology that most of the private sector abandoned two decades ago? Luna: I remember reading that the Social Security Administration still processes benefit calculations on a mainframe running COBOL. They pay something like $70 million a year just to maintain that one system. Lucas: That's the classic example. The SSA's mainframe handles over a billion transactions a year — things like benefit adjustments, earnings records. It was built in the 1980s, and they've been layering patches on top of patches ever since. The annual maintenance contract with IBM runs about $70 million. But that's the visible cost. The hidden cost is that you cannot recruit a generation of developers to learn COBOL and work on a 40-year-old system. The workforce is retiring, and the knowledge is walking out the door. Luna: So what happens when those COBOL programmers retire? Do they just pay more to keep them? Lucas: They do. Agencies have been known to pay retired contractors triple their original salary to come back and maintain the code. There's a famous story from a few years back — the State of New Jersey couldn't process unemployment claims during the pandemic because their mainframe couldn't handle the load and they had to track down a 72-year-old COBOL programmer to fix it. The federal government is in a similar situation, just at a much larger scale. Luna: And it's not just the Department of Defense or the IRS — it's across the board. The Department of Veterans Affairs, the Treasury, even the Department of Agriculture. Lucas: Right, the VA still uses a mainframe for health records. The Treasury uses one for debt management. But let's be fair — there are reasons these systems survive. They are incredibly reliable. Mainframes have uptime measured in decades, not months. And migrating a system that handles billions of dollars in transactions is risky. The IRS tried a modernization effort in the 1990s and it was a disaster — cost over a billion dollars and was eventually scrapped. Luna: So there's institutional fear. But that $4 billion figure — is that just maintenance, or does it include upgrades? Lucas: That's mostly maintenance, plus the cost of the hardware itself. The government rents the mainframes, they don't own them. IBM charges a premium because it's essentially a monopoly for this kind of legacy support. And because the government's procurement process is so slow, by the time a migration plan is approved, the technology has moved on again. It's a classic case of 'if it ain't broke, don't fix it' — except it is broke, just not catastrophically. Luna: Speaking of things that are broke but not catastrophically — has it ever struck you, Lucas, that a show like this, which tries to make sense of this kind of spending, relies on listeners who care enough to support it? We don't run ads, we don't have corporate sponsors. It's just us and the audience. Lucas: It's true. Every episode we dig into how the government spends money, and we're able to do it independently because of listener support. If walking through the economy with us has made something click — or if you've ever used one of our episodes to sound smart at a dinner party — you can support the show at buy me a coffee dot com slash fexingo. Luna: It's a simple way to keep us going. And it means we don't have to answer to anyone but you. So thank you to everyone who already does. Lucas: Alright — back to mainframes. So the question is: what's the path forward? The White House Office of Management and Budget has a 'Cloud Smart' policy that's supposed to encourage agencies to move to modern infrastructure. But the results have been mixed. The Department of Defense, for instance, still runs its main payroll system — called DFAS — on a mainframe that costs over $200 million a year to operate. Luna: Two hundred million? That's more than the entire budget of some federal agencies. Lucas: It is. And that's just one system. The IRS's mainframe for tax processing — the Individual Master File — is from the 1960s. It runs on an IBM z/OS operating system and processes 150 million tax returns a year. The cost to maintain it is about $230 million annually. And here's the kicker: the IRS has tried to replace it three times. Each time, the project ballooned in cost and complexity, and eventually got cancelled. Luna: So there's a paradox. The mainframe is expensive, but replacing it is even more expensive — at least in the short term. Lucas: Exactly. But that calculation misses the opportunity cost. While you're spending $4 billion on mainframes, you're not spending that money on modern systems that could improve service, reduce fraud, or make data accessible. The private sector faced this same choice in the 1990s and early 2000s. Banks, insurance companies, airlines — they all had mainframes. Most of them migrated to client-server and then cloud architectures. They took the upfront hit because the long-term savings were clear. Luna: But the government doesn't have the same incentives. A private company can point to a 5-year ROI and get board approval. A government agency has to go through appropriations cycles, and the Congress that funds the migration may not be the same Congress that sees the savings. Lucas: Precisely. The savings from a migration show up 7 to 10 years out. But the budget is annual. So an agency head who wants to start a modernization project has to cut other programs this year to pay for it, and they may not be around when the benefits materialize. The incentives are misaligned. Luna: And there's another angle — security. These old mainframes weren't designed for the internet. They're patched with security updates, but they're inherently more vulnerable. In 2023, the Government Accountability Office found that several agencies had mainframes that were still running operating systems that no longer receive security patches from IBM. Lucas: Right, because IBM stops supporting certain versions after a while — unless you pay for extended support, which the government often does. That's another line item. But even with patches, the architecture itself is a risk. Modern systems have features like encryption at rest, multi-factor authentication, and audit trails built in. On a mainframe, those are often add-ons, if they exist at all. Luna: So we're paying billions to keep the lights on, and also paying billions more in security risk. Is there any good news? Any agency that's actually managed to migrate? Lucas: There are a few success stories. The Department of Transportation moved its mainframe-based driver's license system to the cloud a few years ago. It took five years and cost about $100 million, but they're saving $20 million a year in maintenance now. That's a 5-year payback. The Centers for Medicare and Medicaid Services — CMS — has also been slowly migrating claims processing off mainframes. But these are exceptions. The vast majority of agencies are still stuck. Luna: It reminds me of the saying: 'The government is the only institution that can spend $4 billion a year on something that doesn't work, and call it budget discipline.' Lucas: That's about right. The thing is, the technology exists to replace these systems. There are companies that specialize in migrating COBOL applications to modern languages. But the government's procurement rules make it almost impossible to hire them quickly. A typical request for proposals takes 18 months. By the time the contract is awarded, the scope has changed. Luna: So we're trapped in a cycle. The mainframes cost too much to maintain, but the process to replace them is too slow and too risky. And meanwhile, the workforce that understands them is retiring. Lucas: Right. The GAO has listed 'modernizing legacy IT' on its high-risk list for over a decade. Every year, a new report comes out with the same recommendations. And every year, the spending continues. It's one of those problems that's not a crisis until it is — and by then, it's too late. Luna: Which raises the question: at what point does it become cheaper to just build a new system from scratch, rather than keep patching the old one? Lucas: That's the billion-dollar question. Literally. For most agencies, the breakeven point is somewhere around 5 to 7 years. But because of the upfront cost and the political risk, they never take the leap. So we keep paying $4 billion a year, indefinitely. And that's not even counting the lost productivity and the opportunity cost of what that money could have bought. Luna: What could $4 billion buy every year? You could build a dozen new schools in every state. You could fund the entire National Endowment for the Arts three times over. Lucas: Or you could spend it on something that makes the government work better for citizens. A modern tax filing system, for instance, where the IRS could pre-fill your return — something that's standard in many other countries. But as long as the mainframe is the backbone, those innovations are off the table. Luna: So the next time you hear about a government IT project that's over budget and behind schedule, remember — the foundation it's built on is probably half a century old. Lucas: And that foundation costs $4 billion a year just to keep from crumbling. For Fexingo, I'm Lucas. Luna: And I'm Luna. Thanks for listening.