Anthropic, the Pentagon and the myth of the coming “SaaSpocalypse”
When an AI unicorn walks away from a $200 million Pentagon deal, it’s not just another contract dispute. It’s a stress test of the entire AI industry’s ethics, business models and dependence on government money. Add in fears of a looming “SaaSpocalypse” and record valuations for defense startups, and you get a snapshot of a market trying to decide what kind of AI future it actually wants. In this piece we’ll unpack what really happened between Anthropic and the U.S. Department of Defense, why OpenAI is now caught in the backlash, and what all of this means for SaaS founders on both sides of the Atlantic.
The news in brief
According to TechCrunch’s Equity podcast, the U.S. Department of Defense (DoD) labeled Anthropic a “supply‑chain risk” after negotiations over a major AI contract broke down. The key dispute reportedly centered on how much control the Pentagon should have over Anthropic’s models, especially for potential use in autonomous weapons and large‑scale domestic surveillance.
Anthropic walked away from a contract worth around $200 million, and the DoD instead turned to OpenAI, which accepted the work. Shortly after the military deal became public, ChatGPT uninstalls reportedly jumped by 295%, suggesting a consumer backlash against the company’s closer alignment with the U.S. military.
The same Equity episode also highlighted several other shifts in the tech landscape: defense startup Anduril is said to be raising at a $60 billion valuation, MyFitnessPal has acquired teen‑built calorie‑tracking app Cal AI, and there’s growing debate over whether we’re heading into a “SaaSpocalypse” as AI destabilizes traditional software‑as‑a‑service business models.
Why this matters
Anthropic’s standoff with the Pentagon is a rare moment where an AI company puts governance ahead of revenue—and pays for it. Walking away from $200 million in guaranteed income in a capital‑intensive sector is not a symbolic gesture; it’s a strategic bet that brand trust and control over deployment matter more than short‑term ARR.
Winners and losers? In the short run, OpenAI wins the contract but risks its reputation. Taking the deal signals to governments that OpenAI is willing to integrate deeply into defense systems. For enterprises and consumers already uneasy about concentrated AI power, that might be the tipping point. The surge in ChatGPT uninstalls is less about lost subscription revenue and more about a narrative shift: from helpful assistant to quasi‑military infrastructure.
Anthropic, meanwhile, loses revenue but gains a clear ethical positioning. Whether you agree with its red lines or not, investors and large customers now know where those lines are. In a market flooded with interchangeable AI wrappers, that kind of clarity is differentiation.
For startups, the signal is mixed. Federal money looks tempting, but it comes with policy capture risk: once defense becomes your biggest customer, your product roadmap and safety posture inevitably bend toward its needs. On the other side, refusing that money may mean slower growth—yet it can preserve optionality for regulated industries, international customers and future EU compliance.
The deeper implication is that AI governance is becoming a commercial variable, not just a PR talking point. How you answer “who gets to use our model, and for what?” is now as core to your go‑to‑market as pricing or uptime.
The bigger picture
This Pentagon–Anthropic episode plugs into several broader currents the Equity team also touched on.
First, the militarisation of AI. Anduril, reportedly raising at a $60 billion valuation, epitomises a new class of defense‑first tech companies. Their pitch: move fast, break the old acquisition system, and put AI into the field before slower incumbents can react. When one AI lab refuses a weapons‑adjacent contract, another—whether OpenAI, Anduril or a classified in‑house project—will usually step in. That dynamic creates what you might call ethics arbitrage: governments shop around until they find a vendor whose safety threshold is low enough.
Second, the SaaSpocalypse narrative. As AI systems handle more of what SaaS used to do—CRM data entry, analytics, support, content creation—investors are understandably asking whether traditional SaaS multiples are sustainable. The MyFitnessPal–Cal AI deal is a neat micro‑example: a legacy fitness app bought a viral, AI‑native calorie tracker built by teenagers. That’s not the end of SaaS, but it is a sign that AI‑first challengers are forcing incumbents to buy rather than build.
Third, capital is flocking to defensible moats. Pinterest raising and spending $1 billion on AI, only to deploy much of it via share buybacks, hints at the anxiety inside ad‑driven platforms: they know they must invest heavily in AI just to stand still. Defense AI, core model labs and infra providers look comparatively safer because their moats are rooted in data access, regulatory entanglement and long‑term contracts, not just product UX.
Taken together, the message is clear: we’re not in a generalized tech winter so much as a violent repricing of what counts as durable. AI labs that embed into state power structures and critical infrastructure will look extremely valuable to some investors—and deeply concerning to many citizens.
The European and regional angle
For European readers, this is more than an American drama. It foreshadows the collision between Brussels‑style AI regulation and Washington‑style defense pragmatism.
The EU’s upcoming AI Act places strict limits on biometric surveillance and high‑risk uses of AI, while debate around lethal autonomous weapons remains heated in European parliaments. If U.S. labs become deeply integrated with the Pentagon—including for surveillance and targeting—European governments and regulators will have to ask: can these same models be legally and politically deployed in the EU?
We’re already seeing European alternatives emerge. Startups like Mistral AI (France), Aleph Alpha and defense‑focused Helsing (Germany) position themselves explicitly around European values, data residency and sovereignty. For them, Anthropic’s refusal of the Pentagon deal is a gift: it validates the idea that there is market demand for value‑aligned AI, not just the most capable model at any cost.
For European SaaS founders, the “SaaSpocalypse” talk should be interpreted less as doom and more as a forcing function. Between the AI Act, GDPR and the Digital Services Act, EU‑based SaaS already operates under tighter constraints. The upside is that products built to clear EU bars on privacy, transparency and safety can become export‑grade as other regions tighten rules.
In other words: while U.S. AI giants negotiate weapons clauses, European companies have an opening to differentiate on trust, compliance and openness, especially in sectors like healthcare, education and public services.
Looking ahead
Expect the Anthropic–Pentagon split to become a reference case in every future negotiation between AI labs and governments. Security agencies will push harder for model weights access, fine‑tuning rights and priority usage, arguing that national security trumps commercial concerns. Some labs will agree; others will carve out red lines similar to Anthropic’s.
We should also expect more consumer blowback when high‑profile AI companies sign defense deals. The 295% spike in ChatGPT uninstalls may or may not have a lasting revenue impact, but it sends a clear signal to marketing teams: military contracts are reputationally expensive, especially outside the U.S.
On the SaaS side, the most likely future is not a sudden “SaaSpocalypse” but an extended SaaS mid‑life crisis. Products that merely wrap existing AI APIs with thin UX will struggle; products that own distribution, workflows or regulated data will endure. Consolidation, like MyFitnessPal snapping up Cal AI, will accelerate as incumbents buy AI‑native challengers rather than risk irrelevance.
For European policymakers, the next 2–3 years will be about implementation: turning the principles of the AI Act and related legislation into technical standards, audits and enforcement. If they move too slowly, foreign defense‑aligned AI platforms may become de facto standards. If they move well, Europe could define what “responsible AI SaaS” looks like globally.
The open questions: Will any major lab publish a binding, externally auditable policy on military use? Will users actually switch en masse to “non‑military” AI alternatives, or is the uninstall spike a short‑lived protest? And can SaaS founders convince investors that AI is a tailwind, not a death sentence, for their recurring‑revenue dreams?
The bottom line
The Anthropic–Pentagon breakdown is not just a contract story; it’s a fault line in how we want AI embedded in power. OpenAI’s decision to take the deal, the rise of defense giants like Anduril and the hand‑wringing over a “SaaSpocalypse” all point to the same reality: AI is moving from experimentation into institutions that don’t easily let go. The real question for founders, regulators and users—especially in Europe—is simple: who do you trust to hard‑wire their values into the systems that will quietly run your world?
