They say there's no such thing as a free lunch. But in 2020 we came close: you could order a delivered pizza on DoorDash for $16. DoorDash promptly paid the restaurant $24 for it. The difference came out of venture capital, and nobody seemed to mind. Least of all you, the person eating the pizza.
This is the oldest pattern in tech: subsidize the product until it becomes the habit, then charge what it actually costs. The surprising part isn't that the subsidy ends. It's how often it works.
The subsidy buys time. The roadmap earns the right to charge.
The bet is always the same. The product isn't ready yet, or the market doesn't know it needs it yet, so you underwrite adoption now and build something worth paying for later.
Uber spent billions on below-cost fares to make ride-hailing ubiquitous. Since 2018, prices have climbed roughly 18% per year. Most riders (including me) are still riding. The subsidy bought time to build a dense two-sided marketplace that outperformed what came before.
This works in B2B as well. Google launched Google Apps for Your Domain in 2006 as a free offering: Gmail, Docs, Calendar under your own business domain. It stayed free for sixteen years. During that time, Google built it into a comprehensive productivity platform, eventually rebranding it as Google Workspace. In 2022, legacy free accounts transitioned to paid plans. Most organizations stayed, because the product had earned its place as core infrastructure.
It also works in the Enterprise. MuleSoft started as an open-source integration engine and became the de facto enterprise service bus. As the platform matured, the company developed Anypoint Platform, a commercially licensed product with the management, governance, and design capabilities enterprises needed at scale. Salesforce acquired MuleSoft for $6.5 billion in 2018. The open-source project remains available. The commercial platform is where the value compounded over time.
In each case, you lose some users on the margin who were only there because it was free. But the subsidized runway did its job.
Now it's AI's turn. And it's coming in hot.
Like many, I've been building agents in OpenClaw using Claude's subscription plan. And, like many, I've been getting a remarkable deal: workloads priced at 20–30x below equivalent API rates. By one estimate, a single $200 Max subscription was delivering $5,000 to $7,500 worth of monthly inference through third-party tools. I enjoyed it while it lasted, but clearly that math was never going to hold. Anthropic's recent move to meter third-party tool access is the correction arriving in real time. No more free lunch.
Every subsidized product has to find equilibrium, the point where what you charge covers what it costs. How fast you get there depends on what's underneath.
SaaS companies could run a slow burn. Building the product might have been expensive (top tier engineer salaries all around), but marginal costs at scale were negligible. Google ran its free tier for sixteen years. MuleSoft's open-source adoption ran for a decade before the commercial transition was complete.
Marketplace businesses that move atoms, like Uber and DoorDash, had real per-unit costs from day one, so the runway was shorter. But these models also exploited idle capacity: people's free time, cars already sitting in driveways, restaurants already itching for new customers. The infrastructure already existed. The platform just activated it.
AI has no such slack in the system. Scaling inference requires new capacity. Potentially $5 trillion in the next 5 years! Buying land, pouring concrete, standing up data centers, securing power, installing cooling, and landing GPUs. These are capital expenditures measured in years, against demand compounding in months. There is no idle GPU sitting in someone's driveway. The subsidy window is compressing fast.
The repricing is inevitable. The rollout is under your control.
If the correction has to happen, and it does, then the only question is how.
When Anthropic announced dedicated credits for programmatic Claude usage in May 2026, the economics were defensible. Agentic workloads were consuming 20–30x their subscription price in compute. No business sustains that. But the announcement framed metered credits as a bonus. Developers immediately recognized it as a reduction. The backlash was harsh.
Anthropic is destroying its developer ecosystem with changes like this which tells their subscribers they can't use the best apps out there on Anthropic's models
— Kun Chen
I love the models Anthropic are offering, but I seriously hope it's a mistake that they're blocking alternative harness providers. Seems very customer hostile.
— DHH
Until we see significant change, it is safe to assume any statement from an Anthropic employee is a lie on a timer.
— Theo Browne
I'll grant you that you can't make all of the people happy all of the time. But this reaction was exceedingly predictable.
The economics were right. The framing was wrong. And when your own users are publicly calling the move "customer hostile" and declaring that your statements are "a lie on a timer," the destruction of trust is a big counterweight against the marginal cost you just saved.
Treat your customers like adults. You won't regret it.
There is nothing structurally novel about this shift, but the pace is unprecedented and this won't be the last example we see. The AI space is rife with subsidization, and a flood of cheap compute is nowhere on the horizon. I can empathize with the pros at Anthropic and chose that positioning. The temptation to sugarcoat and spin is natural. Trust me, I've felt it, I've done it. To err is human. Giving in to this temptation feels like the surest way to win the day, and sometimes it is. But it's not the way to build trust that lasts.
My advice: resist the temptation. Be honest with your customers. Instead of trying to tell them their lower quotas and higher bills are a great new benefit… let them wear the free lunch they got as a badge of honor. For my part I appreciated all of the cheap pizzas and subsidized rides in the last decade, and I also have appreciated burning a heap of subsidized tokens so far in this decade.
So acknowledge the shift. Explain the why. Skip the spin. Recognize that some customers will be worse off. Don't paint them as people abusing the system. Let them stand proud as the ones who got in early and got a free lunch out of the deal.
In my professional capacity I've managed through dozens of these pricing changes. For my part I've never regretted treating customers like the smart, discerning people they are. To all the marketing and communications pros in the industry facing this decision right now, I encourage you to do the same.
Researched in Perplexity · Drafted with Claude Opus 4.6 · Hand crafted in Google Docs · Header image generated in Google Imagen 3