GitHub’s Switch to Usage Pricing Reignites the AI Economics Debate — And Ed Zitron Is Back With Another Bubble Warning

GitHub’s quiet but seismic shift from flat-rate subscriptions to usage-based pricing for Copilot has done what no earnings call or hype cycle could: it has forced the AI industry to confront the uncomfortable math behind the curtain. The move, which began rolling out in late 2025 and accelerated this spring, has triggered a fresh wave of scrutiny — and no one has seized the moment more enthusiastically than longtime AI skeptic Ed Zitron.

GitHub’s Switch to Usage Pricing Reignites the AI Economics Debate — And Ed Zitron Is Back With Another Bubble Warning In a lengthy, characteristically fiery essay published April 28, 2026 on his site Where’s Your Ed At?, Zitron argues that the entire generative-AI business model is fundamentally broken.

“Subscriptions were always a subsidy scam,” he writes, designed to hide the true, eye-watering cost of inference while hooking users on seemingly cheap access before the bill comes due.

The numbers he cites are brutal. Back in October 2023, The Wall Street Journal reported that Microsoft was losing more than $20 per user per month on GitHub Copilot on average — with some heavy users costing the company as much as $80 a month — even while charging just $10.

Three years later, the underlying economics haven’t magically improved; they’ve simply been restructured.

Users are now seeing their Copilot bills spike whenever they actually use the tool heavily, and the complaints have been loud.

Zitron then turns to leaked OpenAI internal projections that paint an even darker picture. The company reportedly expects an 80 % collapse in ChatGPT Plus subscriptions (the $20-a-month tier) this year — dropping from roughly 44 million to just 9 million. The plan? Make up the difference with a much cheaper “Go” tier targeted at price-sensitive emerging markets like India and Brazil, and grow that segment 36-fold. Zitron calls the forecast “unrealistic at best.”

GitHub’s Switch to Usage Pricing Reignites the AI Economics Debate — And Ed Zitron Is Back With Another Bubble Warning But the real heart of his critique is infrastructure. He dives deep into the Stargate project — the massive data-center joint venture between OpenAI, Microsoft, Oracle, and others — pointing out that only a tiny fraction of the announced capacity has actually broken ground.

Announced plans for 114 GW of AI data centers have just 15.2 GW under construction, and even that is running years behind schedule. Zitron notes that while companies have pledged hundreds of billions in capex, actual revenue to service the debt remains negligible. “They’re building castles on quicksand,” he argues, “and the tide is coming in.”

He also hammers the GPU amortization trap. NVIDIA’s relentless annual hardware refreshes mean that even a six-year depreciation schedule for servers is optimistic — by the time the hardware is fully paid off, it’s already obsolete. Electricity, cooling, and land costs only compound the problem.

Here Zitron’s timeline gets a little ahead of itself. The most popular current workhorse, the H100, is only about three years old, and the previous-generation A100 — now six years old — remains surprisingly popular and in demand. There is still no visible collapse in GPU orders; if anything, hyperscalers are still scrambling to secure supply. Demand, for now, has not cratered.

None of this is to say the AI industry’s economics are pristine. Inference remains expensive, margins are razor-thin or negative for many consumer-facing products, and the gap between promised trillion-dollar revenues and today’s reality is enormous. Zitron is right that many companies are still burning cash at an alarming rate while betting on hockey-stick growth that has yet to materialize.

GitHub’s Switch to Usage Pricing Reignites the AI Economics Debate — And Ed Zitron Is Back With Another Bubble Warning What feels familiar — and slightly wearying — is the rhetorical reflex. Every time a hot new technology attracts massive investment and becomes culturally ubiquitous, the same “this is a bubble” narrative reappears.

Cellular phones were once dismissed as an expensive fad that would never achieve widespread adoption.

The internet itself was called a speculative bubble in the late ’90s. Cloud computing, streaming video, and even electric vehicles all faced similar doomsaying.

The difference this time is that the capital at stake is orders of magnitude larger, the energy demands are unprecedented, and the technology is evolving at breakneck speed.

GitHub’s pivot to usage pricing is not the beginning of the end — it’s simply the industry admitting that the old subscription math no longer works when every heavy user can burn through millions of tokens in a single afternoon.

Whether the AI gold rush ultimately proves sustainable or collapses under its own weight remains an open question. What is clear is that the era of “free-ish” AI for everyone is over. The real economics are finally coming into focus — and they are, as Ed Zitron loves to remind us, not pretty. But history suggests that ugly economics don’t always mean the end of a revolution. They just mean the honeymoon is finished, and the real work of building a profitable business has begun.