In partnership with
AI Spotlight · The Spending Paradox
AI Companies Are Spending Like Drunk Billionaires
The biggest AI labs are burning cash at a pace no business in history has matched. Right now, that is great news for users. The part nobody wants to talk about is what happens when the bill comes due.
There is a strange dynamic at the heart of the AI industry right now. The tools keep getting better. The prices keep falling. The features keep expanding. And the companies building them keep losing more money every quarter. This is not a bug. It is a deliberate, calculated strategy. And it is one of the most consequential financial bets in modern technology history.
OpenAI is projected to spend nearly twice what it earns this year, with operating losses expected to reach a record high, spending roughly one dollar and seventy cents for every dollar of revenue it collects. Anthropic is burning through cash at a rate that surprised even its own internal forecasts. The leading AI labs are collectively running some of the largest peacetime cash deficits any private companies have ever operated under, funded by a steady stream of investment capital from the world's largest technology firms and sovereign funds.
This issue explains the two sides of that equation: why the spending war is currently a genuine gift to users, and why it carries risks that most people using AI tools every day are not thinking about.
|
How bad is the burn, really? OpenAI is projected to post a loss of around fourteen billion this year against revenue of roughly thirteen billion, meaning it spends nearly twice what it earns. Its cumulative cash burn through the end of the decade is projected to reach over one hundred billion. Anthropic burned through billions more than its own internal models predicted last year. HSBC analysts have concluded that OpenAI is unlikely to reach profitability before the end of the decade and faces a funding shortfall that would dwarf most companies' entire market caps. |
* * *
The good news first
Why the spending war is the best thing that ever happened to AI users
When companies compete for market share by spending more than they earn, the people who benefit most are the customers. Every dollar the AI labs lose on serving your queries is effectively a subsidy on the cost of intelligence. The tools you use every day, the chatbots, the coding assistants, the writing tools, the research agents, are dramatically underpriced relative to what they actually cost to run.
OpenAI's inference costs, the actual cost of running the model when you type a message, quadrupled in a single year as usage scaled and models grew more capable. Yet the price users pay either stayed flat or fell. That gap between what it costs to serve you and what you are actually charged is the spending war in action. Labs are deliberately absorbing that gap to grow user bases, lock in habits, and demonstrate the value of their platforms to investors and enterprise clients who pay much more.
|
What users are getting right now because of the burn Subsidised intelligence: The actual compute cost of your queries is far higher than your monthly subscription fee. Labs are eating that difference to keep you in the ecosystem. Accelerated capability: Because labs are spending at scale rather than pacing investment to revenue, model improvements are arriving years faster than they would in a profitable, steady-state business. Intense competition: Multiple well-funded labs competing for your attention means no single company can get away with slowing down, cutting features, or dramatically hiking prices without losing users to a rival. Free tiers that would not otherwise exist: The free versions of ChatGPT, Claude, Gemini, and others exist primarily to grow user bases for investor metrics and enterprise upsell pipelines. Users get genuinely useful tools at no cost in exchange for being part of that growth story. |
The honest framing is that you are not just using an AI product. You are a beneficiary of one of the largest investor-funded technology subsidies in history. The question is how long it lasts and what changes when it ends.
Build the AI skills senior engineers need to get ahead
AI is already handling large parts of execution. That shift is not coming later. It is happening now.
What is left, and becoming more valuable, is the ability to design systems, apply AI thoughtfully, and own outcomes in production. That is the work strong teams expect from senior engineers in 2026.
Gauntlet is built for engineers who want to operate at that level. In a single week, 15 hiring partners conducted 246 interviews with challengers onsite in Austin. Apply now.
Must be a US citizen to qualify.
Now the risk
Why the same spending war is a long-term risk nobody is talking about clearly
The strategy every major AI lab is running right now has a name in venture capital: blitzscaling. Grow as fast as possible, ignore near-term profitability, and bet that scale and market dominance will eventually produce returns that justify the losses. It worked for Amazon, which lost money for years before its cloud and logistics businesses turned the model profitable. The AI labs are running the same playbook, but with burn rates that dwarf anything Amazon ever ran, and timelines to profitability that keep getting revised further into the future.
OpenAI's own internal documents show operating losses expected to grow substantially through at least the late 2020s before the company pivots to what it projects will be meaningful profit. Anthropic expects to reduce its cash burn significantly faster, targeting near breakeven by the late 2020s. But both timelines depend on assumptions about user growth, enterprise adoption, and model cost reductions that are not guaranteed.
|
Five risks inside the spending war Risk 1 — The profitability gap never closes: Inference costs are structural, not temporary. Every time a more capable model launches, it costs more to run. Revenue growth has so far not kept pace with cost growth, and there is no guarantee it ever will. Risk 2 — Investor patience runs out: The capital flowing into AI labs comes from investors who expect returns. If growth slows, if a major model fails, or if the broader market cools on AI as a category, the funding pipeline could tighten quickly, forcing cost-cutting that hits product quality and availability directly. Risk 3 — The price floor breaks: When AI labs eventually need to move toward profitability, prices will rise. Free tiers will shrink. Rate limits will tighten. The tools that feel abundant today will start to feel metered. Users who have built workflows and habits around current pricing will face a reckoning. Risk 4 — Consolidation kills competition: If one or two labs survive the spending war and others collapse or get absorbed, the competition that currently keeps prices low and features high disappears. The survivors will have enormous pricing power and no competitive pressure to use it gently. Risk 5 — Infrastructure concentration creates systemic fragility: As AI labs consolidate around a handful of cloud providers for compute, single points of failure grow. Outages, geopolitical disruption, or regulatory action targeting one provider could ripple across the entire AI tool ecosystem at once. |
* * *
Two very different bets
OpenAI and Anthropic are taking opposite paths to survival
Not all AI labs are running the same strategy. The contrast between OpenAI and Anthropic is particularly revealing because it shows two fundamentally different theories of how to survive the spending war.
| Dimension | OpenAI | Anthropic |
|---|---|---|
| Burn strategy | Maximum spend, bet on dominance and AGI timeline | Narrower spend, target faster path to breakeven |
| Revenue focus | Consumer scale plus enterprise, massive infrastructure commitments | Primarily enterprise, tighter operational discipline |
| Profitability target | Meaningful profit projected around late decade, losses growing until then | Burn rate falling sharply year over year, near breakeven targeted sooner |
| Risk profile | High variance: enormous upside if AGI bet lands, catastrophic downside if markets cool | Lower variance: slower upside but more defensible path to sustainability |
For users, the difference matters because it determines which company's products remain available and at what price if the investment climate changes. Anthropic's model is more conservative but may be more durable. OpenAI's model offers the greatest potential rewards but requires near-perfect execution on a timeline that stretches far into the future.
* * *
What to do with this
How to build your AI stack knowing the subsidy eventually ends
The practical implication of all of this is not that you should stop using AI tools. The tools are genuinely good and the value they deliver is real regardless of the financial dynamics behind them. The implication is that you should build your workflows with the assumption that pricing, availability, and access will change, possibly sharply, within the next two to three years.
|
Five principles for building in a subsidised-then-priced market Do not get locked into one provider: Build workflows that can switch models without rebuilding everything from scratch. Vendor lock-in is costless today and extremely expensive if pricing shifts. Use the cheap era to build durable assets: The time when AI is underpriced is the time to create knowledge bases, train custom prompts, build automations, and document your own workflows. Those assets retain value even if the underlying tool changes. Know which tools you actually depend on: There is a difference between tools you use casually and tools that your income or productivity genuinely depends on. The second category deserves more scrutiny of the company's financial sustainability. Watch the enterprise pricing signals: What AI companies charge enterprises is closer to what they actually need to charge everyone to survive. When enterprise pricing rises or new pricing tiers appear, that is a signal about where consumer pricing is eventually heading. Treat open source as your floor: Open-source models are improving rapidly and are free from the financial pressures that drive pricing decisions at proprietary labs. Knowing what is achievable with open models gives you a real alternative if commercial pricing becomes untenable. |
* * *
The honest big picture
The AI spending war is one of the strangest economic experiments in modern history. The biggest, best-funded technology companies in the world are writing blank cheques to train and run models that currently produce less revenue than they cost to operate, betting that the technology will become so essential and so economically valuable that the losses will look trivial in retrospect.
That bet might be right. The historical precedents are not comforting, because nothing at this scale has been tried before. What is clear is that the dynamics cannot hold indefinitely. At some point, the industry will move from growth-at-all-costs to profitability-at-all-costs. The users who are best positioned for that transition are the ones who understood it was coming and built accordingly.
|
You are not just using an AI product. You are a beneficiary of the largest investor-funded technology subsidy in history. The smart move is to build as much as possible while the subsidy lasts, and to build in a way that survives when it ends. |
* * *
|
Before you go When AI pricing eventually rises, what tool or workflow in your stack would you be most unwilling to give up? Hit reply and tell me. The most interesting answers will shape a follow-up issue on how to recession-proof your AI stack before the subsidy era ends. |
Until next time,
AI Spotlight
Practical AI, translated into real work, once a week.