OpenAI Is Losing $1.22 For Every Dollar It Makes, And Now It Wants To Cut Prices..

[00:00:00]For every single dollar OpenAI earns in revenue right now, every subscription, every API call, every enterprise contract, it spends $222, negative 122% operating margin. That is the official Q1 2026 number. Not a rumor, not a critic's estimate, their own financials.

[00:00:18]And this week, the Wall Street Journal reported that OpenAI is now considering drastic price cuts. not raising prices to fix the problem, cutting them to get more customers, who will cost them even more money per transaction than the customers they already have. I want to give you the most generous possible interpretation of this strategy. I genuinely tried. I could not find one because here is the math that nobody in the hype cycle wants to say out loud.

[00:00:43]Uber, one of Open AI and Anthropic's biggest enterprise customers, burned through its entire 2026 AI budget by April. Not December, April. Four months into the year. Per engineer monthly API costs were running between $500 and $2,000. Their CTO said the company is quote back to the drawing board on budgeting. JP Morgan analysts published an internal note this month titled AI bills are out of control. Some JP Morgan employees are spending more on AI tools than their own monthly salary. And the Palunteer CEO, who sells AI software for a living, compared what's happening to an addiction, his words, not mine. This is the state of the AI industry in June 2026. Two companies racing toward trillion dollar IPOs. Neither profitable. Both considering cutting prices on a product that already loses money at current prices because their biggest customers are quietly telling them the economics don't work. Today, I want to walk you through exactly what is happening, what the numbers actually say, why the price cuts make the problem worse, not better, and what this all means for the biggest tech IPOs in history that are supposed to happen later this year. Because the story the industry is telling and the story the numbers are telling are two very different stories right now. Let me start with the financial picture because I think most people still don't have a clear sense of just how far these companies are from a working business model. Open AAI Q1 2026 revenue $5.7 billion. Sounds impressive but the operating margin was negative 122%. That means for every dollar of revenue they spent $222.

[00:02:13]They are projected to spend a combined $65 billion on compute and operations across 2026 alongside Anthropic.

[00:02:21]OpenAI's own timeline to profitability 2030, four years away. Current valuation $852 billion. ChatGpt became the first app to reach 1 billion monthly users in May 2026. 1 billion users. Still losing money on every transaction. Anthropic.

[00:02:37]Better picture, but not as good as the headlines suggest. Annualized revenue surged from $9 billion at the end of 2025 to a reported 47 billion by May 2026, a $422% jump in five months. Q2 2026 was their first profitable quarter valuation 9065 billion edging past OpenAI for the first time largely driven by Claude Code's explosive adoption. But here's the thing about that 47 billion annualized figure. It was calculated during the period when enterprises were burning through AI budgets at maximum speed before anyone had checked their bills. Zero Hedge and others have pointed out that Anthropic quietly annualized the revenue from February through May, the peak token maxing months, right before their IPO filing.

[00:03:24]The question is what the number looks like now that enterprises are reading their invoices. The subscription math underneath all of this. Semi analysis, an AI research firm that is explicitly pro AI, not a skeptic, published an analysis showing that if any subscriber at any tier uses more than 25% of their monthly rate limit, OpenAI has a gross margin of negative 25% on that customer.

[00:03:46]If someone uses 50% of their $200 per month ChatGpt Pro subscriptions capacity, the gross margin is negative 775%.

[00:03:55]Not negative 10%. Not negative 50%, negative 775%.

[00:03:59]The 20 per month subscription tier provides access to tokens worth roughly $400 at API rates. The $200 per month tier provides access to tokens worth approximately $8,000 at API rates. These are not products priced to make money.

[00:04:13]These are products priced to build market share with the assumption that something will change before the math catches up. Something hasn't changed and the math is catching up. There's a term that's gone viral inside Silicon Valley boardrooms in the last 60 days. Token maxing. It describes what happened when enterprises got access to powerful AI agents and started using them at maximum volume burning through tokens as fast as possible to drive productivity without a clear way to measure what they were getting back. Uber is the clearest example. Their engineering organization went from 32% clawed code adoption to 84% in a very short window. Monthly API costs per engineer ranged from $500 to $2,000. The entire 2026 AI budget set at the beginning of the year for 12 months was gone by April. Their CTO described the company as back to the drawing board on budgeting. And he said it was difficult to connect AIdriven coding gains to actual product improvements that customers could see. That last sentence is the one that matters because the ROI problem is not just about cost.

[00:05:12]It's about measurement. You can't easily measure what a dollar of AI tokens produces. A task that costs $50 in tokens one day might cost $500 the next depending on how the model responds, how many follow-up prompts are needed, whether the output requires human review and correction. The cost of a task is genuinely variable in a way that makes budget planning almost impossible. The PHOPS Foundation, an organization that helps companies manage cloud and infrastructure spending, said companies started calling in April saying they were already three times over their entire 2026 token budget. Not over by a little, three times over. In response, the Linux Foundation announced plans for a tokconomics foundation, specifically to bring cost discipline to AI spending.

[00:05:52]That organization did not need to exist six months ago. JP Morgan analysts put it plainly in their note. Investors are discussing whether much of the token spend corporate America is incurring is simply wasted. JP Morgan, the bank that co-wrote the SpaceX IPO, not exactly a group of AI skeptics. So, this is the context in which OpenAI is reportedly considering drastic price cuts. Let me walk you through why this is so strange.

[00:06:17]The Wall Street Journal reported, citing people familiar with the discussions, that OpenAI is weighing significant cuts to what it charges for tokens. The move is described as preemptive, anticipating similar cuts from Anthropic. CEO Sam Alman publicly confirmed the direction at a recent event, saying OpenAI would find a lot of ways we can help people get more value for less spend. He called costs a huge issue for business customers. All of that is true. Costs are a huge issue for business customers.

[00:06:46]The question is whether cutting token prices actually solves the problem.

[00:06:49]Here's what the price cut logic assumes.

[00:06:52]One, that customers are using less AI than they want to because the price is too high and cheaper tokens will cause them to use more. Two, that the additional usage from lower prices will generate enough revenue to offset the lower price per token. But here's what the Uber situation and the JP Morgan data actually show. Customers aren't using less than they want to. They used more than they budgeted for. They burned through entire annual budgets in four months. The problem isn't that tokens are too expensive to use. The problem is that enterprises don't know how to control spending on something this variable and can't measure the return on what they're already spending. Cutting prices doesn't solve either of those problems. It might temporarily reduce the bill for the same amount of usage, but it also removes the pricing signal that was forcing enterprises to think carefully about what they were buying.

[00:07:38]Cheaper tokens might actually increase token maxing and increase the operating losses proportionally. If Uber burned its entire annual budget in four months at current prices and you cut the price in half, Uber either burns through a budget twice as fast or uses twice as much and costs OpenAI twice as much to serve. Neither outcome helps OpenAI's path to profitability. And there's the competitive dynamic to consider. If OpenAI cuts prices, Anthropic matches.

[00:08:04]If anthropic matches, Google responds.

[00:08:07]Google has been cutting Gemini API prices aggressively throughout 2026.

[00:08:11]Google builds its own TPUs which gives them structural cost advantages that OpenAI and Anthropic genuinely cannot replicate. Google was processing 3 trillion tokens per day internally by midMay. And underneath all of this, Deepseek and other Chinese open source models already offer frontier quality AI at roughly 113th the cost of what Open AI and anthropic charge. 113th the floor on intelligence pricing keeps falling towards zero. Any margin recovery at OpenAI or Anthropic becomes, as one analysis put it, a math problem with no clean solution. Now, here's where it gets genuinely strange because both of these companies are heading toward IPOs.

[00:08:47]At the same time, they're considering price cuts on their core product.

[00:08:50]Anthropic filed confidentially with the SEC and is reportedly targeting an October 2026 debut. Open AAI filed confidentially the same week. Both companies valued at $852 billion and $965 billion respectively are going to put their financial statements in front of public market investors for the first time. Public market investors who will see the negative 122% operating margin.

[00:09:14]Who will see that the company is spending $222 for every dollar of revenue? Who will see the debt over $1.1 trillion in combined compute commitments between the two companies? And who will now also see that the company is deliberately cutting the price of its core product right before listing. Think about what that signals to a public market investor. You are a company worth nearly a trillion dollars on paper. You just lost money on every dollar of revenue you made. And your strategy to fix the problem is to charge less per unit of the thing you're selling. The counterargument is that you're cutting prices to gain market share and that market share will eventually generate a volume of transactions that makes the math work. But that argument requires believing that enough new customers will appear at lower prices to offset the revenue lost from existing customers.

[00:09:58]When your existing customers are already burning three times their planned budget at current prices and when open source Chinese alternatives exist at 113th the cost the new customers that lower prices attract are the most price sensitive customers in the market. The ones most likely to switch to deepseek the moment the gap narrows further by 2029.

[00:10:18]Combined projections suggest these companies need to be generating roughly $350 billion in annual revenue to justify their compute commitments.

[00:10:26]That's around $10 billion per month each. Anthropic is currently tracking toward $10 billion in Q2 revenue. That is a quarterly number. They need it to be a monthly number in three years and they're doing that by cutting the price of the product. The former investor Mark Andre has been bullish on AI throughout this cycle. Even within the bull case, the arithmetic requires a kind of vertical revenue growth that has no clear historical precedent in technology, let alone in a market where a major competitor is giving away a comparable product for free. There's one more piece of this that I think gets systematically under reportported in the mainstream AI coverage. Deepseek v4, GLM, KI, Chinese open source models available right now through inference providers deliver frontier quality AI performance at roughly 113th the cost of what open AI and anthropic charge. Not 1/ half, not 2/3, 113th.

[00:11:15]Roughly 80% of US AI startups are already using Chinese open source models for significant portions of their training and infrastructure work.

[00:11:23]They're doing this quietly without press releases because the economics are simply better. As open AI and anthropic race toward a price war with each other, the actual floor on AI pricing is being set not by either of them, but by open source models that neither company can easily compete with on cost. Every time OpenAI cuts its token price, it moves closer to a number that open- source inference providers are already offering. The end point of that race is a price that no proprietary model company can sustain. Google understands this. Sundar Pichai argued at Google IO 2026 that enterprises shifting the majority of their workloads to Gemini Flash, Google's cheapest, fastest model, could save over $1 billion annually compared to current enterprise AI spending. Google can price Gemini Flash aggressively because Google builds its own chips and already runs the model at such massive internal scale that the marginal cost per token is structurally lower. Neither Open AAI nor Anthropic has that structural cost advantage. They both depend on Nvidia and thirdparty infrastructure. Their per token costs are not coming down at the pace the pricing war is moving. I want to be honest about this because the bare case is easy to make and I want to give the bull case its best argument before I close. The bull case goes like this. AI capabilities are still improving rapidly. The models being deployed today are significantly more capable than the models from 12 months ago. At some point, the capability improvement creates entirely new categories of value. things enterprises can do with AI that they literally cannot do without it. When you get to that point, the price sensitivity drops dramatically because there's no alternative.

[00:12:55]Anthropic's first profitable quarter in Q2 2026 is a real milestone. Their revenue growth from 9 billion to 47 billion annualized in 5 months is genuinely extraordinary. Claude Code crossed $1 billion in revenue within 6 months of launch. These are not numbers you dismiss. And the structural demand for AI compute continues to grow. The four largest tech companies, Google, Amazon, Meta, Microsoft, are spending a combined $700 billion on AI infrastructure in 2026. They are not doing that because they think the market is going away. The bullcase says the token maxing problem is a growing pain, not a structural failure. That enterprises will get better at controlling AI spend, that the ROI measurement tools will improve, and that the applications with genuine measurable returns will dominate spending while the wasted usage falls away. That's a coherent argument. I take it seriously.

[00:13:46]But it requires trusting that the revenue growth continues at something close to its current pace while costs come down simultaneously and doing both of those things in the window before public market investors start looking at audited financials. That is a very specific and very narrow path. Let me give you the specific things to watch over the next 3 to 6 months because this story is moving fast. Watch the IPO filings when they go public. When Anthropic and OpenAI file their public S1s, the full financial disclosures that become visible to everyone, we will see the actual numbers for the first time, not the curated announcements, the audited financials, Q2 revenue, Q2 margins, the full debt picture, the compute commitments. That is when the story either gets a lot better or a lot worse. Watch whether the price cuts actually happen and how deep they go.

[00:14:31]The journal reported these are still in discussion. If OpenAI cuts token prices by 20%, that's a competitive adjustment.

[00:14:38]If they cut by 50%. That's desperation.

[00:14:41]The size of the cut will tell you more than the cut itself. Watch what enterprises actually do with their Q3 budgets. The token maxing wave hit in February through April. Enterprises are now resetting their AI budgets for the second half of the year. If Q3 enterprise AI spending comes in significantly below Q2, that will show up in Anthropic's revenue figures before the IPO. If it holds or grows, the bullcase gets stronger. Watch DeepSeek and open source. The Chinese AI labs are not standing still. Every quarter that passes with cheaper open source models narrows the window for proprietary model companies to establish pricing power before the floor collapses. And watch the enterprise contracts specifically.

[00:15:19]Anthropic's enterprise model involves minimum annual spend commitments. The companies that signed those commitments in the first half of 2026 are now reading their mid-year usage reports.

[00:15:29]The renewal conversations that happen in Q3 and Q4 will tell you more about the real state of enterprise AI demand than any revenue announcement. Here is what I think this moment in AI actually tells us. The technology is real. The capabilities are real. The demand is real. What is not yet real is a business model that makes money at scale. When you lose $122 for every dollar of revenue, the path to profitability requires either dramatically higher prices, which your customers are already revoling against, or dramatically lower costs, which your infrastructure position doesn't currently allow, or dramatically higher volume, which makes your losses grow before they shrink. And cutting prices when you already lose money on every transaction, doesn't help with any of those three variables.

[00:16:11]OpenAI CEO Sam Alman has said for years that the company would figure out the business model. Sam Alman is a smart person who has built real things, but we'll figure it out is a different thing than a path that the numbers support.

[00:16:24]What the WSJ price cut story actually reveals is that less than 3 months after moving enterprises to token-based billing, the first time many of these companies saw what they were actually paying enough of them pushed back hard enough that the two most valuable AI companies on Earth started discussing how to lower the price of their core product. That is not a sign of a company with pricing power. That is a sign of a company discovering in real time where the ceiling is. The SpaceX IPO this week, the Open AI and anthropic filings, the price war discussions, we are watching the AI industry go from private markets where you can tell any story you want about your trajectory to public markets where the auditors have opinions. That transition is going to be one of the most interesting things to watch in business for the rest of 2026.

[00:17:06]Subscribe and I'll keep tracking it with you. Every major development between now and those IPO filings, I'll be here breaking it down. Drop a comment. Are you in the bull camp or the bear camp on whether these companies can build a real business model in the next three