OpenAI price cuts turn AI IPO race into margin test

OpenAI’s reported plan to cut token prices as competition with Anthropic intensifies would turn the artificial intelligence IPO race from a valuation contest into an earlier test of pricing power.

Coming just after OpenAI confidentially filed for a US listing, and a week after Anthropic made the same move, the reported cuts would give investors a sharper question than who lists first, according to Reuters reporting on the filings. Anthropic may get the first chance to set the benchmark for frontier-AI revenue. OpenAI, meanwhile, has to explain why lower token prices would widen adoption without compressing already contested margins.

Enterprise buyers hear a different signal. For chief financial officers and technology chiefs, cheaper tokens are not a threat to the AI story. They are the point. CNBC’s analysis of model routing, where companies shift tasks to lower-cost models when premium systems are unnecessary, suggests corporate customers are already pushing the market toward price discipline rather than accepting every frontier-lab bill as strategic inevitability.

Behind that split is the real tension in the listing race. Private investors rewarded OpenAI and Anthropic for scale, technical leadership and scarce exposure to generative AI. Public investors will ask whether tokens can keep premium pricing once rivals copy capabilities, buyers route around the expensive models and compute bills remain large.

Pricing power becomes the story

According to Reuters, the Wall Street Journal reported that OpenAI is considering drastic price cuts because it expects Anthropic to make similar reductions. CNBC said the consumer tiers cited in the report include OpenAI subscriptions around $8, $20 and more than $100 a month, compared with Anthropic Claude plans at about $17 and more than $100 a month.

Sticker prices can look trivial beside the headline valuations. In this case, they are probably the numbers that matter most. AI companies sell access in tokens, the metered units of text processed by their models, so the price of a token is the closest thing the sector has to a visible unit-economics signal. A pre-IPO OpenAI discount would look less like a consumer promotion and more like evidence that the marginal cost of demand is falling faster than the marginal price customers are willing to pay.

Unit economics make the distinction hard to avoid. A software company that cuts prices to accelerate adoption can still defend high gross margins if distribution costs are low. A frontier-AI company cutting token prices while paying for training, inference and data-centre capacity has to prove efficiency gains are large enough to offset the discount. Model performance, chip utilisation and routing economics move into the IPO story from day one.

On that reading, CNBC’s token-economy explainer was less background than warning label. Wall Street will need to compare token volume, inference costs and model mix before it can value OpenAI or Anthropic like ordinary software companies. Revenue growth alone will not settle the question if cheaper tokens simply pull forward usage at lower margins.

Anthropic gets the first price marker

Filing sequence matters because Anthropic may get to define the valuation language. Reuters reported that Anthropic confidentially filed for a US IPO before OpenAI and was valued at $965 billion in late May, while OpenAI’s March valuation stood at $852 billion.

In a market normally framed around OpenAI, that gives Anthropic a rare first-mover role. Kat Liu, an analyst at IPOX, told Reuters that timing itself was part of the opportunity.

Filing shortly after SpaceX allows Anthropic to capitalize on strong investor interest in AI and growth stocks while the window remains favorable.
Kat Liu, IPOX, via Reuters

Liu’s line captures the bullish view: investor appetite is still there, and Anthropic can ride it before fatigue sets in. The same window also raises the bar for disclosure. Public investors are likely to ask how much of Anthropic’s $47 billion annual run-rate revenue cited by CNBC is tied to durable enterprise workloads, how much comes from discounted growth and how quickly price competition could change the answer.

For OpenAI, the position is subtler. Harrison Rolfes of PitchBook told Reuters that OpenAI can watch the market’s reaction to Anthropic’s audited numbers before setting its own terms.

OpenAI now has a free option to watch how institutional investors react to audited frontier AI financials before committing to its own price.
Harrison Rolfes, PitchBook, via Reuters

The advantage is useful, but defensive. A strong Anthropic trade would let OpenAI argue the category has cleared its first public test. A stumble on cash burn or gross margin would force OpenAI to show that its larger brand and product reach translate into better economics, not just a larger addressable market.

The buyer is pushing back

Buyer pushback is where the IPO narrative gets less clean. Enterprise AI spending has moved from experimentation to budget review, and the next phase may reward vendors that let customers spend less per task, not more.

Companies are looking for ways to direct simpler work to cheaper models while reserving premium systems for harder tasks, CNBC’s model-routing analysis said. Semafor separately reported that companies are struggling to measure AI’s return on investment, a sign that boards are asking for proof after two years of pilots and platform deals.

For OpenAI and Anthropic, that changes the pricing conversation. A corporate customer that can send routine summarisation, extraction or coding support to a lower-cost model will not pay frontier prices for every request. Total demand may still rise, but usage would be spread across cheaper systems.

Reported cuts at OpenAI would therefore be a market signal. They would suggest the company expects unit prices to fall and wants to shape the decline rather than react to it. Anthropic would then have to defend its own pricing while courting public investors who may prefer revenue quality to raw growth.

Cloud computing offers a useful analogy, though only up to a point. Amazon, Microsoft and Google trained investors to accept periodic price cuts because scale drove down infrastructure costs and customer workloads deepened over time. Frontier AI is trying to make a similar argument before the economics are as seasoned, and with competitors still spending aggressively on model training.

Capital is crowding the window

Calendar pressure adds another risk. Reuters said OpenAI could reach public markets as early as September. Bloomberg has argued that OpenAI and Anthropic listings would test AI enthusiasm, while SpaceX’s own offering may compete for the same growth-stock capital.

Fund managers do not evaluate mega-listings in isolation. Buying SpaceX, Anthropic and OpenAI in the same window would amount to one large allocation to scarce, high-growth private technology finally entering public markets. If the first deal demands a valuation built on years of perfect execution, the second and third deals inherit that scepticism.

Wired’s analysis of overlapping AI investors framed the rivalry more like Pepsi and Coke than a winner-take-all fight, with backers willing to own both OpenAI and Anthropic. That may work in venture portfolios. Public markets are less forgiving about duplicated exposure when the companies share the same broad customer budgets, infrastructure bottlenecks and regulatory risks.

Reuters cited a cautious OpenAI statement on timing.

It may be a while because there are things we want to do that are likely easier as a private company.
OpenAI, via Reuters

Viewed that way, the caution looks rational. Remaining private lets OpenAI adjust pricing, product bundles and enterprise contracts without explaining every margin effect in quarterly filings. Going public would give employees and investors liquidity, but it would also force a cleaner answer on whether lower token prices expand the market faster than they dilute economics.

What public investors will ask

Gross margin comes first. Not the adjusted version, and not a long-term target in a slide deck. Investors will want to know how much it costs to serve a dollar of inference revenue after model mix, usage intensity and compute commitments.

Customer concentration follows. If large enterprise and developer accounts carry the economics, token cuts could be aimed at defending strategic customers rather than winning casual users. That would send a different signal from a broad consumer price war.

Substitution is next. If cheaper models can handle more corporate tasks, the premium frontier layer has to prove it is still scarce. Anthropic’s Claude, OpenAI’s GPT systems and Google’s Gemini will be compared not by benchmark headlines but by cost per useful output.

Capital-market appetite is the last test. A $965 billion Anthropic valuation and an $852 billion OpenAI valuation leave little room for public buyers to discover that the sector’s pricing is already bending downward. Price cuts can be strategic. They can also be the first evidence that competition is arriving before operating leverage.

A simple version of the AI IPO story says two frontier labs are racing to meet investor demand. The harder version is that they are racing because demand, pricing and margins may never look as generous again. OpenAI’s reported cuts put that harder version in front of the market early.