May 22, 2026
Today, OpenAI is preparing to make history — and the timing could not be more charged. The company behind ChatGPT is filing a confidential IPO prospectus with the Securities and Exchange Commission as early as Friday, May 22, 2026, according to simultaneous reporting from CNBC, Reuters, and Axios. Working with Goldman Sachs and Morgan Stanley, OpenAI is targeting a September public listing at a valuation somewhere between $852 billion and $1 trillion — which would make it the largest technology IPO ever recorded.
But even as the champagne is being chilled in San Francisco’s Mission District, a cold front is moving in from the East. Chinese AI models now account for more than 60% of all traffic on OpenRouter, the most-used developer AI routing platform in the world. In 2024, that number was just 1%. The speed of that shift — from the fringe to the majority in less than 18 months — is not just a curiosity. It is the single most important threat to every assumption baked into OpenAI’s trillion-dollar valuation.
The IPO That Changes Everything
To understand what is happening today, you first have to understand the race that has broken out across Silicon Valley. Three of the most anticipated public offerings in technology history are now targeting a six-month window. SpaceX filed its public IPO prospectus on May 20, aiming for a June Nasdaq listing under the ticker SPCX at a staggering $1.75 trillion valuation. OpenAI is filing confidentially today for a September debut. And Anthropic — which just projected $10.9 billion in Q2 revenue and its first-ever quarterly operating profit — is eyeing October.
SpaceX goes first and sets the comparable for AI infrastructure economics. OpenAI goes second and stakes the valuation benchmark for frontier AI model companies. Anthropic goes third, armed with superior revenue growth and the first operating profit milestone. Coinbase prediction market data puts OpenAI’s probability of listing before Anthropic at 85%.
A confidential IPO filing is a specific legal instrument. It allows a company to begin the SEC review process without disclosing its financials publicly. Investors typically see a public S-1 about two months after the confidential filing, followed by the actual offering another month later. That cadence points to a public S-1 from OpenAI in late July or early August — with trading beginning in September, just as the summer tech conference circuit wraps up.
At $852 billion to $1 trillion, an OpenAI IPO would exceed Saudi Aramco’s 2019 record of $29.4 billion in proceeds. It would be the defining public market event of the 2020s. And yet, lurking beneath the financial spectacle is a question that no S-1 can fully answer: what happens to the business if cheap AI eats the market from below?
The 60% Problem
Here is the number that keeps AI executives up at night. On OpenRouter — the platform where developers plug in different AI models and route requests across providers — Chinese models went from approximately 1% of usage in 2024 to more than 60% by May 2026. Databricks CEO Ali Ghodsi disclosed the figure this week, and it landed like a grenade.
The models driving that shift are not toys. DeepSeek’s V3.2 runs at $0.28 per million input tokens. Kimi K2.6, which scores an impressive 90.5% on the GPQA Diamond benchmark, costs $0.95 per million tokens. Zhipu’s GLM-5.1 completes AI benchmarking firm Artificial Analysis’s standard 10-evaluation workload for a total cost of $544. For the same workload, Claude costs $4,811. ChatGPT costs $3,357. By price, the cheapest Chinese frontier model is roughly nine times cheaper than Claude and six times cheaper than GPT for equivalent benchmark tasks.
That is not a small difference. That is the difference between a $100 monthly API bill and an $870 monthly API bill for the same developer workload. And it explains, almost entirely, why the OpenRouter numbers look the way they do.
The “Advisor Model” Problem
Databricks’ Ghodsi has a name for the architectural pattern now spreading through enterprise AI deployments: the “advisor model.” The concept is straightforward and economically rational. Companies deploy a cheap, capable open-source or Chinese model as their default layer — handling the 80 to 90% of queries that do not require the absolute frontier of capability. They reserve calls to OpenAI or Anthropic for the tasks that cheaper models cannot reliably solve: complex reasoning chains, sensitive legal or medical analysis, high-stakes code generation.
In this architecture, every frontier model improvement immediately faces challenge from below. OpenAI ships GPT-5.5 and enterprise users reach for it for the hard problems. But the easy problems — the volume — now routes to DeepSeek. The result is that frontier model companies capture the premium segment but progressively lose the base.
Anthropic acknowledged exactly this dynamic in a policy paper released this month, stating candidly that US AI models are only “several months ahead” of Chinese counterparts, and that Beijing is “winning in global adoption on cost.” That is an extraordinary admission from the company that operates the most capable AI cybersecurity model in the world. The US advantage in chip access, TSMC supply, and frontier researcher concentration was supposed to be measured in years, not months.
CNBC published a detailed investigation this week arguing that this dynamic could directly derail both the OpenAI and Anthropic IPOs. The logic is clean: a $1 trillion valuation requires assumptions about sustained pricing power at the frontier for years. If the capability gap to Chinese models closes to within five to ten percentage points in the next 12 months — which Anthropic itself implies is plausible — then the addressable market willing to pay a nine-times premium for the marginal last improvement shrinks rapidly. You cannot build a trillion-dollar public company on a market that is consolidating into a commodity.
What OpenAI Is Betting On
OpenAI’s response to the cheap AI narrative is coherent even if it is not entirely comforting. A person familiar with the company’s thinking told CNBC that every frontier model release, including GPT-5.5, had generated what they called “a vertical wall of enterprise demand” — meaning that each new capability level created net-new use cases that the cheaper tier could not satisfy, expanding the pie rather than merely slicing it thinner.
There is real evidence for this view. OpenAI’s ARR has climbed to approximately $25 billion. Anthropic just projected its first-ever quarterly operating profit on $10.9 billion in revenue in a single quarter. These are not numbers that suggest a business in price-led collapse. Enterprise customers spending more than $1 million per year with Anthropic doubled from 500 to over 1,000 between February and April 2026.
But the OpenRouter data is real too. Developer adoption today is enterprise adoption in 18 months. The engineers choosing DeepSeek and Kimi for their personal projects and internal prototypes in May 2026 are the same engineers who will architect their company’s AI infrastructure in 2027. If those engineers have spent 18 months building workflows on cheap Chinese models, the switching cost to Claude or GPT goes up — not down.
The Race to Define a Category
The deeper strategic play behind the OpenAI IPO is not just raising capital. It is about being first to define what a frontier AI company looks like as a public market asset. Whoever files first sets the valuation comparable. Whoever’s S-1 lands first establishes the financial narrative. By filing today, OpenAI is making a bet that its current momentum — $25 billion ARR, ChatGPT at nearly one billion users, the Sora video platform, the new operator partnerships — tells a strong enough story to set the benchmark high before Anthropic’s superior growth metrics hit the market in October.
It is a calculated move. And if it works, October becomes very difficult for Anthropic, which will be valued against an OpenAI that the public market has already priced. If it does not work — if OpenAI’s S-1 reveals uncomfortable numbers, thin margins, or a compute cost structure that even $25 billion in ARR cannot rationalize — then the entire AI IPO window could face headwinds, and Anthropic’s October listing walks into a storm.
That is the gamble. The biggest tech IPO in history, filed today, in the same week that Chinese AI claimed majority market share among developers. The timing is either brilliant or terrifying, depending on which data you believe.
Probably both.
