Does the ChatGPT Health precedent read across to banks? Whether community skepticism reaches OpenAI's finance-vertical ARR

Two camps watched OpenAI push the same money-drawing quarter into the rails of regulated banking. Skeptics saw a ChatGPT Health repeat — consumer distrust over handing over private records that stalls sales to institutions that answer to regulators. Optimists saw noise, and beneath it, procurement teams that buy on contracts and compliance, not Reddit threads. Both cannot be right about what actually slows the finance business. The fight is over the transmission channel: does community doubt reach bank procurement, or does compliance-driven buying simply ride over it?

The distinction matters because OpenAI is heading for a public listing the CFO says will happen in 2027, and the market is being asked to price a vertical — finance — whose revenue is not disclosed at all.

Shared facts

As of September 2026, OpenAI is a private company worth roughly $852 billion at its last secondary sale, with a confidential S-1 filed in June and a targeted 2027 debut that CFO Sarah Friar said could come sooner "if our business continues to inflect." Its run rate recently crossed $40 billion, with enterprise more than half of revenue and growing about 50% quarter-to-date versus 35% overall. Against that stands a projected 2027 cash burn near $63 billion and no projected cash-flow positivity until 2030.

What both camps are reading from the same pages:

• ChatGPT Health (launched January 7, 2026).A consumer feature that connects medical records to the chatbot — explicitly not HIPAA-compliant — and drew documented community distrust from its first day, with privacy warnings in mainstream outlets and even a skeptical Lifehacker "I Don't Trust It" headline on rollout.
• The finance push.A personal-finance experience that lets U.S. users connect bank accounts (a May preview, broadly released in June, powered by Plaid), plus a spring Financial Services Summit and a reported consumer-finance acquisition.
• Named, paying institutions. Morgan Stanley, OpenAI's original strategic financial client, runs an internal advisor tool used daily by over 98% of advisor teams under enterprise terms including zero data retention. Citi pays for OpenAI software to read legal documents, approve account openings, and send trade invoices. Neither paused for the ChatGPT Health backlash.
• Third-party figures on direction, not sentiment. Market share data from the AI-intelligence firm Evident shows OpenAI's share of disclosed bank AI use cases fell from roughly half eighteen months ago to about one-third by the end of 2025, with banks moving to Anthropic, Google, and in-house models.
• Independent finance benchmarks. Box's enterprise eval put one OpenAI model at 76% accuracy on complex financial documents versus 71% for its predecessor. A DualEntry accounting benchmark found even the leading models "still struggle to reach enterprise-grade financial accuracy". An Excel-modelling test rated the top model at 73.7% and, even at that, "not client-ready deliverables".

One absence is itself a fact: OpenAI does not disclose a finance-vertical ARR. The number in the title of every part of this debate is a hope attached to a private company, not a reported figure.

Round one — does community doubt reach bank procurement?

The bull's cleanest evidence is the falsification hiding in plain sight. This quarter's named bank customers exist. Morgan Stanley and Citi signed and run production workloads despite the documented ChatGPT Health distrust that the skeptics say should choke procurement. That is the direct test the bear's channel predicts failing — and it failed. The skepticism-led thesis requires that retail doubt translate into longer sales cycles. If banks are deploying on zero-data-retention contracts and production-scale advisor tools in the same period the consumer product drew its loudest backlash, the transmission path is broken.

The bear answers that this is the wrong sample. Morgan Stanley is a years-old, grandfathered relationship — OpenAI's first strategic financial client, signed before ChatGPT Health existed. It proves OpenAI can sell a financier once, not that a wave of new finance-vertical revenue is arriving. Citi, the bear notes, also buys from Anthropic, Google, and Microsoft; being in the stack is not the same as owning the budget line. The bear concedes the specific channel is weak but insists the real bottleneck sits elsewhere, in accuracy and vendor policy — which is the next round.

Round one: bull, on the evidence margin. A named, paying production check beats an inference from sentiment.

Round two — is the accuracy gap an iteration problem or a compliance barrier?

This is the round where each side faces its strongest fact. The optimist's case is that models are improving quickly and procurement is a lagging indicator — that today's 76%-versus-71% gap closes in a release or two, and that regulated buyers buy the trajectory, not the snapshot.

The bear's heaviest punch is that finance is a category where "almost good enough" is not deployable. A bank cannot put a model to work on the customer's savings if the independent benchmark stops short of client-ready grade, whatever the trend. The DualEntry result — all leading models still short of enterprise-grade financial accuracy — is exactly the kind of finding that lands on a compliance desk's review and lengthens the cycle the way no tweet can. In finance, the seller does not get to call the model ready; the auditor does.

The bull wins the trajectory-of-iteration point as a technical truth, but the bear wins the deployment-round: in a regulated vertical, the benchmark is the gate, and the benchmark is not yet open. Optimists who dismiss the accuracy gap as retail noise confuse the loud complaint with the consequential one.

Round two: bear, narrowly.

Round three — the sign that neither camp is watching

Here is the number that should resolve the fight, because it is about the channel both sides are ignoring. Evident's reading is that OpenAI's share of disclosed bank AI use cases fell from about half eighteen months ago to a third by end-2025, as banks embraced model-agnostic strategies and in-house builds. This is not consumer sentiment reaching procurement. It is the opposite: professional buyers, unbothered by the retail backlash, nevertheless de-bundling OpenAI on their own compliance and cost terms.

That single fact damages both storylines. It says the skepticism-led camp undersells the resilience (bank adoption of AI itself is booming) while the noise camp oversells the franchise (adoption is happening, but increasingly on rival and self-built models). Neither community doubt nor its absence is moving the share line; competition and control are.

Round three: bear, because the measurable direction of finance-vertical penetration is down at the same time the bull needs it up.

What the price demands

Now make both stories pay rent. A private $852 billion valuation asks enterprise — finance being one slice of it — to keep growing near 50% quarter-to-date, and a 2027 listing must convert a $63 billion projected burn into a story investors fund. That is a price that needs every vertical, finance included, to be compounding. The bull asks you to believe the fastest-growing vertical is under-covered and about to inflect. The bear asks you to notice that the only disclosed measure of bank AI penetration is moving away from the supplier, and that the vertical under debate is not reported at all. On the evidence-to-expectation ratio, the burden sits with the bull.

The ruling

Separate the business from the stock, because here they genuinely split. On the specific question in the title — does community skepticism reach bank procurement and stall finance-vertical ARR — the answer is no. The channel is weak; compliance-driven buying overrode it; named, paying institutions are the evidence. The ChatGPT Health precedent is not a reliable leading indicator for regulated-enterprise finance sales.

But the optimists' victory is narrower than they think, because the thing that actually constrains finance-vertical revenue — the accuracy gate and banks' deliberate diversification away from any single lab — is moving the wrong way, and neither is sentiment-driven. So the business thesis (OpenAI can connect regulated finance buyers) holds, while the stock-at-this-price thesis (finance-vertical ARR will inflect and visibly compound into the listing) is unproven against the only proxy that exists.

The verdict: bull on the transmission question, bear on the finance-vertical expectation at this valuation. The skepticism camp loses with the wrong reason; the noise camp wins the point but not the price.

The tripwire. The reveal is the public S-1, the first moment finance-vertical dollars are actually reported — track the filing, then watch whether finance-vertical ARR and the disclosed enterprise growth hold the 50% quarter-to-date pace. In the interim, the earliest confirming indicator belongs to the losing camp: if Evident's share of disclosed bank AI use cases keeps sliding for two more quarters, or the next finance benchmark from DualEntry or Box still stops short of client-ready grade, the bear's engine — accuracy and de-bundling, not Reddit — was right all along. If named finance customers keep arriving through 2027 and the accuracy gate opens, the bull's compliance-buying thesis wins outright. You cannot own OpenAI until it lists. What you can do is keep the scorecard, because the debut will be settled by which measure — sentiment or the audit — was ever the load-bearing one.