What changed since March
| SIGNAL | MARCH 2026 | AUGUST 2026 | EFFECT ON VITALITY |
|---|---|---|---|
| Musk litigation | Trial pending Mar 30; up to $134B sought | All claims dismissed May 18; appeal filed | Friction down sharply. The single largest tail risk in the March assessment is off the board. |
| Audited economics | Projected $14B loss for 2026 | $20.9B operating loss on $13.07B revenue in 2025 | Friction up, Claim Integrity down. Burn is 1.6× revenue, and the correction arrived by leak rather than disclosure. |
| Valuation & capital | $730B on a $110B round | $852B on a $122B round (Mar 31) | External engine deeper. But $35B of Amazon's $50B is contingent on an IPO or an AGI milestone — the fuel supply has a trigger on it. |
| Competitive position | Most valuable private company; enterprise lead assumed | Second to Anthropic on valuation ($965B) and enterprise LLM spend (27% vs 40%) | Friction up. The lead narrative no longer matches the position. |
| Consumer share | 64.5% of chatbot web traffic | ~53% (May); Gemini ~27–28%, Claude ~8–9% | Traction Quality down. Erosion at this rate is the clearest evidence that part of the traction is rented. |
| Monetisation | Ads in testing since Feb 9 | $100M annualized within six weeks; $2.5B projected for 2026 | Internal engine up. The first high-margin, owned-inventory revenue OpenAI has had. |
| Leadership | 20+ senior departures since 2024 | Weil, Peebles, Narayanan (Apr); Simo steps down (Jul); head of safety systems Heidecke departs (Jul) | Friction Quality down. The pattern flagged in March continued, and reached the safety function again. |
| Governance | CA/DE AG investigations open | Resolved; OpenAI Foundation holds ~26% (~$130B) with safety-committee powers | Friction down. Structure is now settled and defensible. |
| Liability | Copyright suits unresolved | Wrongful-death cluster growing: Raine, Scott (May), Parish (Jun); ~two dozen product-liability cases | New friction category. Consumer-harm exposure replaces founder litigation as the live legal risk. |
| Public-market path | No filing | Confidential S-1 filed Jun 8; >$1T target; NYT reports lean toward 2027 | Claim Integrity easier to assess, timing uncertain. |
Six Dimensions
Traction
Magnitude is accelerating. CFO Sarah Friar told staff on July 29 that annualized recurring revenue in July alone exceeded the whole of the second quarter[OAI-TRC-01]. Growth is attributed across three products rather than concentrated in one: the GPT-5.6 model series, the ChatGPT Work enterprise agent, and Codex. ChatGPT is still the most-used consumer AI product by web traffic[OAI-TQ-01]. Advertising went from a February test to $100M annualized within six weeks. The March round closed at $122B.
Breadth is the distinguishing feature. Consumer, enterprise, API, developer tooling and now advertising all carry material volume. No other AI company has traction across that many surfaces at once, and none has a comparable consumer installed base to monetise.
Magnitude and breadth are not in question. What the traction is made of — how much of it survives without capital and discounting — is assessed separately below.
Traction Quality
The question: if the capital and the discounting stopped tomorrow, how much of this would remain?
Compounding share. Advertising is the strongest single item — it monetises inventory OpenAI already owns, at high margin, with no incremental acquisition cost. ChatGPT Work and Codex are embedding into enterprise workflows, which manufactures switching cost. The brand is still the category default for consumers, which is real unpaid pull.
Rented share. An 80% price cut on GPT-5.6 Luna is traction purchased with margin. Consumer share fell from roughly 76% to about 53% of gen-AI web visits in twelve months while spending rose — demand that moves that fast on a competitor's release was not deeply held. Enterprise share sits at 27% against Anthropic's 40%[OAI-TQ-02], and coding, the highest-value workload, went decisively the other way.
The 22-point consumer share loss is the cleanest available evidence on this dimension: traction that erodes at that rate under competitive pressure is, by definition, partly rented. Mixed rather than Weak because the advertising business and the enterprise agents are genuine compounding assets forming underneath the churn.
Friction
Financial — worse than assumed. Leaked audited statements, independently verified by the Financial Times, show a $20.92B operating loss on $13.07B of 2025 revenue[OAI-FIN-02]. Total costs reached $34B, of which $19.18B was R&D — R&D alone exceeded the entire top line. The headline $38.5B net loss includes a $41.55B non-cash charge from the restructuring, so the operating figure is the one that matters.
Competitive — worse. Anthropic passed OpenAI on both valuation ($965B vs $852B) and enterprise LLM spend (40% vs 27%), and holds roughly 54% of enterprise coding-model spend. A price war is now open: OpenAI cut GPT-5.6 Luna pricing by up to 80% to counter Gemini Flash and Claude.
Legal — reshaped, not removed. The Musk suit was dismissed in full on May 18 when a nine-member jury found the claims time-barred[OAI-LEG-01]; Musk called it a "calendar technicality" and is appealing. The California and Delaware AG reviews concluded, leaving the OpenAI Foundation with roughly 26% of the company (~$130B) and a special committee empowered on safety. In their place, a consumer-harm docket is forming: Raine, the Scott overdose case filed in May, and Parish in June, inside a cluster of roughly two dozen suits arguing chatbots should be regulated as products.
Organizational — moderate and partly intentional. Pre-IPO consolidation closed side bets and concentrated effort. That is friction being converted into focus.
Friction Quality
Downgraded from Strong. March held that OpenAI was metabolising most of its friction into capability, with one destructive exception: safety-capacity erosion. Five months on, the productive half held and the destructive half got worse.
Productive, confirmed. Competitive pressure produced a faster model cadence and a differentiated enterprise product. Financial pressure produced advertising and a price cut only a company with scale advantages can absorb. Legal pressure produced a settled governance structure with an endowed foundation and a safety committee. Pre-IPO consolidation converted scattered effort into focus.
Destructive, worsened. The head of safety systems departed in July[OAI-CAP-04], adding to a multi-year pattern of safety leadership loss — as wrongful-death and product-liability suits accumulate, which is precisely when demand for internal risk detection rises. The company is shedding sensors while the environment gets noisier. Nothing in the period suggests this friction is producing learning; it is producing exposure.
Economic Locus
What share of forward motion the business generates itself, and which way that share is moving.
External. The 2025 audited accounts settle the level: a $20.92B operating loss on $13.07B of revenue means roughly $1.60 of injected capital funded every $1.00 earned. $122B was raised in March. Compute runs largely on partner balance sheets — Microsoft, Amazon, Nvidia, SoftBank. And the external engine is itself conditional: $35B of Amazon's $50B is contingent on an IPO or an AGI milestone rather than committed outright, which puts a trigger on the fuel supply.
Internal. Forming, and faster than expected. Advertising is the first genuinely high-margin, owned-inventory revenue OpenAI has had. Revenue is accelerating against a research cost base that does not scale linearly with it, which is the shape that eventually produces operating leverage. Codex and ChatGPT Work carry better unit economics than consumer inference.
Direction of travel is positive; the level is not. This is the dimension that produces the verdict: a company can be enormous, well-capitalised and admired and still run mostly on fuel it does not make.
Note: what March scored as Capacity — balance sheet depth, compute partnerships, talent stock — is now read as evidence here. Capital depth is not strength in itself; it is what the external share is made of.
Claim Integrity
How well stated reality tracks measured reality — and, critically, which direction the corrections travel and whether they arrive voluntarily.
The defining event of this period is the gap between the publicly framed $14B loss and the audited $20.92B operating loss, and the fact that it closed by leak rather than by disclosure. Corrections that travel in one direction and arrive involuntarily are the signature of framing that has been running ahead of the books.
The forward claims remain unretracted and remain extreme: $280B of revenue by 2030 from $13.07B in 2025, and advertising from $2.5B to $100B over the same period. Ads are tracking well against the near-term curve, which raises confidence about 2026–27 and says nothing about 2030.
Strained rather than Weak for two reasons. Nothing here is fabrication. And the direction is improving by force of process — the AG settlement produced documented governance, and the S-1 will convert every ambiguous number into a statement made under legal liability. This dimension should be materially easier to score by the next update.
Traction: Very strong − Friction: Strong
Net balance: Positive, and roughly unchanged in magnitude — but the composition has shifted from legal risk to economic risk.
The audited figures confirm and sharpen what March inferred: every dollar of growth in 2025 was purchased with roughly $1.60 of external capital. Revenue is now growing faster than it was, which improves the trajectory. It does not change the mechanism.
Modified by: Traction Quality is Mixed — a meaningful share of the traction is rented, and the 22-point consumer share loss is the proof. Friction Quality is downgraded to Mixed. Economic Locus is External-dominant but improving. Claim Integrity is Strained.
How the verdict is derived
With Traction Quality and Economic Locus scored explicitly, the durability verdict is computed from the dimensions rather than asserted alongside them. The rules are evaluated terminal-first — Collapsing, then Fragmenting, then Compounding, then Coasting — and the first condition met is the verdict. They are set out below in constructive order for readability; the evaluation order governs.
| VERDICT | CONDITION | RESULT | WHY |
|---|---|---|---|
| Compounding | Traction Quality high AND Economic Locus internal-trending | Not met | Traction Quality is Mixed and the engine remains external-dominant. |
| Coasting | Traction high, Economic Locus external-dominant | MET | Traction Very strong; operating loss exceeds revenue by ~60%; growth funded by injection. |
| Fragmenting | Friction Quality destructive-dominant | Not met | Friction Quality is Mixed — the productive share is still larger, and coherence held. |
| Collapsing | Economic Locus external AND external source withdrawing | Not met | Capital access unimpaired; $122B raised at a higher mark five months ago. |
DURABILITY VERDICT
Coasting
Held from March, for changed reasons. The existential legal threat resolved favourably and revenue growth accelerated — both arguments for an upgrade. Against them: the audited loss is materially larger than the figure the last assessment relied on, and the company surrendered its lead in enterprise while the assessment period ran. Structural health is not deteriorating. It is still not compounding. The engine is bigger, and it still runs on fuel from outside.
Why not an upgrade
Compounding requires that growth be increasingly self-funded. In 2025 it was not: operating losses exceeded revenue by roughly 60%. The 2026 acceleration is real, but the capital intensity of the underlying compute commitments is rising alongside it, and the largest tranche of the March round — $35B of Amazon's $50B — is contingent on an IPO or an AGI milestone rather than committed outright.
Why not a downgrade
Fragmenting would require organizational coherence to break. It has not. Shipping cadence held through a leadership transition, the governance question was settled rather than litigated into crisis, capital access is unimpaired, and the April departures were substantially the result of a deliberate consolidation. The company is losing races, not losing control.
What would move the verdict
Upward: gross margin crossing 50%, a successful listing at or above $852B, and enterprise share stabilising. Downward: a further senior safety departure, an adverse ruling in the product-liability cluster, or a 2026 operating loss that exceeds 2025's despite higher revenue.
Four Scenarios
Compounding
20%
The IPO completes at or above the private mark. Advertising scales toward its 2027 target and carries structurally higher margin than inference. Gross margin crosses 50% as inference costs fall faster than price. Enterprise share stabilises around a defensible floor rather than continuing to slide toward Anthropic. Safety leadership is rebuilt rather than backfilled.
Coasting
50%
Revenue keeps compounding fast and losses persist at scale. The listing happens but prices below the private mark, or slips to 2027. Anthropic holds the enterprise lead; OpenAI holds consumer. The price war compresses everyone. Profitability recedes toward 2030. The company remains a very large, very well-capitalised, loss-making platform — larger every year, and no closer to its own engine.
Fragmenting
22%
The consumer-harm docket becomes the defining story. An adverse ruling classifies chatbots as products, or a high-profile safety failure lands while the safety function is thin. Enterprise customers accelerate diversification. Leadership churn compounds into a credibility problem during a listing process. The brand shifts from "AI leader" to "the AI company with the lawsuits."
Collapsing
8%
Reduced from 10%. The catastrophic legal path largely closed with the May verdict. What remains: capital markets turning against pre-profit AI spending before the listing completes, a severe safety incident triggering regulatory intervention, open-weight models closing the capability gap fast enough to make premium pricing unviable, or several of these compounding.
REPORTED FACTS
Financial: Leaked audited statements, verified independently by the Financial Times, show a $20.92B operating loss on $13.07B of revenue for 2025, with $34B of total costs including $19.18B of R&D. The $38.5B headline net loss includes a $41.55B non-cash charge tied to the nonprofit-to-PBC conversion. Total assets are roughly $50B, about half in cash.
Capital: OpenAI closed a $122B round on March 31, 2026 at an $852B valuation[OAI-CAP-01] — Amazon $50B (of which $35B is contingent on an IPO or an AGI milestone), Nvidia $30B, SoftBank $30B, with Microsoft also participating. A confidential draft S-1 was submitted to the SEC on June 8, 2026, with Goldman Sachs and Morgan Stanley leading; reporting on June 25 indicated a lean toward a 2027 listing rather than September 2026.
Legal: On May 18, 2026 a nine-member jury unanimously found Musk's claims time-barred and returned no liability for Altman, Brockman, or OpenAI. Musk described the outcome as a "calendar technicality"; his counsel confirmed an appeal. The Delaware AG issued a Statement of No Objection to the recapitalisation; the resulting OpenAI Foundation holds roughly 26% of the company (~$130B) and can appoint for-profit directors and act on safety through a special committee. Separately, wrongful-death and product-liability suits including Raine, Scott (May 2026), and Parish (June 2026) are pending within a cluster of roughly two dozen cases.
Competition: ChatGPT's share of gen-AI web visits fell to roughly 53% by May 2026 from about 76% a year earlier; Gemini rose to about 27–28% and Claude to about 8–9%. Anthropic closed a $65B Series H at a $965B post-money valuation on May 28, 2026, and holds roughly 40% of enterprise LLM spend against OpenAI's 27%. OpenAI cut GPT-5.6 Luna pricing by up to 80%.
Product and people: Advertising launched to free and Go tiers on February 9, 2026 and crossed $100M annualized within six weeks[OAI-TRC-02]; investor projections run $2.5B (2026) to $100B (2030). Kevin Weil, Bill Peebles, and Srinivas Narayanan departed in April 2026 amid side-project shutdowns; Fidji Simo stepped down from her full-time role in July 2026 after medical leave; Johannes Heidecke, head of safety systems, announced his departure in July 2026.
REASONABLE INFERENCES
The narrative-to-audit gap is the most important new signal. When the publicly discussed loss figure is $14B and the audited operating loss is $20.9B, and the correction arrives by leak rather than by disclosure, the reasonable inference is not fraud — it is that external framing has been running consistently ahead of internal reality. That matters more now than it did in March, because a company entering a public listing has to close that gap under oath.
Losing the enterprise lead is a structural event, not a quarterly one. Enterprise LLM spend is sticky: it is embedded in workflows, procurement, and security review. A vendor that reaches 40% share against a first mover's 27% has usually won something that does not revert on a single model release. OpenAI's consumer position remains dominant; its enterprise position now requires recapture rather than defence.
The safety-attrition pattern has outlived every alternative explanation. In March it could still be read as turnover in a fast-scaling company. After the departure of the head of safety systems in July — during a period when consumer-harm litigation is accumulating — the pattern is better explained as a persistent structural deprioritisation than as coincidence.
The Musk verdict is worth less than it appears. The jury did not find that the conversion was proper; it found that Musk sued too late. That resolves OpenAI's exposure to Musk. It does not establish a precedent that protects the company from other challengers, and an appeal is pending.
Advertising is the most underrated positive. $100M annualized in six weeks from a standing start, on inventory the company already owns, is a genuinely high-margin revenue stream layered onto an existing user base. If it scales even to half its 2027 projection, the loss trajectory changes shape.
FRAMEWORK INTERPRETATION
Coasting is now a firmer verdict than it was in March, not a softer one. Five months ago the verdict rested partly on uncertainty — the trial, the AG reviews, the projections. Much of that uncertainty has resolved, and it resolved in a pattern: legal risk down, economic risk up, competitive position down, revenue growth up. Those largely cancel. What remains is the same core fact the framework identified in March — the company grows on injected capital rather than on its own cash generation — now confirmed by audited numbers rather than inferred from projections.
Friction Quality is the dimension that moved. Downgrading it from Strong to Mixed is the substantive change in this update. OpenAI is still converting most of its friction into capability; the price war, the governance settlement, and the ad business are all evidence of that. But the destructive pattern flagged in March did not correct — it advanced, and it advanced into a period of rising liability. A company can afford to lose market share. It is much harder to afford losing the people who tell you when something is about to go wrong, in the same year that courts begin deciding whether your product is legally a product.
The IPO is the forcing function. Everything ambiguous in this assessment — margin trajectory, real loss run-rate, compute obligations, litigation reserves — becomes disclosed and enforceable when the S-1 goes public. The next Signal update should be materially higher confidence for that reason alone, and Claim Integrity should be scorable against filed accounts rather than leaks.
Evidence ledger
| ID | VALUE & CLAIM | SOURCE | CLASS · USED IN |
|---|---|---|---|
| OAI-FIN-01 | $13.07B 2025 revenue | Leaked audited statements, independently verified by the Financial Times (reported by Ed Zitron; summarised by MLQ News) 2026 | Audited Traction · Friction · Economic Locus · Claim Integrity |
| OAI-FIN-02 | $20.92B 2025 operating loss | Leaked audited statements, FT-verified (MLQ News) 2026 | Audited Friction · Economic Locus · Claim Integrity · Verdict |
| OAI-FIN-03 | $38.5B Headline 2025 net loss, including a non-cash conversion charge of $41.55B | Leaked audited statements, FT-verified (State of Surveillance summary) 2026 | Audited Friction · Claim Integrity |
| OAI-FIN-04 | $19.18B of $34B total costs 2025 R&D spend, exceeding total revenue | Leaked audited statements, FT-verified (MLQ News) 2026 | Audited Friction |
| OAI-CAP-01 | $122B raised at $852B Funding round closed at a post-money valuation | Bloomberg; corroborated by CNBC and Forbes 2026-03-31 | Press Economic Locus · Traction |
| OAI-CAP-02 | Amazon $50B · Nvidia $30B · SoftBank $30B Round composition; $35B of Amazon's tranche contingent on an IPO or an AGI milestone | CNBC 2026-03-31 | Press Economic Locus |
| OAI-CAP-03 | Filed 8 June 2026 Confidential draft S-1 submitted to the SEC; reporting indicates a lean toward a 2027 listing | Reported via Forbes IPO tracker and The New York Times (25 June) 2026-07 | Press Claim Integrity · Scenarios |
| OAI-TRC-01 | July ARR > Q2 total CFO Sarah Friar told employees that annualized recurring revenue in July exceeded the whole of Q2 | CNBC 2026-07-29 | Company Traction |
| OAI-TRC-02 | Launched 9 Feb 2026 Advertising launched to free and Go tiers; crossed $100M annualized within six weeks; investor projection of $2.5B for 2026 | The AI Insider; projections per investor materials reported by humAI 2026-02-26 | Press Traction · Traction Quality · Economic Locus |
| OAI-TQ-01 | ~76% → ~53% (May 2026) Share of gen-AI web visits fell over twelve months while spend rose; Gemini ~27–28%, Claude ~8–9% | Similarweb data, reported via Enterprise DNA and eMarketer 2026-05 | Estimate Traction Quality · Friction |
| OAI-TQ-02 | 27% Share of enterprise LLM spend, against Anthropic's 40% | Menlo Ventures enterprise LLM market report 2026 | Estimate Traction Quality · Friction |
| OAI-CMP-01 | Up to 80% GPT-5.6 Luna pricing cut to counter Gemini Flash and Claude | VentureBeat 2026 | Press Traction Quality · Friction |
| OAI-LEG-01 | Verdict 18 May 2026 A nine-member jury unanimously found Musk's claims time-barred; no liability for Altman, Brockman or OpenAI. Musk called it a 'calendar technicality'; counsel confirmed an appeal | NBC News; corroborated by NPR, CNBC and Al Jazeera 2026-05-18 | Primary Friction · Scenarios |
| OAI-LEG-02 | Resolved Delaware AG issued a Statement of No Objection to the recapitalisation; the OpenAI Foundation holds roughly 26% (~$130B) with safety-committee powers | Delaware Department of Justice 2025-10-28 | Primary Friction · Friction Quality |
| OAI-LEG-03 | Raine; Scott (May 2026); Parish (June 2026) Wrongful-death and product-liability suits pending within a cluster of roughly two dozen cases | Lawsuit Informer case tracker; Nolo legal encyclopedia 2026-07 | Press Friction · Friction Quality · Scenarios |
| OAI-CAP-04 | 6 named departures in 2026 Senior departures: Kevin Weil, Bill Peebles and Srinivas Narayanan (April); Fidji Simo stepped down from her full-time role (July); Johannes Heidecke, head of safety systems (July) | CNBC (April); Bloomberg (Simo); PYMNTS (Heidecke) 2026-04-17 | Press Friction · Friction Quality |
About the two quality dimensions
Traction quality. Not all traction is equal. Some compounds — it creates switching costs, retained value and demand that survives without subsidy. Some is rented — bought with capital, price cuts, or someone else's distribution, and it leaves when the payment stops. OpenAI has both in large quantities. Advertising and the enterprise agents compound; the consumer base that shed 22 points of share in twelve months while spending rose does not.
Not all friction is bad. Some friction is the necessary cost of learning, adapting, and improving. What matters is whether friction produces learning or just produces waste.
Most of OpenAI's friction is still productive. The price war is forcing inference efficiency. The governance fight produced a settled, endowed structure with real safety powers. The loss position produced advertising. Competitive defeat in enterprise is producing a serious enterprise product. All of that is friction being metabolised into capability.
The critical exception has not corrected. Safety-focused leadership has continued to leave, and in July that reached the head of safety systems — while wrongful-death and product-liability cases accumulate. This friction produces no learning. It removes the capacity to detect problems before they become crises, at exactly the moment courts are beginning to decide how much those crises cost. A company that loses its error-detection system is not getting leaner. It is getting blinder.
Method note
This assessment is based on public information and AI-assisted interpretation. It is designed as a directional monitoring tool, not a definitive audit, legal finding, or substitute for structured due diligence. Confidence depends on the quality, recency, and breadth of available public evidence. Evidence, inference, and interpretation are separated throughout. Scenario probabilities are structured estimates, anchored on the current verdict and updated against explicit watch items. They are not the output of a formal Bayesian model: no priors, likelihood ratios or calibration record are claimed.
Where figures are disputed across sources — notably OpenAI's current annualized revenue, which is reported between roughly $25B and higher depending on definition and date — this report cites the most conservative well-sourced figure and flags the range rather than selecting the most favourable one.
Consumer user counts are deliberately not cited. Third-party estimates for ChatGPT vary by hundreds of millions depending on methodology and date, and none met the evidence standard for a ledger entry. Web-traffic share is used instead, on the same basis as the Anthropic report.
Sources: Every load-bearing figure in this report is listed individually in the Evidence ledger above — 16 claims, each with its source, publication date and evidence class, and the dimensions it was used to score. Claims are referenced inline by ID. All sources were retrieved on 6 August 2026; a machine-readable export is available on request.
Framework: Aisocrat Signal · NSV (Net Strategic Vitality) Framework
Report generated: August 6, 2026 · supersedes March 26, 2026
Next scheduled update: October 2026, or event-triggered on public S-1 filing, IPO pricing, a ruling in the product-liability cluster, or a further senior safety departure.