--- title: "OpenAI's 20bn revenue gap sparks uproar; real impact looks limited — a sentiment false alarm." type: "Topics" locale: "en" url: "https://longbridge.com/en/dolphin/post/44254261.md" description: "The bearish headline that OpenAI's ARR went from 70bn to 50bn triggered a global pullback in AI plays. Dolphin Research offers a quick take on where expectations diverge.1) The 70bn vs 50bn gap mostly reflects definitional differences rather than a real shortfall. Many leading model vendors periodically disclose ARR, but methodology matters — gross vs net revenue, annualizing a 28-day avg x12 vs a single recent week x52, etc. Disclosure timing matters too, e.g., whether it follows a blockbuster flagship launch or a promotion; without that context, cross-company comparisons are not apples-to-apples." datetime: "2026-10-09T08:13:30.000Z" locales: - [en](https://longbridge.com/en/dolphin/post/44254261.md) - [zh-CN](https://longbridge.com/zh-CN/dolphin/post/44254261.md) - [zh-HK](https://longbridge.com/zh-HK/dolphin/post/44254261.md) author: "[Dolphin Research](https://longbridge.com/en/dolphin.md)" generator: "portal-rs" --- # OpenAI's 20bn revenue gap sparks uproar; real impact looks limited — a sentiment false alarm. The headline risk of OpenAI's ARR shifting from RMB 70bn to $50bn triggered a global pullback in AI names. Dolphin Research weighs in on the expectation gaps behind the prints. **1) The RMB 70bn vs. $50bn gap is largely a methodology difference, not a real miss** While leading model vendors periodically disclose ARR, the definitions (gross vs. net, trailing 28-day avg.\*12 vs. last-week\*52) and timing (fresh flagship release or promo underway) matter a lot. Without consistent methodologies, cross-company comps are meaningless and prone to headline whiplash. **Anthropic reports ARR on a gross basis** that includes revenue ultimately shared with external channels. OpenAI reports ARR on a net basis, stripping out revenue shares, i.e., excluding MSFT's take (20% of total ARR) and AMZN Bedrock's cut **(per Anthropic–Bedrock, the channel keeps 50% of gross rev. after inference costs)**. Back in Apr, OpenAI CRO Denise Dresser told staff that Anthropic's revenue looked 'inflated' because it included cloud-channel splits. That comment directly pointed to the gross-vs-net divide. Under the new post-2026 agreement, for every $1 OpenAI earns, it pays MSFT $0.20 through 2030, with a cumulative cap at $38bn. This embeds a structural 20% take-rate until the cap binds. For Bedrock's effective take on gross, assume 35% inference cost; the channel share for O&A equals roughly (100%−35%)\*50% = 32%. That is a directional estimate for the distribution cut post-cost. However, Anthropic's 2025 IPO filing shows combined distribution fees paid to Amazon and Google at ~16% of gross rev. (= RMB 351mn fees / RMB 2.16bn cloud-channel gross). This is well below the 32% heuristic. The gap stems from the filing covering all revenue flowing through AWS and Google Cloud, which includes Bedrock-like distribution (TaaS, where rev. share sits) and Anthropic's direct API sales hosted on AWS/Google where there is no distribution rev., only cloud hosting costs. In short, not every cloud-routed dollar bears a distribution split. **Overall, OpenAI and Anthropic face similar channel economics.** Beyond differing reliance on distributors (pre-1H26, Anthropic leaned more on cloud channels), the main incremental piece is MSFT's 20% economic interest tied to OpenAI's early equity. That 20% flows through as a rev. share on OpenAI's net definition. Now to what RMB 70bn and $50bn each represent. Note that neither figure came from OpenAI's official disclosure. **(1) The RMB 70bn print is gross, not OpenAI's net** The RMB 70bn headline surfaced last month via Reuters: 'OpenAI's current ARR is near RMB 70bn, up ~70% since Q3, with enterprise doubling.' That implies Jun ARR near RMB 40bn. Co-founder Greg Brockman and CFO Sarah Friar told staff that year-end 2025 ARR would exceed RMB 20bn, with Jul running ~2x that pace. Bloomberg later reported via sources that Jul ARR was RMB 40bn, which aligns with that internal color. Given this came from senior management, the Jul RMB 40bn should be net. If the FT-derived Jun RMB 40bn ARR were also net, Jul would be flat MoM, which contradicts third-party alt-data trends. TickerTrends shows ~10% MoM growth in Jul (it blends trend tracking with model fitting that incorporates official signals, so its estimate likely aligns closer to a net definition). True stagnation was in Mar–Apr when Claude Opus 4.6 plus coding agents were surging. After OpenAI shipped GPT 5.5 at end-Apr as the default, expanded Codex, and pushed into ChatGPT mobile, momentum re-accelerated and pricing cuts helped. Against Anthropic, which launched Fable but remained pricey with modest penetration, OpenAI started to visibly gain share. Hence, the late-Jun/early-Jul RMB 40bn and last month's RMB 70bn likely referenced gross ARR, while the early-Aug disclosure of Jul RMB 40bn was on a net basis. The ~20% delta reflects MSFT's cut, with cloud-channel distribution still small as Google and AMZN Bedrock only began scaling ChatGPT distribution in Apr and Jun, respectively. Valuation cross-checks also point to RMB 70bn being gross: At P/ARR, OpenAI's latest round at RMB 1.4tn vs. RMB 70bn ARR implies ~20x; Anthropic at RMB 100bn gross ARR and a RMB 2tn IPO valuation also implies ~20x. **(2) The $50bn figure is most likely net** The $50bn ARR comes from the FT, citing investor communications that OpenAI's ARR is near $50bn. This is quasi-official and consistent with OpenAI's net methodology that excludes MSFT and cloud channel shares. Cloud distribution likely remained modest in Sep. Gross from Bedrock and other cloud channels probably contributed less than 10% of overall GMV (SemiAnalysis pegs Anthropic's Bedrock flow at ~15%). Starting from $50bn net ARR, add back MSFT's 20% to get $62.5bn. Then assume Vertex/Bedrock and other external platforms contribute 10% of flow; with a ~25% channel take on that slice, add back 10%\*25% = 2.5%, yielding $62.5bn/(1−2.5%) ≈ $64bn. **That is still ~10% (about $6bn) below the $70bn headline, but far less alarming than the FT's framing of a '$20bn shortfall.'** Absolute levels can swing near-term supplier expectations. But over a longer horizon, trend matters more: OpenAI's momentum looks healthy recently (~10% avg. MoM in Q3), while Anthropic has hit some friction (~3% avg. MoM in Q3), and the sector continues to grow. Putting this together, Dolphin Research reconstructs recent OpenAI ARR under different methodologies for reference only. Method choices explain most of the optics. **2) AI penetration is still low; near-term cooling doesn't alter the long-term upside** After coding hit >50% penetration in its addressable base, the industry is seeking the next scalable use case to carry monetization and the upstream supply chain narrative. Co-working, an extension from coding into multi-agent collaboration, moved to center stage post-Jun. A quick reality check: co-working's TAM is larger, but its rollout will likely be bumpier than coding. The adoption hurdles are higher and the payback is less immediate. AI coding took off because efficacy is easy to verify and quantify—does the code run, does the app ship; the AI can self-validate in many cases. Economically, it benchmarks cleanly against IT dev labor, which has high hourly rates after two decades of internet growth, making the substitution math straightforward for enterprises. Co-working's TAM is clearly bigger—Gartner estimates agents could influence ~$234bn of enterprise app software spend by 2030, ~20% of SaaS. But the path is more complex than coding. Beyond proving efficacy, enterprises must retain more internal data assets on agent platforms, which raises thornier security debates. **Security risk vs. cost economics requires time for enterprises to evaluate and balance.** Near term, ROI is still framed as labor productivity, but the bigger prize is top-line lift, not just cost cuts. Dolphin Research will publish deeper work on AI office use cases soon; stay tuned. **These are near-to-mid-term issues.** Over the long run, AI's runway remains substantial. A recent a16z report shows: Among U.S. users, broad AI penetration (used at least once) has reached ~50%. **But only ~25% use it daily, and the pay rate is sub-3%** (various sources put paid penetration at 2–4%). In other words, despite rapid model iteration and AI assisting on frontier science, stickiness for most people remains low in both frequency and time. Longer term, that implies significant unrealized application value. **Bottom line: Fixating on the print reflects a lower risk appetite** The FT's '$20bn gap' headline sounds scary, but the real hit is much smaller. Even so, U.S. equities sold off overnight and HK/China opened weak today (Korea and Taiwan were shut; China also faced worries around U.S. curbs on upstream materials), before Asia flows stabilized. Markets' hair-trigger focus on O&A and other top model vendors' ARR tells you expectations are split and risk appetite is down. We remain constructive on AI long term, but Q3 earnings season and Anthropic's IPO filing will test real demand and business models. In a high-rate regime, dispersion-driven volatility will be higher, so diversify exposure appropriately to reduce concentration risk. \ Risk disclosure and statement: [Dolphin Research Disclaimer and General Disclosure](https://support.longbridge.global/topics/misc/dolphin-disclaimer) ### Related Stocks - [ANTH.NA](https://longbridge.com/en/quote/ANTH.NA.md) - [OpenAI.NA](https://longbridge.com/en/quote/OpenAI.NA.md) - [MSFT.US](https://longbridge.com/en/quote/MSFT.US.md) - [GOOG.US](https://longbridge.com/en/quote/GOOG.US.md) - [GOOGL.US](https://longbridge.com/en/quote/GOOGL.US.md) - [GOOGN.US](https://longbridge.com/en/quote/GOOGN.US.md) - [AMZN.US](https://longbridge.com/en/quote/AMZN.US.md) - [MSFD.US](https://longbridge.com/en/quote/MSFD.US.md) - [MSFU.US](https://longbridge.com/en/quote/MSFU.US.md) - [MSFY.US](https://longbridge.com/en/quote/MSFY.US.md) --- > **Disclaimer: This article is for reference only and does not constitute any investment advice.**