OpenAI is worth somewhere between $1.2 trillion and $1.5 trillion, depending on who you ask. The gap is the whole story.
The view
In 2025 OpenAI booked $13.07 billion of revenue and spent $34 billion. Six months later investors offered to buy into the company at $1.2 trillion, and OpenAI said no, because it wants $1.5 trillion.
Both of those numbers can be defended. That is what makes this interesting.
Our conclusion is that OpenAI is a genuinely extraordinary operating business and, at the prices currently being discussed, an unattractive security. Those two statements do not conflict, and the distance between them is the most useful thing we can offer a reader.
We value the equity on what we think it is worth in roughly five years, then work backwards to what you could pay today. That produces an entry band rather than a target price. Buy below about $650 billion. Hold between $650 billion and $900 billion, which is where the last completed round sits. Trim between $900 billion and $1.2 trillion. Avoid above $1.2 trillion.
The $1.5 trillion OpenAI is asking for sits well inside avoid. So does the $1.2 trillion investors offered.
What would change our mind: audited gross margin above 50% with a credible path to operating leverage, or clarity on the Microsoft revenue share beyond 2030, or evidence that enterprise retention holds at unsubsidised prices. Each of those is checkable, and we say when.
Part 1. Start with the arithmetic
OpenAI booked $13.07B of revenue in 2025 and spent $34B to earn it. Every dollar of revenue came with $2.60 of cost attached.
The FY2025 numbers are not from a filing. They come from audited documents obtained by the journalist Ed Zitron and independently verified by the Financial Times in June 2026. OpenAI does not publish GAAP statements, so this is the best record available, and everything downstream of it inherits that caveat.
Revenue of $13.07 billion against total costs of $34 billion means every dollar earned arrived with $2.60 of cost attached. The operating loss was $20.92 billion. The headline net loss of $38.53 billion is larger still, but most of that gap is a $41.55 billion noncash fair value charge on convertible interests and warrant liabilities. That is an accounting remeasurement, not cash leaving the building, and anyone citing the net loss as a burn rate is reading it wrong.
The more useful figure sits in between. Gross profit was roughly $5.57 billion on that revenue, a margin near 42.6%, up from 28.4% the year before. Genuine improvement. It is also nowhere near enough, because research and development alone ran $19.18 billion, more than three times gross profit.
This is the shape of the business. The gross margin is improving and the cost of staying at the frontier is growing faster.
Part 2. How $180 billion got raised
From $157B in October 2024 to a $1.5T ask today. The gap between what investors offered and what OpenAI wants is $300B, larger than the whole company was worth two years ago.
OpenAI has raised more than $180 billion since 2015, and the pace of revaluation has been the story rather than the amounts. October 2024 at $157 billion. March 2025 at $300 billion. October 2025 at $500 billion, on an employee share sale rather than new capital. March 2026 at $852 billion, when the company closed $122 billion of committed capital, the largest single financing in Silicon Valley history.
Then the line goes flat, and the flat part is the interesting part.
In August 2026 OpenAI ran a $7 billion employee tender and priced it at the same $852 billion mark. No step up. A company that wanted a higher number for optical reasons had an easy opportunity to take one and declined.
Six weeks later, investors approached the company about a new round at more than $1.2 trillion. Note the direction: the talks were initiated by investors, not by OpenAI, and the company countered at $1.5 trillion, pointing to growth in Codex and the reception of its newest models. The Financial Times reported the $1.2 trillion figure on 15 September. The New York Times reported the $1.5 trillion counter the following day.
Nothing has closed. Both figures could move. But the gap between them is $300 billion, which is larger than the entire valuation of OpenAI two years ago, and it exists because two sophisticated parties are looking at the same business and reaching different conclusions about the next five years.
Part 3. The three numbers everyone confuses
$13B, $40B and $350B all describe the same company. One is booked revenue, one is a month multiplied by twelve, and one is a forecast made in July.
You will see OpenAI described as a $13 billion company, a $40 billion company and a $350 billion company, all in the same week, all correctly.
Recognised revenue is what was earned in a defined period. That was $13.07 billion for 2025, $5.7 billion in Q1 2026 and $6.7 billion in Q2.
Run rate takes a recent month and multiplies by twelve. Bloomberg reported that measure passing $40 billion in August 2026. It is a statement about current pace, not about money booked.
Forecast is a claim about the future. Reuters and the FT reported a company planning presentation projecting revenue rising from $36 billion in 2026 to $350 billion in 2030, alongside $278 billion of cumulative negative free cash flow over the same period and roughly $856 billion of compute and infrastructure spending. That presentation was prepared in July, which means the newest number in circulation has the oldest vintage of the three.
The near term burden is worth doing by hand. First half 2026 revenue was $12.4 billion. Hitting $36 billion for the year requires $23.6 billion in the second half, averaging $11.8 billion a quarter, which is about 76% above the second quarter's actual result. Not impossible after a strong model launch. Not a formality either.
Part 4. What the money bought
More than $180B raised, and commitments with nearly every major chip and cloud provider. Adding these together counts the same data centres several times over.
OpenAI’s infrastructure commitments read like a sovereign industrial programme. More than $100 billion to AWS over eight years. Over $300 billion across five years in the Oracle and Stargate arrangements. A $250 billion Azure commitment. Letters of intent with NVIDIA worth up to $100 billion. Custom silicon programmes with Broadcom and AMD. Contracts with CoreWeave and Cerebras.
The temptation is to add these up. Resist it.
A single data centre can house chips bought under a silicon agreement, sell capacity under a cloud contract, and appear inside a Stargate programme total. Adding all three counts the same racks three times. Nominal gigawatts are also not interchangeable across chips or workloads, and capacity suited to a training run may be poorly matched to latency sensitive inference.
The genuine risk in this map is not the size of the numbers. It is their shape. Take or pay commitments create operating leverage in both directions. During growth they guarantee capacity. When demand or pricing disappoints, they become a fixed burden that does not care about your revenue. A rapid gain in model efficiency, which would be good news for the product, can strand capacity procured on older assumptions.
One contract matters more than the rest. Microsoft’s April 2026 amendment made the model licence nonexclusive through 2032 and ended Microsoft’s revenue share payments to OpenAI, while OpenAI’s payments to Microsoft continue through 2030 subject to a cap. The percentage and the cap are not public. PitchBook has called that cap the single variable that makes a positive free cash flow model possible, estimating $70 to $97 billion of avoided revenue share through 2030. Terms after 2030 are unpublished. An investor is being asked to underwrite a business whose largest contractual economics are invisible past the fourth year of their holding period.
Part 5. Who owns it is not who controls it
The nonprofit Foundation owns about 26% and appoints the board. Public shareholders would own the cash flows without any say over who runs the company.
The entity you would buy is OpenAI Group PBC, a Delaware public benefit corporation. The entity that controls it is the OpenAI Foundation, a nonprofit holding roughly 26% of the economics and the right to appoint and remove the Group board.
Public benefit corporation is a legal form, not a statement about who can trade the shares. The practical consequence is that economic ownership and control have been formally separated. You can own the cash flows without any ability to replace the board or to require that management prioritise financial returns when they conflict with the safety mandate.
That mandate is not decorative. On 12 September, Anthropic's Dario Amodei called publicly for the industry to slow the rate at which model capabilities advance. Sam Altman backed him the same day, as did Elon Musk. AI stocks fell on the Monday. Altman then told Fortune that a 2026 listing would be ill advised given the safety debate.
Read that sequence as an investor. The controlling shareholder's mandate produced a public commitment that moved the sector, and the CEO cited it as a reason not to list. Whatever you think of the substance, the governance structure demonstrably influences capital markets decisions. That is a real cost and it should carry a real discount.
The counter argument deserves space. Mission control may preserve enterprise trust and prevent liabilities that would dwarf the discount. Altman is widely reported to hold little or no personal equity, which is unusual alignment for a founder taking a company public. Neither effect is zero, and the prospectus will need to make the rights explicit.
Part 6. The two months that changed the argument
OpenAI's share of developer spend on OpenRouter went from 20% to 50% against Anthropic in two months. A lead won by one model launch can be lost to the next.
Through the first half of 2026 the competitive story was going against OpenAI. Growth flattened near a $25 billion run rate. Anthropic was taking enterprise and coding share, and Menlo Ventures' survey work had OpenAI's share of enterprise LLM spend falling from 50% in 2023 to 27%, while Anthropic climbed to 40%.
Then GPT-5.6 shipped in July and GPT-6 Astra on 3 September, and the picture inverted fast. Gavin Baker, who runs Atreides Management, posted that OpenAI's share of spend routed through OpenRouter went from 20% to 50% against Anthropic between June and September. He called the gains extraordinary. Reuters reported on 19 September that Anthropic is weighing the release of a new frontier model specifically to respond.
Two caveats before anyone extrapolates. OpenRouter is one developer routing hub and its mix is not the enterprise market. And a share recapture achieved in two months by a model launch can be reversed in two months by the next one, which is precisely the problem with a moat built on model quality.
Astra also carries a detail most launch coverage flattened into a single headline number. Its advertised context window is 1.05 million tokens, and its standard price is $10 per million input tokens and $50 per million output. Once a prompt passes 272,000 input tokens, the entire request reprices at $20 and $75, not just the excess. The same threshold applies across OpenAI's GPT 5.6 family, so this is a pricing policy rather than an Astra quirk. Roughly three quarters of the advertised capacity is sold at a price that does not appear in the headline. For anyone modelling revenue per token, that is a real input.
Part 7. Reverse the question
At $852B, OpenAI needs about $258B of 2030 revenue to return 15% a year. At the $1.5T it is asking for, it needs about $455B, well above its own $350B plan.
The usual way to value a company like this is to pick a revenue multiple and argue about it. That approach fails here, because nobody knows what multiple applies to a business with 42% gross margins, a $278 billion planned cash burn and a nonprofit board.
So we run it backwards. Instead of asking what OpenAI is worth, we ask what each price is asking you to believe.
The mechanics are simple enough to check. Assume you buy today and sell at the end of 2030, a little over four years. Require 15% a year, which is a reasonable hurdle for a single illiquid position with this risk profile. Assume a 22% net margin at maturity, a 30 times exit multiple, and that you keep 80% of your ownership after future financings dilute you. Terminal value is revenue times margin times multiple. Multiply by retained ownership, discount back, and you have the most you can pay today.
At $852 billion, the last completed mark, OpenAI needs roughly $258 billion of revenue in 2030 to return 15% a year. At the $1.2 trillion investors offered, it needs about $364 billion. At the $1.5 trillion OpenAI wants, it needs roughly $455 billion.
Now put those against the company’s own plan, which projects $350 billion of 2030 revenue.
The $1.2 trillion offer requires OpenAI to modestly beat its internal bull case just to clear a 15% return. The $1.5 trillion ask requires it to beat that plan by 30%. You are not being asked to believe the plan. You are being asked to believe the plan is conservative, while accepting the execution risk without compensation.
Every input here is arguable. A higher exit multiple, less dilution, or a lower required return all improve the picture, and you should push them around yourself. The model is attached so you can. What does not change under reasonable assumptions is the ordering: at the prices being discussed, the margin of safety is negative.
Part 8. Four futures
A company reaching $200B of revenue at a 20% margin would be a historic success, and a buyer at $852B would still earn under 3% a year.
Scenarios are more honest than a target price, because they show you where the answer comes from.
In the adverse case, competition commoditises the profitable workloads, open weight models cap pricing, and OpenAI reaches $100 billion of 2030 revenue at a 10% net margin. The equity is worth $89 billion to today's buyer. From $852 billion that is a loss of roughly 32% a year.
In the middle case, the business does well. Revenue reaches $200 billion, margins settle at 20%, the market pays 30 times earnings. Today's shares are worth around $528 billion. From the March mark that returns 2.8% a year, which is less than a Treasury bill currently pays.
Sit with that one, because it is the most instructive number in this report. A company that reaches $200 billion of revenue at a 20% net margin by 2030 would be one of the great commercial achievements in corporate history, and an investor entering at $852 billion would have earned almost nothing for the risk.
Our own base case is slightly more constructive. We model $250 billion of revenue at a 22% margin, which values today's equity near $725 billion and returns roughly 7.5% a year from the March mark. Still below our hurdle. From $1.5 trillion it is mildly negative.
The favourable case is where the bulls live and it is not absurd. Revenue hits the plan at $350 billion, margins reach 30% because custom silicon and utilisation finally bite, and the market pays 35 times. Today's equity is worth $1.8 trillion, and even a buyer at $1.5 trillion earns 27% a year. This outcome requires almost everything to go right at once, which is exactly why it should not be the base case.
Part 9. The bands
Buy below $650B, hold to $900B, trim to $1.2T, avoid above that. Both the $1.2T offer and the $1.5T ask sit in avoid.
Buy below roughly $650 billion. Here the required 2030 revenue drops under $200 billion, comfortably inside the plan, and you are paid for the execution risk you are taking. Given the last mark was $852 billion and investors are bidding $1.2 trillion, this level is unlikely to appear outside a broad market dislocation. Which is the point of having a line.
Hold between $650 billion and $900 billion. The March mark sits in here. Fair rather than cheap. If you already own it through a fund or a secondary, there is no case for panic and no case for adding.
Trim between $900 billion and $1.2 trillion. You are paying for the plan to land. Existing holders should be taking something off.
Avoid above $1.2 trillion. Both numbers currently under discussion sit here. At $1.5 trillion you are paying more than the company’s own bull case and receiving the downside for free.
One important caveat on all of this. These bands describe a valuation, not a share price, because the fully diluted cap table is not public. Forge Global marked the shares at $721.85 in late August, implying roughly $894 billion on about 1.24 billion shares, but we have no way to confirm that share count and we do not build on it. Anyone quoting a per share IPO target is guessing at a denominator nobody has published.
Part 10. What actually breaks the thesis
Generic risk sections are useless, so here are the specific ones, in the order we weight them.
The Microsoft cap after 2030. PitchBook has described it as the single variable that makes a positive free cash flow model possible. The percentage and the cap amount are not public, and after 2030 terms are unpublished entirely. A renegotiation that goes badly changes the model more than any competitive development.
Model parity arriving faster than monetisation. The Astra recovery proved a share swing of 30 points is achievable in two months. That cuts both ways. Anthropic is reportedly preparing a response, and Google’s distribution does not require anyone to choose Gemini deliberately.
The open weight price floor. Llama, DeepSeek and their successors set a near zero price for good enough. Gross margin is a function of how much work remains above that floor, and that is an empirical question nobody has answered.
Financing correlation. Suppliers and investors are willing to fund expansion precisely when expectations are high. SoftBank carries a $40 billion bridge maturing March 2027. Amazon’s $35 billion second tranche is contingent on an IPO by end 2028 or an AGI milestone. If sentiment turns, funding availability and business prospects deteriorate together.
Safety as a financial variable. Astra was the first OpenAI model rated Critical for cyber capability under the preparedness framework, with full features gated behind a partner programme. A controlling nonprofit board plus capability gating means product decisions can be made for reasons that reduce near term revenue, and shareholders cannot vote on them.
Duration. Break even is guided to around 2030. A 2027 listing means three years of public losses at mega cap scale. That profile is priced very differently in a risk off tape than a risk on one.
Part 11. How to take exposure, and what to watch
There is no OpenAI ticker and there will not be one in 2026. Altman has said so on the record. The practical routes are indirect, and each one is a different security with a different payoff.
Microsoft gives you roughly a 27% as converted stake plus Azure pull through, in a company where OpenAI is a mid single digit percentage of equity value. The cleanest liquid proxy and the most diluted.
Nvidia is the toll road. It holds $30 billion of OpenAI equity and sells into every lab’s training and inference bill, which means it wins whether or not OpenAI specifically wins.
Amazon owns pieces of both major labs, which makes it a barbell rather than a bet.
SoftBank is the highest torque and the highest noise, carrying leverage against the position.
ARK’s ETFs hold a reported $240 million of OpenAI across ARKK, ARKW and ARKF from the March round, giving retail the only easily accessible direct stub, at roughly 3% look through weights and with active manager NAV volatility on top.
Secondary platforms offer accredited investors direct paper at wide spreads and a significant information disadvantage.
The monitoring list. Five things would change our view, and four of them are observable without a prospectus.
Whether the private round closes, and at what number. A close at $1.2 trillion tells you institutional demand is real at a level our model says is too high. No close tells you the opposite.
Whether OpenRouter share holds through Anthropic’s next model. Two months of share gain is a data point. Six months is a moat.
Anthropic’s November listing and its first print. It will set the comparable multiple for every frontier lab, and it arrives before OpenAI does.
Any disclosure of the Microsoft revenue share cap, which would resolve the largest unquantifiable in the model.
And the first audited cash flow statement, whenever it arrives, which will replace run rate mythology with recognised revenue, stock compensation and related party detail.
The conclusion
OpenAI built the most recognisable product of this technology cycle and one of the fastest revenue ramps in corporate history. The business is real, the reacceleration after Astra is real, and anyone telling you demand is fake is not reading the same evidence we are.
The security is a different matter. You would be buying a claim on cash flows that begin around 2030, from a company controlled by a nonprofit board, dependent on a contract you cannot read past the fourth year, competing against a rival with a higher run rate and a tighter multiple that is likely to list first.
At $852 billion that trade is defensible. At $1.2 trillion it is expensive. At $1.5 trillion you are underwriting the bull case and being paid nothing for the risk that it does not arrive.
The $300 billion gap between what investors offered and what OpenAI asked is not a negotiation detail. It is two sets of sophisticated people disagreeing about whether this company grows into the most valuable software business ever built or merely becomes a very large one. Our model says the investors are closer to right, and that even they are paying a full price.
We will be watching the November comparable.
Source register
Graded. A means issuer, counterparty or fund primary disclosure. Primary does not mean audited or unbiased. B means identified reporting on private information. C means indexed public commentary, usable within its stated limits.
IPO status and process
OpenAI, confidential draft S-1 submitted. 8 Jun 2026. A. https://openai.com/index/openai-submits-confidential-s-1/
Reuters via Investing.com, OpenAI IPO will not happen in 2026, Altman. 12 Sep 2026. B. https://www.investing.com/news/stock-market-news/openai-ipo-will-not-happen-in-2026-amid-ai-safety-fears-altman-says-4898687
FT via PYMNTS, OpenAI eyes $1.2T valuation in pre-IPO round. 16 Sep 2026. B. https://www.pymnts.com/news/artificial-intelligence/2026/openai-eyes-1-2-trillion-valuation-in-pre-ipo-funding-round/
Bloomberg, OpenAI weighing round at over $1.2T. 16 Sep 2026. B. https://www.bloomberg.com/news/articles/2026-09-15/openai-weighing-funding-round-at-over-1-2-trillion-valuation
Forbes, OpenAI weighing round at up to $1.5T (NYT DealBook sourced). 16 Sep 2026. B. https://www.forbes.com/sites/siladityaray/2026/09/16/openai-is-reportedly-weighing-new-funding-round-at-15-trillion-valuation/
Qz, talks early, investor initiated, figure may not hold. 16 Sep 2026. B. https://qz.com/openai-funding-round-1-trillion-valuation-ipo-091626
Financials
Ed Zitron, Where’s Your Ed At, reported FY2024 and FY2025 audited statements. 15 Jun 2026. B. https://www.wheresyoured.at/exclusive-openai-financials/
FT, OpenAI spending hit $34bn last year. 15 Jun 2026. B. https://www.ft.com/content/e15b0d7e-ff6b-4f16-ba7a-4068feddb828
Quartz via Yahoo Finance, 2025 financials leaked, corroboration. 16 Jun 2026. B. https://finance.yahoo.com/markets/stocks/articles/openai-2025-financials-leaked-38-121508294.html
Bloomberg Law, run rate tops $40B. 13 Aug 2026. B. https://news.bloomberglaw.com/artificial-intelligence/openais-revenue-run-rate-tops-40-billion-ahead-of-ipo
Sacra, company profile, Q1 and Q2 2026 revenue. Aug 2026 update. B. https://sacra.com/c/openai/
Reuters via Investing.com, $278B cumulative burn to 2030, July presentation vintage. 18 Sep 2026. B. https://www.investing.com/news/economy-news/openai-expects-to-burn-through-almost-280-billion-by-2030-ft-reports-4907970
FT, OpenAI expects to burn $280bn by 2030. 18 Sep 2026. B. https://www.ft.com/content/6011d061-eee3-4193-b3b7-8ee4155f538c
Financing history
OpenAI, $122B raise at $852B. 31 Mar 2026. A. https://openai.com/index/accelerating-the-next-phase-ai/
OpenAI, Scaling AI for everyone, initial $110B announcement. 27 Feb 2026. A. https://openai.com/index/scaling-ai-for-everyone/
OpenAI, Scale the benefits of AI, $6.6B at $157B. 2 Oct 2024. A. https://openai.com/index/scale-the-benefits-of-ai/
Reuters via Yahoo, $500B employee share sale. 2 Oct 2025. B. https://finance.yahoo.com/news/openai-hits-500-billion-valuation-050859895.html/
SoftBank, 2026 commitment, three tranches. 27 Feb 2026. A. https://group.softbank/en/news/press/20260227
SoftBank, first tranche executed. 1 Apr 2026. A. https://group.softbank/en/news/press/20260401_0
SoftBank, second tranche executed. 1 Jul 2026. A. https://group.softbank/en/news/press/20260701
SoftBank, $40B bridge facility, maturity Mar 2027. 27 Mar 2026. A. https://group.softbank/en/news/press/20260327
Amazon, $50B structure and $100B AWS expansion. 27 Feb 2026. A. https://press.aboutamazon.com/2026/2/openai-and-amazon-announce-strategic-partnership
Structure and Microsoft
OpenAI, Our Structure, Foundation control and recap ownership. Oct 2025. A. https://openai.com/our-structure/
Microsoft, next phase of the partnership, April amendment. 27 Apr 2026. A. https://blogs.microsoft.com/blog/2026/04/27/the-next-phase-of-the-microsoft-openai-partnership/
OpenAI, next chapter of the Microsoft partnership. 28 Oct 2025. A. https://openai.com/index/next-chapter-of-microsoft-openai-partnership/
Compute
NVIDIA, 10GW letter of intent, up to $100B. 22 Sep 2025. A. https://nvidianews.nvidia.com/news/openai-and-nvidia-announce-strategic-partnership-to-deploy-10gw-of-nvidia-systems
OpenAI and AMD, 6GW programme and warrant. 6 Oct 2025. A. https://openai.com/index/openai-amd-strategic-partnership/
OpenAI and Broadcom, 10GW custom accelerator. 13 Oct 2025. A. https://openai.com/index/openai-and-broadcom-announce-strategic-collaboration/
CoreWeave, contract expansion to ~$22.4B. 25 Sep 2025. A. https://investors.coreweave.com/news/news-details/2025/CoreWeave-Expands-Agreement-with-OpenAI-by-up-to-6-5B/default.aspx
OpenAI and Cerebras, 750MW through 2028. 14 Jan 2026. A. https://openai.com/index/cerebras-partnership/
SoftBank, Stargate announcement. 22 Jan 2025. A. https://group.softbank/en/news/press/20250122
SoftBank, Oracle and OpenAI, five new Stargate sites. 24 Sep 2025. A. https://group.softbank/en/news/press/20250924
Product and competition
OpenAI, safety overview GPT-6 Astra, Critical cyber rating. 3 Sep 2026. A. https://openai.com/index/safety-overview-gpt-6-astra/
OpenAI, advancing the price performance frontier with GPT-5.6. 30 Jul 2026. A. https://openai.com/index/advancing-the-price-performance-frontier-with-gpt-5-6/
Gavin Baker, OpenRouter share 20% to 50% versus Anthropic. Sep 2026. C.
Techstrong, Anthropic weighing a counter model, citing Reuters. 19 Sep 2026. B. https://techstrong.ai/articles/anthropic-weighs-counterstrike-against-openais-gpt-6-astra-ahead-of-2-trillion-ipo-report/
Codersera, GPT-6 Astra pricing verified against OpenAI’s rate card, 272,000 token threshold. 3 Sep 2026. B. https://codersera.com/blog/gpt-6-astra-pricing-api-costs-2026/
CloudZero, GPT-6 pricing and the long context repricing rule. Sep 2026. B. https://www.cloudzero.com/blog/gpt-6-pricing/
Menlo Ventures, 2025 State of Generative AI in the Enterprise. Dec 2025. A. https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/
Anthropic comparison
Bloomberg citing the FT, adjusted operating profit and 80% plus gross margin before revenue share and training. 13 Sep 2026. B. https://www.bloomberg.com/news/articles/2026-09-13/anthropic-sees-adjusted-operating-profit-this-quarter-ft-says
Deep Quarry (Olga Usvyatsky), the FT measure excludes stock based compensation. Sep 2026. C.
Bloomberg, $15B revolving credit facility, Morgan Stanley, Goldman, JPMorgan, Citi. 3 Sep 2026. B. https://www.bloomberg.com/news/articles/2026-09-03/anthropic-nears-finalizing-15-billion-pre-ipo-credit-facility
Bloomberg, Anthropic run rate surpasses $65B. 17 Aug 2026. B. https://www.bloomberg.com/news/articles/2026-08-17/anthropic-revenue-run-rate-surpasses-65-billion-ahead-of-ipo
Axios, Anthropic run rate and IPO timing. 17 Aug 2026. B. https://www.axios.com/2026/08/17/anthropic-revenue-run-rate-ipo-openai
Bloomberg, annualized revenue to top $100B in 2026, NYT sourced. 18 Sep 2026. B. https://www.bloomberg.com/news/articles/2026-09-18/anthropic-s-annualized-revenue-to-top-100-billion-in-2026-nyt
Axios, Anthropic tops $100B pace, November trading. 18 Sep 2026. B. https://www.axios.com/2026/09/18/anthropic-100-billion-revenue
AIwire, Anthropic $65B Series H at $965B. 29 May 2026. A syndicated. https://www.hpcwire.com/aiwire/2026/05/29/anthropic-raises-65b-in-series-h-funding-at-965b-post-money-valuation/
Manager commentary
Reuters via The Star, Cathie Wood on the industry slowdown pledge. 15 Sep 2026. B. https://www.thestar.com.my/business/2026/09/15/ark039s-wood-says-global-ai-leaders039-pledge-is-positive-for-technology-development
The Next Web, Wood on extinction warnings, ARK exposure across four companies. Sep 2026. B. https://thenextweb.com/news/cathie-wood-ark-ai-pledge-extinction-ridiculous
Coatue CTEK, fund holdings disclosure. 31 Aug 2026. A. https://www.coatue.com/ctek/performance
Robinhood, Ventures Fund I purchase of OpenAI common. 22 Apr 2026. A. https://robinhood.com/us/en/newsroom/rvi-openai/
ARK Invest, state of AI infrastructure. 25 Mar 2026. A. https://www.ark-invest.com/articles/analyst-research/the-state-of-ai-infrastructure-demand-costs-custom-silicon
Oaktree, Howard Marks, Is it a bubble. 9 Dec 2025. A. https://www.oaktreecapital.com/insights/memo/is-it-a-bubble
Legal
AP, Musk lawsuit time bar verdict. 18 May 2026. B. https://apnews.com/article/0b9b0bfaffe96f2c930341f52dfe4f8c
AP, AI slowdown antitrust complaint. 18 Sep 2026. B. https://apnews.com/article/960af4308161eaf4ed13c383b0ce1c1b