Access Restricted

Key Takeaways

• Axios reported on 29 September 2026, citing unnamed sources, an OpenAI annualised run rate of nearly $70 billion, up more than 70% since the start of Q3.^[1]
• FactSet Insight estimated on 23 July 2026 that incremental debt reached 32% of capex at five hyperscalers.^[5]
• The US 10-year Treasury yield closed at 5.28% on 7 October 2026.^[2]

Nearly $70 billion. That is how big OpenAI’s annualised revenue run rate looked in the Axios scoop of 29 September 2026, and the sources were unnamed.^[1] A run rate is the latest period’s revenue multiplied up to a full year. It is not recognised revenue, and it is not profit.

Quick Answer

OpenAI’s reported run rate of nearly $70 billion, per Axios citing unnamed sources, shows fast revenue growth. It does not show profit, costs or compute commitments. With the 10-year Treasury yield at 5.28% on 7 October, those gaps shape the Artificial Intelligence (AI) bubble debate.^[1]

Checklist: four things to check in any AI revenue headline

• Source: press-reported by Axios from unnamed sources, not audited.
• Costs: expenses, profit or loss and compute commitments, none reported.
• Financing: incremental debt at 32% of capex over the last twelve months, per FactSet.
• Discount rate: 10-year yield at 5.28% at the 7 October close.

So is AI revenue keeping pace with AI spending? We followed the money. Four layers turned up: the reported run rate, the capital being spent, how it is financed, and the rate used to value what it might earn. None of them settles the AI bubble debate; together they show what is known and what is missing.

OpenAI Revenue: What Axios Reported, and What It Left Out

Business revenue was up more than 100% over the same period, Axios said, enterprise sales had more than doubled since July, and consumer revenue added in the third quarter alone beat all of 2025.^[1]

Now the fine print. Every figure came from sources Axios did not name, and Reuters, which relayed the report, cited an unnamed person too.^[3] OpenAI has not published these numbers in anything we have verified, and neither have we. Axios said it could not immediately learn details about expenses.^[1]

That last sentence is the whole plot.

Reported	Not reported
Run rate of nearly $70 billion (Axios, 29 Sep 2026)	Expenses
Up more than 70% since the start of Q3	Profit or loss
Business revenue up more than 100%; enterprise sales doubled since July	Compute commitments
Consumer revenue added in Q3 above all of 2025	Audited revenue
Talks to raise at least $30 billion at about $1.4 trillion pre-money (Bloomberg, The Information)	A signed term sheet or completed round

Table 1: What was reported and not reported about OpenAI, as of 8 October 2026. Source: Axios (29 September 2026); Bloomberg, as reported by The Next Web (1 October 2026). Run rates are very good at being misread.

Why a Run Rate Is Not Profit, and How It Colours AI Valuations

Think of a run rate as a speedometer reading, not an odometer: pace at one moment, not distance covered. Reuters noted that such metrics often multiply a single month’s revenue by 12, so one strong month stretched across a year can read as steadier than it was.^[3]

It also differs from recognised revenue, which a company books once it has delivered a product or service. Neither number says what OpenAI revenue cost to earn. Profit is what remains after costs, and no costs were reported.

Valuation talks add a second number to the pile. Bloomberg and The Information reported on 29 September 2026 that OpenAI was in talks to raise at least $30 billion at about $1.4 trillion before the new money. Bloomberg called the round a bridge in place of an initial public offering (IPO, the first sale of a company’s shares to the public); The Information expects a listing in 2027.^[4] March 2026 was $122 billion at $852 billion including new money, so the two are not like for like; roughly, $1.4 trillion is about 64% higher.^[4] These are early talks with no term sheet signed, not a completed round.

One more figure, and it is ours, not a company disclosure: the $1.4 trillion pre-money figure divided by $70 billion is about 20 times the reported run rate. We take no view on whether 20 times is high or low. AI valuations are simple to quote and hard to compare, so we show the working.

The missing costs have to be paid from somewhere, so follow the money.

How More Than $690 Billion of Capex Is Being Financed

The money leads to the hyperscalers: very large cloud and data-centre operators. FactSet Insight’s 23 July 2026 estimates cover five of them, not OpenAI, in three layers.^[5]

Layer one is capex. Capital expenditure (capex) is spending on long-lived assets such as data centres and chips. FactSet put aggregate fiscal 2026 capex for the five at more than $690 billion.^[5]

Layer two is free cash flow, the cash left after capex, expected to be near zero or negative for all but two of the five.^[5]

Layer three is debt. Incremental debt as a share of capex rose from 9% in fiscal 2024 to 32% over the last twelve months before June 2026, and aggregate group debt was about $700 billion.^[5]

For a close-up, take one bond. Amazon.com (NASDAQ:AMZN) had a $25 billion issue that was 2.5 times oversubscribed, meaning investors offered 2.5 times the amount sold, against 3.2 times in March.^[5]

Limits: these are July estimates, and none says how much is tied to OpenAI.

Spending that runs ahead of revenue and leans on borrowing has a famous precedent, so we went back to find it.

Telecom Fibre in 1999 and 2000: When Spending Ran Ahead

Rewind to 1999 and 2000, when telecom companies built fibre-optic networks at speed. A Federal Reserve Bank of Richmond review of the period described uncoordinated, large up-front investment, with voice and data traffic growing more slowly than the new supply.^[6]

The same review said carriers’ revenue did not cover operating costs and network debt, and firms including Global Crossing and Williams Communications filed for bankruptcy.^[6]

The Nasdaq Composite closed at a then-record 5,048.62 on 10 March 2000 and did not close above that level until 5,056.06 on 23 April 2015, according to Fortune.^[7]

That is a debate point, not a forecast. Today’s spenders, financing and rates differ, and nothing here implies the pattern repeats.

The difference this time is the rate attached to all that money.

Treasury Yields and AI Stocks: What a 10-Year Above 5% Does to the Maths

The US 10-year Treasury yield closed at 5.28% on 7 October 2026.^[2] Our guide to the 10-year Treasury yield covers the basics.

Why should AI stocks care? Because of the discount rate: the rate used to turn expected future earnings into a value today. A higher rate means a lower present value for distant earnings, all else equal. As arithmetic only, $100 due in 10 years is worth about $59.78 today at a 5.28% discount rate and about $67.56 at 4.00%. That is an illustration, not a prediction about yields or any company.

It simplifies: firms are usually valued at the Treasury yield plus a risk premium, which the sum leaves out. A higher rate bites hardest on distant earnings.

The S&P 500 and the Nasdaq Composite both closed at record highs on 6 October, according to Yahoo Finance, which said the Nasdaq was helped by Nvidia (NASDAQ:NVDA).^[8] Yahoo Finance also showed a Bank of America chart from strategist Michael Hartnett, in which peak market concentration in most past bubbles capped at about 40%; Yahoo’s own writer drew the comparison with AI-driven concentration today.^[8] That is one point in the debate, not a conclusion.

Three of the four layers are numbers anyone can look up, so where do you look?

Where the Useful Surface Area Sits Today

Here is where we land. The capability is real enough to show up in revenue. But headline revenue is what companies announce or sources leak; cost, financing and the discount rate decide how it reads.

Revenue is the part that gets announced; cost, financing and the discount rate sit in filings and bond markets.

Three places are worth bookmarking, as reading material and not instructions: company filings for capex, debt and cash flow; bond-market reports for how oversubscribed an issue was; and the Treasury’s daily par yield curve for the discount-rate layer.

None of these says whether AI stocks are expensive or cheap, and we are not claiming they do. They show which numbers are reported, which are estimated, and which are missing.

Several of those numbers get a fresh reading in the coming weeks.

What to Watch Next

All times are GMT+8, 24-hour.

• US CPI release, Wednesday 14 October, 20:30: US consumer price index, part of the rate backdrop.
• ASML Q3 results, Wednesday 14 October, 13:00: ASML Holding (NASDAQ:ASML) lists 07:00 Netherlands time; converted by Vantage.^[10]
• TSMC Q3 results, Thursday 15 October, 14:00: Taiwan Semiconductor Manufacturing Company (NYSE:TSM) lists 14:00 Taiwan time.^[9]
• FOMC decision, Thursday 29 October, 02:00: the Fed’s rate decision (meeting 27 to 28 October).
• Alphabet Q3 results, Wednesday 28 October, 04:30 on Thursday 29 October: confirmed by Alphabet (16:30 US Eastern call); other hyperscaler dates not yet confirmed.
• Nvidia Q3 FY27 results, mid to late November: date and time not yet announced.

Stop Loss and exposure. CFD positions linked to AI-related shares or indices carry headline risk. A Stop Loss limits a loss to a planned size. It does not prevent loss, and in fast markets or around price gaps it may not fill at the chosen level, so the loss can be larger than planned. Our guide to risk management techniques goes further.

Leverage and position sizing. CFDs are leveraged products. Leverage magnifies gains and losses alike, so a small price move can have a large effect on an account balance. Depending on the account type and jurisdiction, losses may exceed the amount deposited. Sizing a position so a Stop Loss limits the planned loss to a small percentage of account equity, such as 1% or 2%, limits the damage any single outcome can do.

Back to that scoop. Nearly $70 billion is a reported number from unnamed sources, and it describes revenue only. The AI bubble debate it feeds is increasingly about two other layers as well: how the spending is financed, and the discount rate applied to what it may earn. With the 10-year Treasury yield at 5.28% at the 7 October close, the technology is one part of the question and no longer the only one.

Related Reading

• OPENAI News: latest Vantage coverage of OpenAI.
• Bond Market News: yields, auctions and the rate backdrop.
• NVIDIA News: latest Vantage coverage of Nvidia.

RISK WARNING: CFDs are complex financial instruments and carry a high risk of losing money rapidly due to leverage. You should ensure you fully understand the risks involved and carefully consider whether you can afford to take the high risk of losing your money before trading.

Disclaimer: The information is provided for educational purposes only and doesn’t take into account your personal objectives, financial circumstances, or needs. It does not constitute investment advice. We encourage you to seek independent advice if necessary. The information has not been prepared in accordance with legal requirements designed to promote the independence of investment research. No representation or warranty is given as to the accuracy or completeness of any information contained within. This material may contain historical or past performance figures and should not be relied on. Furthermore estimates, forward-looking statements, and forecasts cannot be guaranteed. The information on this site and the products and services offered are not intended for distribution to any person in any country or jurisdiction where such distribution or use would be contrary to local law or regulation.

References

[1] “Scoop: OpenAI’s annual recurring revenue nears $70B – Axios” https://www.axios.com/2026/09/29/scoop-openais-annual-recurring-revenue-nears-70b Accessed on 8 October 2026.

[2] “Daily Treasury Par Yield Curve Rates – U.S. Department of the Treasury” https://home.treasury.gov/resource-center/data-chart-center/interest-rates/TextView?type=daily_treasury_yield_curve&field_tdr_date_value_month=202610 Accessed on 8 October 2026.

[3] “OpenAI’s annualized recurring revenue nears $70 billion, source says – The Star (Reuters)” https://www.thestar.com.my/tech/tech-news/2026/09/29/openai039s-annual-recurring-revenue-nears-70-billion-axios-reports Accessed on 8 October 2026.

[4] “OpenAI seeks $30bn at a $1.4tn valuation as revenue nears $70bn – The Next Web (citing Bloomberg and The Information)” https://thenextweb.com/news/openai-valuation-1-4tn-30bn-round-revenue-70bn Accessed on 8 October 2026.

[5] “Hyperscalers Tap External Financing as AI Capex Outruns Cash Flow – FactSet Insight” https://insight.factset.com/hyperscalers-tap-external-financing-as-ai-capex-outruns-cash-flow Accessed on 8 October 2026.

[6] “Beyond the Turmoil: The Telecommunications Industry in the Fifth District – Federal Reserve Bank of Richmond” https://www.richmondfed.org/publications/research/econ_focus/2003/winter/cover Accessed on 8 October 2026.

[7] “Nasdaq surpasses Internet-bubble peak to set new record – Reuters, via Fortune” https://www.fortune.com/2015/04/23/nasdaq-supasses-internet-bubble-peak-to-set-new-record Accessed on 8 October 2026.

[8] “Stock market today: S&P 500, Nasdaq hit record highs as Nvidia, AMD lead tech higher – Yahoo Finance” https://finance.yahoo.com/markets/live/stock-market-today-tuesday-october-6-dow-sp-500-nasdaq-080526166.html Accessed on 8 October 2026.

[9] “Quarterly Results, Q3 2026 – TSMC Investor Relations” https://investor.tsmc.com/english/quarterly-results/2026/q3 Accessed on 8 October 2026.

[10] “Quarterly Results – ASML Investor Relations” https://investor.asml.com/quarterly-results Accessed on 8 October 2026.