US Venture Hit $516B in 9 Months, 44% Above the 2021 Record

“US venture invested $516b in the first nine months of 2026. That is already 44% above the full-year record set in 2021,” Theory Ventures investor Tomasz Tunguz wrote on 23 October 2026, citing the PitchBook-NVCA Venture Monitor. In the same essay he argues that AI applications reached 10% of infrastructure spending in their fourth year, a mark the web took seven years to pass.

His essay, “How Large is the AI Circle?”, asks whether AI spending is circular, with startups paying startups. This piece sets out his figures, the sources he names for them, and the caveats he attaches, next to our own reporting on who pays for AI.

Business Pill · FINANCIAL CONDITIONS

A short Business Engineer explainer on financial conditions: how the cost and availability of money change. It teaches the general idea only and says nothing about the figures in this piece.

What Tunguz Measured

Tunguz borrows his frame from the economist Carlota Perez: “Every major technology build-out requires two phases, according to Carlota Perez: the infrastructure to power it, & then the applications to pay for it.”

For the web, he writes that “Between 1996 & 2005, American telecom companies spent $1.5t building the internet in today’s dollars. Meanwhile, applications riding on that infrastructure booked $180 billion of revenue.” That is about 12 cents of application revenue per dollar of infrastructure, by our arithmetic ($180 billion divided by $1.5 trillion).

The capex figure comes from USTelecom’s historical broadband capex, he notes. The revenue figure comes from a 24-company panel of SEC filings, from Amazon and Yahoo to Netflix and Salesforce, which he calls a floor for 2001 to 2005.

His summary of the comparison: the web build-out “took seven years to get application revenue above 10% of capex, & ten to reach 48%. AI has reached that 10% mark in four years, roughly half the time.” His chart puts the AI era at 1%, 2%, 4% and 10% in its first four years, from 2023, with the 2026 bar marked by an asterisk.

What Was Posted on X

Tomasz Tunguz of Theory Ventures posted his essay as an X article on 23 October 2026. He opens with Carlota Perez’s two-phase frame, infrastructure first and then the applications that pay for it, and says AI is reaching the web’s mark of application revenue against capex in about half the time. The figures in this piece come from that essay and its footnotes.

CNBC quotes venture capitalist Jenny Lee, speaking at the DBS Asia Future Forward Conference, saying the average holding period before a company goes public is eight to 10 years, but the later-stage companies she now sees are preparing for IPOs at four to five years old. The view is hers; the piece sets it beside the venture totals above.

The Numbers, Side by Side

The table lists the figures from Tunguz’s essay and its footnotes. The two eras are matched on years since each build-out began, as he does.

Figures as published by Tomasz Tunguz, 23 October 2026. He adjusts for inflation with CPI-U, and the 2026 venture figure covers nine months.

The Venture Money

The $516 billion is all US venture investment, not AI alone. In a footnote, Tunguz gives the full-year 2021 record as $357.9 billion and adds that “AI companies took 82.7% of 2026 deal value, the highest share in the dataset.”

He also writes that “Cumulative venture investment in this wave is about six times the first four years of the web era.” His footnote puts that at $1.24 trillion against $201.8 billion, and adds that against the web era’s entire decade the same total is 2.0x.

The Payback Claim

Tunguz writes that “the unit economics from companies like SpaceX suggest that the paybacks on the infrastructure from this revenue are short, about a year.” His footnote quotes SpaceX’s third-quarter S-1: “at current pricing, new compute deployments are generating a payback of less than a year.”

His own working, from a $46-per-watt build, gives 1.21 years before tax and 0.97 years with bonus depreciation, which he calls “a timing benefit rather than a cost reduction.”

How Big Is the Circle

The circularity question is his own: “The major risk is whether the ecosystem is an ouroboros: is all the spend just within startups?”

His answer rests on Ramp data: “Ramp’s data show that 80% of the largest AI spenders are venture-backed. Those firms now account for more than 13% of American employment, up from about 5% during the web era. The circularity is real, but the circle is a sixth of the economy.”

The two figures in that passage differ. A sixth is about 16.7%, by our arithmetic, while 13% is closer to one-eighth. His footnote gives the 13% as an employment share for 2017 to 2021 from Census data analysed by Ando, and calls the comparison with the web era “directional”.

He lists caveats on the Ramp figure himself. The chart “measures dollars spent, not tokens”. Ramp’s sample is “skewed toward tech”. And “venture backing is not the same as venture dollars paying for the tokens”. He also notes that Apollo and Ramp “both have an interest in it being vivid.”

Connecting the Coverage

Our Ramp piece reported the data Tunguz builds on. Ramp economist Ara Kharazian wrote that “80% of spend on OpenAI and Anthropic is driven by 1% of firms”, and Apollo’s chart puts the venture-backed share of the top 1% of AI spenders at 80.4%, against 13.1% for the median spender.

Our Chamath piece covered the sharper version of the circularity claim: Chamath Palihapitiya wrote on X on 21 October that “roughly, $0.70-75 of every dollar of AI spending right now comes from AI companies spending VC investment dollars.”

Our YipitData piece covered a different measure at one lab: YipitData writes that “the top 1% of customers make up nearly 50% of Anthropic spend.”

On the infrastructure side, our financing analysis set Andy Jassy’s “2027 is largely sold out” against a week of rising borrowing costs. Our services piece reported that Anthropic and OpenAI now sell deployment help through separate businesses, Ode and the Deployment Company.

The Structural Read

As we read it, the essay turns the circularity question into a size question: Tunguz does not dispute that much AI spend comes from venture-backed firms, he argues those firms are now a larger part of the economy, at more than 13% of US employment by the figure he cites.

As we read it, his own footnotes mark where the argument is thinnest: the Ramp chart measures dollars spent rather than tokens, and venture backing is not the same as venture dollars paying for them.

As we read it, the payback figure carries a condition: SpaceX’s S-1, as he quotes it, ties the less-than-a-year payback to current pricing, and his own working gives 1.21 years before tax.

Three Implications

FOR READERS OF AI REVENUE Tunguz puts application revenue at 10% of infrastructure capex in the fourth year of the AI build-out, against 5% in the web era’s fourth year, per his chart.

FOR VENTURE INVESTORS PitchBook-NVCA data, as he cites it, put US venture at $516 billion in nine months of 2026, with AI companies taking 82.7% of deal value.

FOR THE CIRCULARITY DEBATE Chamath Palihapitiya estimated $0.70-75 of every AI dollar comes from VC money; Tunguz puts venture-backed firms at more than 13% of US employment. The two measure different things.

The Business Engineer Lens

This story maps onto the Business Engineer framework The AI Capex Map & The State of AI Hyperscalers.

The framework describes its subject as “How the money flows, where it cracks, what it buys, who can absorb it”.

As we read it, Tunguz’s ratio of application revenue to capex is a measure of what the money buys, and his circle of venture-backed firms is a measure of who absorbs it. The framework was published in May 2026 and does not discuss his essay.

What Is Not Established

• What the asterisk on the AI-era 2026 bar means. The essay’s text does not explain it.
• How much of AI application revenue is paid with venture money. Tunguz writes that venture backing is not the same as venture dollars paying for the tokens.
• Whether the SpaceX payback holds at other prices. The S-1 quote is conditioned on “current pricing”.
• Whether the employment shares are comparable across eras. Tunguz calls the comparison directional, from two different sources.
• The full-year 2026 venture total. The $516 billion covers nine months.

The Bottom Line

Tunguz’s figures put AI applications at 10% of infrastructure spending in four years against seven for the web, and US venture investment at $516 billion in nine months of 2026. His own footnotes add that venture-backed buyers are not the same as venture money paying, and that his circle of venture-backed firms is about 13% of US employment.