OpenAI's missing $20B | PitchBook News Summary

1. Key Themes

Theme: OpenAI's growth is accelerating, but so are its losses, and the valuation math is under pressure

The $20B ARR discrepancy is an accounting artifact, not lost revenue

The headline run rate was revised down by nearly a third, which the author attributes to investors adopting a different accounting convention.

"For nine days, markets traded on a $70 billion run rate for OpenAI. OpenAI has since told investors its net run rate neared $50 billion at the end of September."

"The $20 billion difference appears to be an accounting quirk—the $70 billion figure was reportedly the result of its investors' efforts to match Anthropic's accounting method."

Revenue is accelerating sharply

"OpenAI's run rate rose from an estimated $29 billion to $50 billion in Q3, putting Q3 revenue around $10 billion, roughly 50% above Q2."

Losses are scaling just as fast

"Operating loss, including stock-based compensation, reportedly hit $12.3 billion in Q2, or $1.84 per revenue dollar. First-half operating losses of $21.6 billion exceed all of 2025's $20.9 billion."

Valuation multiples have compressed, and a $1.4T raise would re-expand them

"The March round's $852 billion price was 34x OpenAI's run rate then and is 17x today's run rate. A reported raise at $1.4 trillion would be 28x. Anthropic's expected $2 trillion IPO target is 31x its gross July run rate, with Q2 revenue above $11.5 billion and an adjusted profit."

Theme: AI is splitting the market. Capital flows to the physical economy while software re-rates downward

AI infrastructure spending is lifting energy and B2B deal activity

"2026 has been the year of the AI trade, with the physical economy winning out in global M&A. While hyperscaler stocks soar to record highs, the rest of tech is on the outside looking in, as capital flows to the parts of the economy building AI infrastructure: services, energy and materials."

"Through three quarters, energy M&A activity is already at 138% of full-year 2025 levels, with deals getting larger and more expensive."

"B2B, a broad sector encompassing industries such as professional services, construction and manufacturing, is on pace for its third year above $1 trillion in deal value, with multiples rising by 1.2 turns since 2024."

IT multiples have fully de-rated

"Meanwhile, IT multiples have re-rated, down 3.7 turns since 2024 as the tech premium has been completely eroded."

Theme: Compute capacity scarcity is driving massive capex, but public-market appetite is uneven

Firmus's postponed IPO shows the strain on the infrastructure build-out

"Australian data center operator Firmus postponed its IPO, which was set to raise up to $5.5 billion. Firmus is among the neoclouds and hyperscalers with massive capex plans to address the compute capacity shortage."

"Australia-based data center operator Firmus, which is backed by Nvidia, postponed its IPO and is considering another private funding round instead, Bloomberg reported."

Venture dollars are flowing to the physical stack behind AI and frontier hardware

"Veir, which specializes in superconducting power delivery systems for data centers, secured a $110 million Series C led by Matter Venture Partners and Tyche Partners."

"Quantum computer builder Oratomic raised a $475 million Series B led by Arch Venture Partners, Spark Capital, Khosla Ventures, Index Ventures, General Catalyst and Bezos Expeditions, at a valuation of $5.4 billion."

2. Contrarian Perspectives

The next valuation correction will hit OpenAI's rumored $1.4T round, not the broader AI trade The author argues the market has already partially corrected on OpenAI's revised numbers, and that the pending raise is priced richer relative to run rate than the March round. Evidence: the March price was 17x today's run rate, while $1.4T would be 28x, a sharp re-expansion despite widening losses.

"The market has corrected since OpenAI's new disclosure. We expect the next correction to hit the $1.4 trillion valuation, unless Q3 results show losses narrowing and growth accelerating."

The tech selloff is indiscriminate, which creates an opportunity for PE take-privates Rather than treating all software as AI victims, the author argues specialist sponsors can separate truly threatened companies from those caught in the selloff. Evidence: Accelerant is being taken private below its IPO price only a year after listing.

"A broad decline in multiples sets up favorably for sponsors to re-enter the market, as we saw with Thoma Bravo's take-private of insurance tech platform Accelerant only one year after its IPO. The deal, set to close in 2027, is priced below Accelerant's IPO price."

"We expect more take-privates from tech-focused sponsors over the next six months. A cheaper market lets specialists do what they do best: separate companies genuinely threatened by AI from those caught in the selloff."

3. Companies Identified

OpenAI

Description: Leading AI lab and ChatGPT maker.
Why mentioned: Central case study of the run-rate restatement, accelerating revenue, and widening losses.
Quote: "OpenAI's run rate rose from an estimated $29 billion to $50 billion in Q3, putting Q3 revenue around $10 billion, roughly 50% above Q2."

Anthropic

Description: AI lab and OpenAI competitor.
Why mentioned: Its accounting method was the source of OpenAI's inflated run-rate figure, and it is headed for a very large IPO with an adjusted profit.
Quote: "Anthropic's expected $2 trillion IPO target is 31x its gross July run rate, with Q2 revenue above $11.5 billion and an adjusted profit."

Firmus

Description: Australian data center operator backed by Nvidia.
Why mentioned: Postponed a $5.5B IPO, a signal of strain in capital-intensive AI infrastructure.
Quote: "Firmus, which is backed by Nvidia, postponed its IPO and is considering another private funding round instead, Bloomberg reported."

Thoma Bravo / Accelerant

Description: Thoma Bravo is a tech-focused PE sponsor. Accelerant is an insurance tech platform.
Why mentioned: The take-private is the case study for sponsors re-entering a de-rated tech market.
Quote: "Thoma Bravo's take-private of insurance tech platform Accelerant only one year after its IPO. The deal, set to close in 2027, is priced below Accelerant's IPO price."

Oratomic

Description: Quantum computer builder.
Why mentioned: Largest VC round of the day.
Quote: "raised a $475 million Series B led by Arch Venture Partners, Spark Capital, Khosla Ventures, Index Ventures, General Catalyst and Bezos Expeditions, at a valuation of $5.4 billion."

Atomic Machines

Description: Developer of digital manufacturing systems for building small devices.
Why mentioned: Emerged from six years in stealth with a large launch round.
Quote: "After six years in stealth, Atomic Machines... launched with $250 million from investors including OneIM, Gigafund and XTX Ventures."

Arena Intelligence

Description: Startup that evaluates AI models and ranks them on a leaderboard.
Why mentioned: Raised at a $3.1B valuation, signaling demand for model evaluation.
Quote: "raised a $200 million Series B led by Lightspeed and Khosla Ventures at a $3.1 billion valuation."

Veir

Description: Superconducting power delivery for data centers.
Why mentioned: Infrastructure bottleneck play (power).
Quote: "secured a $110 million Series C led by Matter Venture Partners and Tyche Partners."

Hone

Description: Developer of AI agents that perform long-term business tasks.
Why mentioned: Large seed round at a notable valuation, led by top-tier firms.
Quote: "raised a $60 million seed round led by Benchmark and Index Ventures at a $285 million valuation."

Universal Quantum

Description: UK-based fault-tolerant quantum computing startup.
Why mentioned: Quantum momentum alongside Oratomic.
Quote: "secured a $100 million Series A co-led by DCVC and Firgun Ventures."

Spectrum Equity

Description: Growth equity investor.
Why mentioned: Closed its 11th fund focused on software, AI and data services.
Quote: "closed its 11th fund on $2.5 billion, which is focused on software, AI and data services investments."

Iambic Therapeutics

Description: Nvidia-backed biotech using AI in drug development.
Why mentioned: AI-drug-discovery IPO candidate.
Quote: "an Nvidia-backed biotech using AI in drug development, is looking to raise up to $159.4 million in its US IPO."

Avaada Electro

Description: Brookfield Renewable Partners-backed solar manufacturer.
Why mentioned: Large India IPO tied to the energy build-out.
Quote: "is planning to launch its India IPO next month, and could raise about 76 billion rupees ($786 million)."

Bridgepoint Group

Description: London-based PE firm.
Why mentioned: Hit its fundraising target early, notable in a tough fundraising environment.
Quote: "reached its €28 billion fundraising target for the 2024-26 cycle one quarter ahead of schedule."

4. People Identified

Harrison Rolfes

Description: Senior Research Analyst, Private Company Coverage at PitchBook.
Why mentioned: Author of the OpenAI run-rate analysis.
Quote: "The new figure reveals an acceleration of both revenue and losses."

Steven Buibish, CFA

Description: Director, US PE Research at PitchBook.
Why mentioned: Author of the AI market-split and PE opportunity analysis.
Quote: "We expect more take-privates from tech-focused sponsors over the next six months."

Zach Frankel

Description: 37-year-old investor profiled by The Information.
Why mentioned: Early check-writer in Ramp and Cognition.
Quote: "Zach Frankel, 37, wrote the first checks in expense software maker Ramp and AI coding company Cognition."

5. Operating Insights

Normalize revenue metrics before comparing or benchmarking. The $20B swing came entirely from differing run-rate methodology (net vs. gross). Operators and investors should align definitions before comparing growth multiples.

"The $70 billion figure was reportedly the result of its investors' efforts to match Anthropic's accounting method."

Track loss per revenue dollar, not just growth. The author frames OpenAI's economics as unit-loss intensity, a useful lens for any capital-hungry company.

"Operating loss, including stock-based compensation, reportedly hit $12.3 billion in Q2, or $1.84 per revenue dollar."

Software operators should expect take-private interest and position their AI defensibility. With sponsors screening for companies "genuinely threatened by AI" versus those "caught in the selloff," demonstrating AI resilience may determine who gets bought and at what price.

"A cheaper market lets specialists do what they do best: separate companies genuinely threatened by AI from those caught in the selloff."

6. Overlooked Insights

Firmus is considering going back to private markets rather than listing. A postponed multi-billion IPO with a fallback to private funding suggests public investors may be less willing to underwrite neocloud capex than private ones.

"postponed its IPO and is considering another private funding round instead, Bloomberg reported."

Anthropic's profitability contrast with OpenAI. The newsletter notes in passing that Anthropic has an adjusted profit at over $11.5B quarterly revenue, while OpenAI's losses are widening, a potentially important competitive divergence.

"with Q2 revenue above $11.5 billion and an adjusted profit."