Competing ARR Estimates & Fuzzy Valuation Math in the Spotlight as Markets Get Edgy

1. Key Themes

Theme: Revenue-metric opacity is distorting AI valuations and moving public markets

Gross vs. net ARR confusion created a $20B "gap" that spooked AI stocks

The OpenAI/Anthropic ARR discrepancy was largely a definitional mismatch, not a collapse in business. The two companies calculate annualized revenue differently, and a headline built on that mismatch moved markets.

"AI stocks got hammered Thursday after the Financial Times reported that OpenAI was spinning its annualized revenue figure to investors in order to match Anthropic, and that its ARR was actually $20 billion below a widely cited $70 billion figure."

"OpenAI has long been complaining that its revenue figure looks feeble compared to Anthropic's because it's been using a net figure while Anthropic has been using gross (ie Anthropic's number includes the sales of models by partners like Amazon and Microsoft, even though the partners take a percentage of those sales)."

Public-company accounting is framed as the cure for "drip, drip" leaks

Dan Primack argues an IPO would force apples-to-apples reporting. This is an implicit signal that GAAP comparability becomes a catalyst once these labs go public.

"This OpenAI/FT stuff is the sort of thing that would be eliminated if both it and Anthropic go public. Apples/apples GAAP accounting for all to see. Rather than drip, drip cherry-picked financial leaks."

Tranched rounds make "headline valuation" a potentially misleading number

Multi-tranche financing is becoming normal, and the lack of per-tranche valuation disclosure creates risk for employees, founders, and investors.

"If multi-tranche rounds are the new standard, then sharing the valuation for each tranche should be the standard too—instead of misleading employees and the public into thinking a company…" (Natasha Mascarenhas)

"Deals at multiple valuations are becoming normalized, and the optics are one of the reasons... there are real risks for investors, employees, and founders alike if the true valuation figures aren't made clear."

Theme: Western open-source model startups are launching with a "sovereignty" pitch, but demand is not yet following

Three new Western open models launched in one week

Reflection (US), Mistral (France), and Aleph Alpha (Germany) all positioned against Chinese open models.

"American and European companies this week took aim at red-hot Chinese rivals such as DeepSeek with the announcements of three new open models: Reflection's Beam in the US, Mistral's Large 4 in France, and Aleph Alpha's Kolibri in Germany. All three startups were openly positioning their new models as the ones to pick if you're concerned about AI sovereignty."

Buyers mostly don't care about national origin, and Chinese models still lead usage

The sovereignty message is a tough sell without regulation. Chinese models hosted by US inference providers are the practical default.

"Investors we spoke with this week said few of their portfolio companies were concerned about the national origins of their open-source models at this point, other than those working with sensitive trade or government data. They'd prefer a Western model in theory, but many will gladly use a Chinese one hosted by a US inference provider like Baseten, Fireworks, or Modal."

"Top Chinese models like GLM 5.3, Kimi K3, and DeepSeek V4.1 remain the most popular on the cloud platforms Baseten and Fireworks."

Chinese capital formation shows no slowdown

"DeepSeek is close to raising $12 billion in funding co-led by Tencent and battery company Contemporary Amperex Technology Co. per Bloomberg."

Theme: The open-source LLM business model remains unproven

Training costs are high while pricing pressure favors the application layer

"Low prices are good for application-layer founders, of course, but LLM training remains very expensive."

Precedent exists for exits from the race

"Databricks CEO Ali Ghodsi told Newcomer this summer that by the time Databricks released its own open-weight foundation model in 2024, he was ready to move on from such an expensive and laborious task that could only produce a product that quickly became obsolete."

"Meta is still building open-source models, though it did pivot with its first-ever closed-source release this year, Muse Spark. The new open-source startups will need to be creative and nimble to avoid a similar change of course."

Theme: Venture dollars are at record highs but concentrated, with Q3 cooling

"It's been a year of massive, concentrated bets in venture, pushing dollar totals to new heights."

"Deal value for US startups has already surpassed 2021's record-breaking total by around 44%, according to the latest quarterly PitchBook-NVCA Venture Monitor report..."

(The headline of this section notes Q3 "slows to more normal pace"; the rest is paywalled.)

---

2. Contrarian Perspectives

Open source is the more secure option, and regulating it would create a frontier-lab monopoly

Bill Gurley argues against the prevailing safety framing. Open models get stress-tested by the community, and regulation targeting open source would entrench incumbents.

"Open models, he argues, are more secure than proprietary systems because the wider community stress-tests them — a concept that proved true in an earlier era of computing. Any regulation based on the potential dangers of open source will likely manufacture a monopoly for frontier labs, he says. 'IBM was not allowed to outlaw Dell.'"

Cheap open models are not a threat to frontier labs; the pie grows for everyone

Contrary to the "open source kills OpenAI/Anthropic" narrative, Menlo's Deedy Das sees complementarity.

"Cheaper, high-performance open models are great for the application-layer companies, and not such a big threat to the frontier model-makers."

"I find it all very strange that people pit them against each other because I think there's clearly a world where the pie grows faster and everyone grows."

Open source may not stay cheap

The assumed pricing advantage of open models may erode once customization and usage are factored in.

"Another investor we spoke with said implementing open-source models can get expensive quickly, given how much time and infrastructure spending is needed to customize them for specific tasks. Outsourcing it to an inference provider can be cheaper. But the total cost is still a function of how much you call on it, and the tokens add up."

---

3. Companies Identified

OpenAI

• Description: Frontier AI lab, currently in a funding round.
• Why mentioned: Center of the ARR controversy; net vs. gross revenue reporting.
• Quotes: "Bloomberg later reported that OpenAI is expecting to reach $70 billion annualized revenue by the end of the year — with the assumption that this is net and not gross."

Anthropic

• Description: Frontier AI lab and OpenAI's closest competitor.
• Why mentioned: Reports gross ARR including partner-cloud sales, which drives the comparison confusion.
• Quotes: "Investors and customers believe that Anthropic and OpenAI are running more or less neck-and-neck."

Reflection AI

• Description: US open-model startup; released Beam.
• Why mentioned: Leading the Western open-source push with efficiency claims.
• Quotes: "Beam's core claim is that it's even more efficient on inference than Chinese model maker Z.ai's highly touted GLM 5.2, though that isn't Z.ai's latest offering."

Mistral

• Description: French AI startup; released Large 4.
• Why mentioned: One of three Western open-model launches positioned on sovereignty.
• Quotes: "Mistral's Large 4 in France."

Aleph Alpha

• Description: German AI startup; released Kolibri.
• Why mentioned: Third Western open-model launch.
• Quotes: "Aleph Alpha's Kolibri in Germany."

DeepSeek

• Description: Chinese open-model developer.
• Why mentioned: Reportedly raising ~$12B, co-led by Tencent and CATL.
• Quotes: "DeepSeek is close to raising $12 billion in funding co-led by Tencent and battery company Contemporary Amperex Technology Co."

Z.ai (GLM), Kimi, DeepSeek models

• Description: Leading Chinese open models.
• Why mentioned: Still the most-used on inference platforms.
• Quotes: "Top Chinese models like GLM 5.3, Kimi K3, and DeepSeek V4.1 remain the most popular on the cloud platforms Baseten and Fireworks."

Baseten, Fireworks, Modal

• Description: US inference providers.
• Why mentioned: They enable startups to use Chinese open models while keeping hosting in the US.
• Quotes: "many will gladly use a Chinese one hosted by a US inference provider like Baseten, Fireworks, or Modal."

Manus

• Description: Singapore-based AI provider.
• Why mentioned: Benchmark-backed, acquired by Meta, deal unwound by Chinese regulators; founders bought it back and it raised fresh capital.
• Quotes: "Manus this week raised more than $500 million in fresh capital after its founders bought it back."

Meta

• Description: Tech giant.
• Why mentioned: Case study in strategic wavering on open source.
• Quotes: "it did pivot with its first-ever closed-source release this year, Muse Spark."

Databricks

• Description: Data/AI platform company.
• Why mentioned: Example of a company that exited the open-weight foundation model race.
• Quotes: "he was ready to move on from such an expensive and laborious task that could only produce a product that quickly became obsolete."

Benchmark

• Description: Venture firm.
• Why mentioned: Backed Manus and cashed out in the Meta deal.
• Quotes: "Benchmark even backed Singapore-based AI provider Manus and cashed in when it was bought by Meta."

Bessemer Venture Partners

• Description: Venture firm.
• Why mentioned: Partnering with Newcomer on the AI 100 list.
• Quotes: "We're thrilled to be producing this year's AI 100 list alongside Bessemer Venture Partners."

---

4. People Identified

Dan Primack

• Description: Journalist/commentator (Axios).
• Why mentioned: Sharp framing of the ARR issue and how going public would solve it.
• Quotes: "Apples/apples GAAP accounting for all to see. Rather than drip, drip cherry-picked financial leaks."

Natasha Mascarenhas

• Description: Bloomberg reporter.
• Why mentioned: Called for per-tranche valuation transparency.
• Quotes: "If multi-tranche rounds are the new standard, then sharing the valuation for each tranche should be the standard too."

Bill Gurley

• Description: Venture investor and vocal open-source advocate.
• Why mentioned: Substack defense of open source as security and business strategy.
• Quotes: "IBM was not allowed to outlaw Dell."

Deedy Das

• Description: Menlo Ventures partner.
• Why mentioned: Argues open models help app-layer companies without threatening frontier labs.
• Quotes: "I think there's clearly a world where the pie grows faster and everyone grows."

Ali Ghodsi

• Description: Databricks CEO.
• Why mentioned: Explained why Databricks left foundation-model building.
• Quotes: "ready to move on from such an expensive and laborious task that could only produce a product that quickly became obsolete."

---

5. Operating Insights

Normalize gross vs. net before comparing revenue metrics

When benchmarking against competitors, or reporting to investors, confirm the revenue definition. Partner-channel sales can inflate or deflate headline run-rate depending on convention.

"The pair calculate the figure in different ways, with Anthropic including the revenue from sales via cloud partners such as AWS and Google Cloud…"

Disclose per-tranche valuations if you raise in tranches

Transparency protects employees and avoids misleading the market about your true valuation.

"There are real risks for investors, employees, and founders alike if the true valuation figures aren't made clear."

Model total cost of ownership before betting on open-source models

Customization, infrastructure, and token volume can erase the sticker-price advantage. Consider managed inference hosting.

"Implementing open-source models can get expensive quickly... Outsourcing it to an inference provider can be cheaper. But the total cost is still a function of how much you call on it, and the tokens add up."

---

6. Overlooked Insights

A lawsuit over Groq's deal with Nvidia could threaten the "reverse acquihire" playbook

Mentioned only in the summary bullets, but potentially significant for founders and investors who rely on license-and-hire structures as an exit path.

"A new lawsuit over Groq's deal with Nvidia could throw a wrench in the reverse acquihire strategy."

Chinese regulators can unwind US-acquirer deals involving Chinese-origin founders

The Manus case shows cross-border M&A risk: even a completed sale to Meta was reversed, with founders ultimately buying the company back.

"Manus... was bought by Meta, though that deal was unwound by Chinese regulators. Manus this week raised more than $500 million in fresh capital after its founders bought it back."