---
title: "Baseten's $13 billion tranche prices a company that cuts its own rates every 15 days"
description: "Bloomberg says AI investors now split rounds into tranches so later money prints a higher price; Baseten's second tranche marked it at $13 billion in the same week it cut brand-new token prices by up to 72 percent."
publisher: "The Inference"
section: "Deals"
published: 2026-10-26T14:14:54.851Z
modified: 2026-10-26T14:14:54.851Z
canonical: https://theinference.org/article/baseten-s-13-billion-tranche-prices-a-company-that-cuts-its-own-rates-every-15-days
language: en
keywords: "Venture Capital, AI Inference, Valuations, Funding Rounds"
---

# Baseten's $13 billion tranche prices a company that cuts its own rates every 15 days

> Bloomberg says AI investors now split rounds into tranches so later money prints a higher price; Baseten's second tranche marked it at $13 billion in the same week it cut brand-new token prices by up to 72 percent.

## The markup and the cut landed the same week

[Baseten](https://theinference.org/markets/companies/baseten), the San Francisco company whose inference platform runs the models behind Cursor, [Notion](https://theinference.org/markets/companies/notion), [Abridge](https://theinference.org/markets/companies/abridge) and Clay, [closed $1.5 billion in two tranches](https://aiweekly.co/alerts/bloomberg-ai-vcs-split-rounds-into-tranches-to-raise-valuations-mid-deal): first at an $11 billion valuation, then at $13 billion, led by Altimeter, Conviction and Spark Capital [1][2][8]. The Wall Street Journal reported the dual-tier structure in June; PYMNTS relayed it under the headline "[as inference demand surges](https://www.pymnts.com/news/investment-tracker/2026/baseten-nears-1-5-billion-funding-round-as-inference-demand-surges/)" [3].

Inference, the business of running AI models and billing per token, is where the money is moving this cycle [3]. Anyone holding that exposure through secondaries, through perpetual futures on private AI names, or through listed stocks standing in for them is marked off prints like these.

The same week the tranche story ran, Baseten's own rate card moved the other way. Beam, a 501-billion-parameter open-weight reasoning model from the lab Reflection, joined Baseten's Model APIs catalog on October 17, 2026, at $1.30 per million input tokens and $4.30 output [4][5][6]. Four days later those rates fell to $0.40 and $1.20, up to 72 percent off, [now live on the pricing page](https://www.baseten.co/pricing/) [4][5][6]. It was Baseten's third cut and seventh price change in 90 days, roughly one every 15 days [4].

*Chart: **Reflection Beam's launch prices lasted four days** Range chart showing Reflection Beam list prices falling from $1.30 to $0.40 input, $0.13 to $0.04 cached input, and $4.30 to $1.20 output per million tokens*

*List price per million tokens on Baseten Model APIs: October 17, 2026, launch rates vs the October 21 cut. Sources: PricingSaaS diff; Baseten pricing page. [4][6]*

## Two prices for one company, weeks apart

Bloomberg's report describes the mechanism as deliberate: split the round so later money prints a higher valuation mid-deal [1]. Ineffable Intelligence, founded by former [Google DeepMind](https://theinference.org/markets/companies/google-deepmind) scientist David Silver, raised $11 million at about $55 million pre-money [2]. A month later it took $1.1 billion at $4 billion pre-money, more than 70 times the price, with Sequoia buying at both levels [2].

[Etched](https://theinference.org/markets/companies/etched), which builds inference chips, printed $10.3 billion in July and $21 billion in August, led by [Jane Street](https://theinference.org/markets/companies/jane-street), the quant fund that tested its hardware and now runs its own rack [7].

The top number is the one that gets reported, and the one later buyers are asked to pay. "Everyone pretends they only did the higher valuation," posted Brendan Foody, chief executive of the AI-hiring startup [Mercor](https://theinference.org/markets/companies/mercor), in June, adding that every top firm does it [2]. The sharper point is Amplify partner Sarah Catanzaro's: the company's real worth is the blended valuation, a weighted average that usually goes unreported [2].

*Chart: **One round, two prices: the second tranche prints higher** Range chart of reported valuations by tranche: Baseten $11B to $13B, Etched $10.3B to $21B, Ineffable $0.055B to $4B*

*Reported valuation of each tranche, billions of dollars. Baseten Series F per Bloomberg via AI Weekly and Forbes; Etched per TechCrunch; Ineffable pre-money per Forbes. [1][2][7]*

## The denominator reprices every two weeks

Bloomberg credits a 20-fold revenue jump for Baseten's step-up [1]. But inference revenue is price multiplied by tokens, and Baseten resets price on a clock: on September 22 it cut [DeepSeek](https://theinference.org/markets/companies/deepseek) cached input 77 percent, to $0.007 per million tokens [4].

The counterpoint holds partly. Volume has outrun price so far: annualized revenue ran at roughly $200 million entering the first quarter and $600 million leaving it, per June reporting relayed by PYMNTS [3]. Cheap tokens are also the product; when Baseten [raised $300 million at $5 billion in January](https://www.businesswire.com/news/home/20260123035833/en/Baseten-Raises-%24300M-at-a-%245B-Valuation-to-Power-a-Multi-Model-Future), anchored by [NVIDIA](https://theinference.org/markets/companies/nvidia), Shiv Rao, chief executive of the medical AI note-taker Abridge, credited it with "constantly finding ways to lower our costs" [8].

*Chart: **Baseten's printed valuation has climbed 2.6 times since January** Column chart showing Baseten's reported valuation rising from $5 billion in January 2026 to $11 billion and then $13 billion across the two tranches of its latest round*

*Reported valuation marks in 2026, billions of dollars: the January round, then the first and second tranches of the Series F reported this month. Sources: Business Wire; Bloomberg via AI Weekly; Forbes. [8][1][2]*

[The Inference traced in September](https://theinference.org/article/baseten-locked-three-years-of-gpus-whitefiber-s-convert-holders-carry-the-bill) who funds the GPU commitments underneath. None of it prints the blend.

## Who pays the printed number

Employees hired after a tranche strike options near the top mark, "taking more of the risk and capturing less of the upside," Foundation Capital's Jaya Gupta says [2]. Secondary buyers and holders of perps or listed proxies inherit the same headline. The working rule: discount marks built in tranches until token price lists stop falling. Two tells decide it: Baseten's rate card, on a 15-day cadence since July [4], and whether the next tranche print arrives with its blended valuation attached.

## Takeaway

Baseten's second tranche printed $13 billion the same week it cut brand-new token prices up to 72 percent; the blended valuation, which never gets reported, is the number that matters.

## Sources

1. [AI Weekly, Bloomberg: AI VCs Split Rounds Into Tranches to Raise Valuations Mid-Deal, October 2026](https://aiweekly.co/alerts/bloomberg-ai-vcs-split-rounds-into-tranches-to-raise-valuations-mid-deal)
2. [Forbes, Rashi Shrivastava and Iain Martin: AI Startups With No Revenue Are Using This Tactic To Supersize Their Valuations, 25 June 2026](https://www.forbes.com/sites/rashishrivastava/2026/06/25/ai-startups-with-no-revenue-are-using-this-tactic-to-supersize-their-valuations/)
3. [PYMNTS, Baseten Nears $1.5 Billion Funding Round as Inference Demand Surges, 18 June 2026](https://www.pymnts.com/news/investment-tracker/2026/baseten-nears-1-5-billion-funding-round-as-inference-demand-surges/)
4. [PricingSaaS Pulse, Baseten cuts Reflection Beam rates up to 72%, four days after launch, 21 October 2026](https://pricingsaas.com/news/article/baseten-slashes-brand-new-reflection-beam-rates-up-to-72-days-after-launch)
5. [Baseten blog, Beam: Efficient reasoning for coding and agentic workloads, October 2026](https://www.baseten.co/blog/)
6. [Baseten, Cloud Pricing page, fetched 26 October 2026](https://www.baseten.co/pricing/)
7. [TechCrunch, Julie Bort: Etched's valuation doubles to $21B in a month, 18 August 2026](https://techcrunch.com/2026/08/18/etcheds-valuation-doubles-to-21b-in-a-month/)
8. [Business Wire, Baseten Raises $300M at a $5B Valuation to Power a Multi-Model Future, 22 January 2026](https://www.businesswire.com/news/home/20260123035833/en/Baseten-Raises-%24300M-at-a-%245B-Valuation-to-Power-a-Multi-Model-Future)


---

Baseten's $13 billion tranche prices a company that cuts its own rates every 15 days — The Inference. Canonical: https://theinference.org/article/baseten-s-13-billion-tranche-prices-a-company-that-cuts-its-own-rates-every-15-days