The Neolab List:
106 AI Research Labs, Ranked by Valuation

Why this call: Corporate SoftBank IPO vehicle for robots that build data centers; not a researcher-led lab.

Known for: Planned SoftBank robotics-AI company; possible ABB robotics asset inclusion

In more detail: A planned SoftBank company that would use AI and autonomous robots to build and run data centers in the U.S. It has no public product yet.

Revenue: Unknown (not yet operating)

Backers: SoftBank; potentially ABB robotics assets; DigitalBridge/Stargate ecosystem links

Valuation notes: Reuters/FT (Apr 29, 2026): SoftBank plans to list Roze in the U.S., executives seeking ~$100B. Reuters (May 26): SoftBank hired Goldman Sachs, JPMorgan, Mizuho and Morgan Stanley; listing possibly alongside SB Energy. No filing as of Oct 24, 2026. $100B is an IPO target, not a transaction.

Sources: reuters.com · finance.yahoo.com

Why this call: Elite researchers, no revenue, very compute-heavy, software only, record early raise.

Known for: Google's most senior researchers: Dean (former Chief Scientist), Ghemawat (MapReduce/Bigtable), Vinyals (AlphaStar, Gemini), Le (seq2seq)

In more detail: Builds AI systems that automate research by proposing, running and judging thousands of experiments at once. It starts with machine learning research, with medicine and cybersecurity to follow.

Revenue: No

Backers: Radical Ventures, Khosla Ventures (co-leads), Lightspeed, Kleiner Perkins, Doerr Capital, Alphabet

Valuation notes: Founded Aug 5, 2026. PitchBook logs $1B at ~$10B post (Sep 14, 2026; valuation estimated); Business Insider first reported it in negotiation. Business Insider (Sep 12): talks at ~$50B, rumored only.

Sources: thenextweb.com · aiweekly.co

Why this call: Researchers, compute and capital all pass; $100M+ run-rate now weakens the low-revenue test.

Known for: Huge seed; ex-OpenAI CTO; PyTorch founder; Tinker

In more detail: Trains multimodal AI models and tools to customize them. Its Tinker platform lets developers fine-tune models, and its first model, Inkling, is open-weight.

Revenue: Yes (run-rate >$100M per TechCrunch's source, Sep 2026; some reports >$200M)

Backers: a16z (lead), Accel, Nvidia, AMD, Jane Street, others; Accel (in talks to lead), Nvidia (~$2.5B in talks)

Valuation notes: Talks (Sep 3–5, 2026): Accel to lead $5–6B at ≥$40B pre-money, Nvidia ~half; came after a $50–60B attempt fell apart and three co-founders (incl. Lilian Weng) returned to OpenAI. Last closed: $2B seed at $12B post (Jul 2025). Shipped Inkling, a 975B-parameter open-weight model, Jul 2026.

Sources: reuters.com · reuters.com · techcrunch.com · techmeme.com

Why this call: Researcher-heavy, pre-revenue, compute-hungry physical-AI lab; billionaire-led with M&A ambitions, so not a typical startup.

Known for: Jeff Bezos' new AI operating role; physical economy focus

In more detail: Builds AI that helps engineer and manufacture physical products in fields like aerospace, cars and chips. It has not released a public product yet.

Revenue: No

Backers: Jeff Bezos; JPMorgan and BlackRock reportedly participating; other investors undisclosed

Valuation notes: Bloomberg (Apr 23, 2026): closed a ~$10B round at ~$38B (no lead; JPMorgan, BlackRock participated) on top of the $6.2B launch raise; >$16B raised in total. Sheet's $41B kept (tracker figure, e.g. neolabs.fyi/Juniper) but not found in primary reporting. Plans a separate vehicle to acquire manufacturing companies.

Sources: reuters.com · reuters.com · bloomberg.com

Why this call: Research lab with huge raises and heavy compute, but Alphabet-controlled with a drug pipeline.

Known for: DeepMind spinout; AlphaFold 3 co-developer; IsoDDE drug design engine

In more detail: Designs drugs with AI from target to candidate, building on AlphaFold 3. Its IsoDDE design engine drives its own pipeline and work with partners Lilly, Novartis and J&J.

Revenue: Unknown (~$82.5M upfront from Lilly and Novartis partnerships)

Backers: Thrive Capital (Series B lead), Alphabet, GV, CapitalG, MGX, Temasek, UK Sovereign AI Fund

Valuation notes: Bloomberg (Oct 8, 2026): early talks at ≥$40B (one source says up to $50B). Last closed: $2.1B Series B (May 12, 2026), valuation undisclosed; Series A (~$600M, Mar 2025) at $1.79B post. Sheet's '$15–20B' has no source.

Sources: bloomberg.com · isomorphiclabs.com

Why this call: Researcher-led, pre-revenue, compute-hungry, model-only, enormous capital.

Known for: OpenAI founder; huge valuation; secrecy

In more detail: Researches how to build safe superintelligence and says it will ship no products before reaching that goal. It has not released any model publicly.

Revenue: No (no product released)

Backers: Greenoaks (lead), Sequoia, a16z, Lightspeed, Alphabet, Nvidia, others; Nvidia ($5B, Jul 2026)

Valuation notes: Last priced: ~$32B post (Greenoaks-led, 2025). Bloomberg/Reuters (Jul 27, 2026): Nvidia committed ~$5B with access to Vera Rubin chips; no new valuation disclosed, so true value is likely higher. ~$7B+ raised before Nvidia.

Sources: reuters.com · bloomberg.com · techcrunch.com

Why this call: Ex-DeepMind open-weight frontier lab, compute-hungry and heavily funded; exit talks pending.

Known for: American open-source models by ex-DeepMind team

In more detail: Builds open-weight frontier language models in the U.S. as an alternative to Chinese open models. Its first, Beam, is a large reasoning model announced in October 2026.

Revenue: Unknown (no figure found)

Backers: Nvidia, Eric Schmidt, Citi, 1789 Capital, Lightspeed, Sequoia, DST Global, others

Valuation notes: WSJ (Mar 2026) / CEO on CNBC (Apr): ~$2.5B raise at $25B pre-money; full close unclear. FT (Oct 10, 2026): Nvidia in early talks to acquire, acqui-hire/license, invest more, or supply compute, no terms; ~$30B price is podcast chatter only. Nvidia has invested $800M. Released Beam open-weight model Oct 5, 2026.

Sources: techcrunch.com · reuters.com · thenextweb.com · news.bloomberglaw.com

Why this call: Established, large revenue.

Known for: Europe's OpenAI answer; open-weight models; Le Chat; ASML-backed

In more detail: Builds open-weight and commercial language models plus the Le Chat assistant, and sells AI and compute that European firms and governments can control. Customers include ASML, CMA CGM and Stellantis.

Revenue: Yes

Backers: ASML (lead, ~11%), Nvidia, a16z, General Catalyst, DST, Index, Lightspeed, Bpifrance; Samsung Electronics (Series D lead), Scaleup Europe Fund (EQT), PSG Equity

Valuation notes: Closed €3B Series D at >€21B post (~$24B) on Sep 8, 2026, per counsel Orrick; replaces June talks figure. >$400M ARR, targeting $1B by end-2026. Also $830M debt (Mar 2026).

Sources: orrick.com · sifted.eu

Why this call: Scaled product company with ~$500–600M revenue and ~800 staff.

Known for: Leading AI voice platform; founders ex-Palantir (Staniszewski) and ex-Google (Dabkowski); eyeing 2028 IPO

In more detail: Builds voice AI for lifelike speech, transcription, dubbing and voice agents, sold as apps and an API. Customers include Deutsche Telekom and Revolut.

Revenue: Yes (~$600M TTM per PitchBook; ~$500M ARR early 2026)

Backers: Sequoia (Series D lead), a16z, ICONIQ, BlackRock, Nvidia, Wellington, T. Rowe Price, EQT

Valuation notes: $22B post via a ~$300M secondary share sale (Wellington, T. Rowe Price, EQT), completed Oct 1, 2026 per PitchBook; Bloomberg reported the tender talks in July. Last primary: $500M Series D at $11B (Feb 2026, Sequoia). A Series E with EQT is rumored.

Sources: dealroom.co · bloomberg.com

Why this call: Established enterprise company with large revenue.

Known for: "Attention Is All You Need" co-author; sovereign/enterprise AI; Command models

In more detail: Builds secure language models and an AI agent platform, North, for businesses and governments that must keep data private. Customers include RBC, Oracle and Fujitsu. It is merging with Aleph Alpha.

Revenue: Yes

Backers: Radical Ventures, Inovia, Nvidia, AMD, Salesforce, PSP, Cisco, Fujitsu, BDC; Schwarz Group (€500M planned), Canadian-government-backed consortium (up to $3B, talks)

Valuation notes: Last priced: $7B ($600M Series D, Sep 2025). Definitive merger with Aleph Alpha signed Sep 16, 2026 (combined company keeps Cohere brand; Berlin + Toronto HQs), reported at ~$20B; awaiting regulators. Talks to raise up to $3B at as much as $20B (Sep 11). ~$240M 2025 revenue.

Sources: siliconangle.com · bloomberg.com

Why this call: Researcher-led robot foundation models, compute-heavy; the Zebra acquisition pushes it toward operations and revenue.

Known for: Robotic foundation models by CMU professors

In more detail: Builds Skild Brain, a general AI brain that can control many kinds of robots in factories, warehouses and inspection jobs. Partners include Nvidia and ABB.

Revenue: Unknown (no figure found)

Backers: SoftBank, Nvidia/NVentures, Amazon, Lightspeed, LG VC, Samsung, Jeff Bezos, others; Salesforce Ventures, Macquarie, 1789 Capital

Valuation notes: Closed $1.4B Series C led by SoftBank at >$14B (Jan 14, 2026), confirmed by TechCrunch, SiliconANGLE, Bloomberg. No new round since. Skild bought Zebra Technologies' robotics automation unit (Apr 2026).

Sources: skild.ai · techcrunch.com

Why this call: Coding-model lab training its own models with huge raises and heavy compute; Nvidia licensed its tech and hired 109 staff in Aug 2026.

Known for: Raised a lot for coding models; GitHub Copilot lineage

In more detail: Builds AI coding models and agents that companies and governments can run inside their own secure systems. Its Laguna family includes open-weight models. Nvidia licensed its technology in 2026.

Revenue: Unknown

Backers: Bain Capital Ventures, Nvidia, eBay Ventures, others; Citi Ventures, Redpoint, Adams Street

Valuation notes: Aug 2026: Nvidia invested ~$1B at $12B pre and paid ~$6B for a non-exclusive tech license; 109 employees moved to Nvidia (Groq-style structure). Co-founders keep running Poolside independently. Effectively a partial exit.

Sources: reuters.com · forkast.news

Why this call: Researcher-founded robot foundation model lab, pre-revenue, compute-heavy, model-focused.

Known for: Robotic foundation models by Stanford / Google robotics researchers

In more detail: Trains general-purpose AI models that let robots learn many tasks from data gathered across different machines. Its open-source models, such as π0 and π0.5, are now offered to developers through an API.

Revenue: No

Backers: CapitalG (lead), Lux, Thrive, Jeff Bezos, Index, T. Rowe Price; Founders Fund and Lightspeed (in talks)

Valuation notes: Last closed: $600M at $5.6B (Nov 2025, CapitalG-led). Bloomberg (Mar 27, 2026): talks for ~$1B at >$11B; no confirmed close as of Oct 2026. API beta launched Oct 13, 2026. Denied a rumor that Anthropic was buying it (Jul 2026).

Sources: bloomberg.com · outsetcapital.com · bloomberg.com

Why this call: Research-driven with giant raises, but wet-lab heavy rather than purely software.

Known for: Huge Series A

In more detail: Builds AI Science Factories, automated labs where AI designs and runs experiments in biology, chemistry and materials science. It plans to sell this discovery platform to outside companies.

Revenue: Unknown

Backers: Flagship Pioneering, General Catalyst, ADIA, NVentures, CalPERS (Series B anchor), Analog Devices, IQT, ARK, March Capital

Valuation notes: Bloomberg (Jun 3, 2026): talks for ~$2B Series B at ~$8.5B, anchored by CalPERS and NVentures. Harmonic shows $2.785B raised with last funding Jul 31, 2026 (implies close); PitchBook still 'rumor'. Last confirmed: $350M Series A at ~$1.3B (2025). ~415 staff.

Sources: news.bloomberglaw.com · sacra.com

Why this call: Profitable, bootstrapped, product-led.

Known for: Bootstrapped; zero VC; ~$4.7M revenue/employee; Discord-native

In more detail: Makes AI image and video generators sold by subscription, with its V8 image model released in 2026. Meta licenses its technology, and it is also building a full-body ultrasound scanner.

Revenue: Yes

Backers: None, fully bootstrapped, no outside funding

Valuation notes: No priced round; ~$10B is an analyst estimate (~$500M ARR at 20x). Still bootstrapped and profitable. Jun 18, 2026: unveiled the 'Midjourney Scanner', a full-body ultrasound machine (health pivot), aiming for 50,000 units; no funding disclosed.

Sources: news.bloomberglaw.com

Why this call: Researcher-founded, compute-heavy world-model lab with little disclosed revenue; AMD deal pending.

Known for: Fei-Fei Li (ImageNet, Stanford HAI); Marble 3D world-generation product

In more detail: Builds world models that turn text, images or video into editable 3D worlds. Its app and API, Marble, came first, and its newer Atlas model is in early access. Being acquired by AMD.

Revenue: Unknown ('generating revenue', no figure)

Backers: Nvidia, Autodesk, Emerson Collective, Fidelity (Series C leads), Radical Ventures, a16z, NEA (Series B leads), AMD Ventures, Temasek, Samsung, Cisco

Valuation notes: AMD signed a definitive agreement to buy 100% for ~$8.2B in stock (announced Sep 28, 2026; pending regulators); Fei-Fei Li becomes AMD EVP and Chief Scientist. Before: $1B Series C at $5.4B post (Feb 20, 2026); a Greycroft secondary (Mar 2026) implied $7.0B.

Sources: techcrunch.com · nextplatform.com

Why this call: Researcher-led and pre-revenue; wet labs make it less purely software.

Known for: ex-OpenAI VP and ex-DeepMind materials lead

In more detail: Pairs AI models with robotic labs that run experiments on their own to discover new materials, starting with the search for better superconductors.

Revenue: Unknown (no evidence of revenue)

Backers: a16z, DST, Nvidia, Accel, Elad Gil, Jeff Dean, Eric Schmidt, Jeff Bezos; Felicis, AMP (leading 2026 talks)

Valuation notes: Seed: $300M at ~$1.3B (Sep 2025). Talks: Bloomberg ~$7B (Mar 25, 2026), Forbes $500M+ at $7.5B led by AMP (May 7, 2026). No close found.

Sources: techcrunch.com · forbes.com

Why this call: New architecture and a giant early raise, but already a commercial product.

Known for: Almeida ex-OpenAI (InstructGPT/ChatGPT/GPT-4) and ex-Google; Jev, a non-autoregressive 'System One' decision model

In more detail: Builds AI models that return decisions instead of text, such as a choice, score or yes/no answer with a confidence level. Its first model, Jev, handles fast, cheap sorting and routing inside software.

Revenue: Unknown (claims 1M+ users, ~1/3 of Fortune 500)

Backers: a16z (Series A lead), Sequoia, DCVC (seed lead), Horizon Ventures

Valuation notes: Solid: ~$870M Series A led by a16z at $7.5B post (Oct 9, 2026; Bloomberg, PitchBook, Harmonic). Earlier The Information reported talks at ≥$10B. Seed: ~$40M at ~$200M (Sep 2026). ~$914M raised in total.

Sources: news.bgov.com · cryptobriefing.com

Why this call: Researcher-founded and compute-heavy, but reportedly profitable with >$10M revenue.

Known for: Mirage + Oasis world models

In more detail: Builds real-time video and world models that create or change video live as you interact. Its models include Oasis, a playable AI game world, Mirage, and Lucy for live video editing.

Revenue: Yes* ($20M TTM 2024 and 'profitable' per PitchBook)

Backers: Radical Ventures (Series B lead), Nvidia, Toyota Ventures, Sequoia, Benchmark, Zeev Ventures, Aleph

Valuation notes: WSJ (Oct 11, 2026): Amazon in talks to buy at ~$7B (PitchBook 'rumor'; Amazon declined comment). Earlier Anthropic discussed ~$6B all-stock (Aug 2026) and walked away in Sep. Last closed: $300M Series B at ~$4B post (May 20, 2026). ~$450M raised in total.

Sources: thenextweb.com

Why this call: Researcher team, pre-revenue, compute-heavy agents and world models trained on gameplay video, huge early raises.

Known for: Massive gameplay-video dataset

In more detail: Trains AI agents that learn to act in games and the real world from billions of gameplay clips from the Medal app. It sells these agent models, with robotics among its targets.

Revenue: No

Backers: Khosla Ventures, General Catalyst, Raine; Valor, Atreides (Sep 2026 round)

Valuation notes: Two 2026 rounds: $320M Series A at $2.3B led by Khosla (Jun 2026), then $220M at $6.2B with Valor and Atreides (Sep 29–30, 2026) per neolab trackers and Crunchbase/TechCrunch coverage; ~$674M raised in total (Harmonic). Seed was $133.7M (Oct 2025).

Sources: linkedin.com · sources.news · juniper.xyz · cleverhack.com

Why this call: Models plus consumer hardware, founder-led rather than researcher-led, already shipping product.

Known for: Figure AI / Archer founder uses $100M personal funds

In more detail: Builds a personal AI assistant that anticipates what you need, powered by its own multimodal models, with dedicated hardware planned. Hark Pro launched on web, iOS and Android in October 2026.

Revenue: Unknown (first product 'Hark Handoff' launched Aug 2026)

Backers: Brett Adcock (~$100M personal); $700M Series A investors not fully disclosed

Valuation notes: TechCrunch (May 21, 2026): closed $700M Series A at $6B, replacing the earlier 'personal funding only' picture.

Sources: observer.com · techcrunch.com

Why this call: Research lab on data-efficient learning and GPU megakernels, pre-revenue.

Known for: Young research-first team; contrarian non-scaling thesis

In more detail: Researches AI models that learn from far less data than today's systems, taking cues from how the brain learns. It has no product yet.

Revenue: No

Backers: GV, Sequoia; others not fully disclosed; Index and Lux (Series A co-leads), Kleiner Perkins, Valor, Atreides, A*, Nvidia

Valuation notes: TechCrunch/Bloomberg (Oct 21, 2026): closed $1B Series A at ~$6B (pre/post not stated), ~4x the $1.5B seed valuation ($180M seed, Jan 2026).

Sources: gv.com · sequoiacap.com · techcrunch.com

Why this call: Revenue-scale consumer product, not a research lab.

Known for: Leading AI song generator; founders from Kensho; Warner Music Group partnership (Nov 2025)

In more detail: Makes an app that creates full songs with vocals and lyrics from a text prompt. Its current model is v5, and new models licensed through a Warner Music deal are planned.

Revenue: Yes (~$300M TTM per PitchBook)

Backers: BOND (Series D lead), Menlo Ventures (Series C lead), IVP, Forerunner, USV, Lightspeed, Matrix, Nat Friedman, Daniel Gross

Valuation notes: Solid: $400M+ Series D led by BOND at $5.4B post (Jun 3, 2026) per MusicTech and PitchBook. Series C: $250M at $2.45B (Nov 2025).

Sources: musictech.com

Why this call: Established with meaningful revenue, though pivoting toward world models.

Known for: Physics-aware AI video; pivoting into world models; Adobe partnership

In more detail: Builds AI models that generate and edit video, and is expanding into world models that simulate interactive environments. Its models include Gen-4.5 and GWM-1, and partners include Lionsgate.

Revenue: Yes (~$180M ARR est.; +$40M ARR in Q2 2026)

Backers: General Atlantic (lead), Nvidia, Fidelity, AllianceBernstein, Adobe Ventures, Felicis, AMD Ventures

Valuation notes: $315M Series E at $5.3B (Feb 10, 2026), ~doubling prior. Has meaningful revenue (media/enterprise), so fails low-revenue screen. Established, not a neolab.

Sources: aiweekly.co

Why this call: Pre-revenue RL research lab with a record seed.

Known for: AlphaGo/AlphaZero RL lead; Europe's largest seed round

In more detail: Building a reinforcement learning superlearner, an AI that discovers knowledge from its own experience instead of learning from human data. No product or model has been released yet.

Revenue: No

Backers: Sequoia, Lightspeed, Nvidia, Google, DST Global, Index, EQT, BOND, UK Sovereign AI Fund, British Business Bank, others

Valuation notes: Reuters/Bloomberg/company: $1.1B seed at $5.1B (Apr 27, 2026), Sequoia and Lightspeed co-led. Nvidia partnership May 13, 2026. Solid.

Sources: reuters.com · sequoiacap.com · bloomberg.com

Why this call: Ex-Anthropic research team automating AI research, pre-revenue.

Known for: ex-Anthropic scientific reasoning team

In more detail: Trains frontier AI models built to do AI research, aiming to automate the research loop of an entire AI lab. It has not released a model yet.

Revenue: No

Backers: Andreessen Horowitz and Kleiner Perkins (seed co-leads; Kleiner leading the $5B talks), Nvidia; Google Cloud ($100M+ compute deal)

Valuation notes: Seed closed: $200M at ~$1B (Jun 25, 2026). Bloomberg (Sep 22, 2026): talks to raise up to $1B at ~$5B, Kleiner leading, a16z returning. Plans a frontier model by early 2027 on Google TPUs.

Sources: theinformation.com · the-decoder.com · bloomberg.com

Why this call: Researcher-led, pre-revenue, compute-heavy, model-first, $1.1B at founding.

Known for: Cofounder of xAI

In more detail: Builds a platform for training and serving open-weight models with reinforcement learning, sold as River Cloud and an API. Its longer-term aim is personal AI agents that learn each user's preferences.

Revenue: No

Backers: General Catalyst, AMP PBC (leads), Nvidia, AMD Ventures, Temasek, Y Combinator, Cisco Investments; Babuschkin (up to $100M personally)

Valuation notes: Medium: $1.1B raised (Aug 11, 2026) at ~$5B per AI Weekly and other aggregators (Harmonic imputes ~$5.5B). Bloomberg covered the Jun 10 unveiling. PitchBook has no record yet.

Sources: aiweekly.co · bloomberg.com · dealroom.co

Why this call: Researcher-founded self-improving AI lab, pre-revenue, compute-heavy.

Known for: Self-improving AI lab; four-month-old company reported to raise $500M+

In more detail: Builds AI systems that automate how AI models are trained and improved. Its first model, Flint, is an open-weight coding model that runs on a laptop and was built largely by its automated system.

Revenue: No

Backers: GV (lead), Nvidia; round reportedly oversubscribed; Greycroft (co-lead), AMD Ventures; Amazon ($410M compute deal)

Valuation notes: Launched May 13, 2026 with $650M at $4.65B (company-stated; SiliconANGLE), superseding the FT's earlier '$500M at $4B pre'. $410M AWS compute deal (Jul 2026); Flint coding model (Oct 21, 2026).

Sources: ucl.ac.uk · sifted.eu · app.dealroom.co · siliconangle.com

Why this call: Researcher-led world-model lab, pre-revenue, huge seed.

Known for: Yann LeCun's post-Meta world-model lab; one of Europe's largest seed rounds

In more detail: Builds world models that learn how the physical world works from video, aiming for AI that can reason and plan beyond what language models do. Its first model, AMI Video, is still in development.

Revenue: No

Backers: Cathay Innovation, Greycroft, Hiro Capital, HV Capital, Bezos Expeditions, Temasek, Nvidia, SBVA, others; Shorooq

Valuation notes: Reuters/FT reported a $1.03B seed at $3.5B pre-money; implied post-money is roughly $4.5B.

Sources: reuters.com · ft.com · techcrunch.com

Why this call: Researcher-led, pre-revenue, huge early raise, but hardware, not mainly software.

Known for: Naveen Rao founded MosaicML (sold to Databricks for $1.3B) and co-founded Nervana (sold to Intel)

In more detail: Designs analog AI chips and servers that aim to run AI far more efficiently, borrowing ideas from how the brain works. It has shown Un-0, an image model running on a simulation of its chip.

Revenue: No

Backers: a16z and Lightspeed (co-leads), Lux, DCVC, Databricks

Valuation notes: Solid: $475M seed at $4.5B post ($4.03B pre), Dec 2025, first tranche of up to $1B (TechCrunch, PitchBook). No 2026 round or talks found.

Sources: techcrunch.com

Why this call: Researcher-founded, pre-product, heavily seeded.

Known for: ex-xAI researcher; early Google leader

In more detail: Builds AI models designed to understand people and help them work together. Its first release, Persimmon, simulates how real people behave in group chats and is offered as a research preview.

Revenue: No

Backers: SV Angel, Georges Harik, Nvidia, Jeff Bezos, GV, Emerson Collective, others

Valuation notes: Reuters/Bloomberg: $480M seed at $4.48B (Jan 20, 2026); Goodwin (Mar 16) says the seed grew to >$500M at ~$4.5B. Released 'Persimmon' model Sep 2026. No follow-on found.

Sources: reuters.com · goodwinlaw.com

Why this call: Ex-OpenAI research team working on automated AI research/continual learning, pre-revenue.

Known for: ex-OpenAI VP/researcher; 'world's most automated AI lab'

In more detail: Building a highly automated AI lab that develops new learning methods and model designs so AI keeps improving after it is deployed. No model has been released yet.

Revenue: No

Backers: Nvidia; others not disclosed (Acrew listed in an SEC filing per Axios)

Valuation notes: The Information: $100M at $1B closed (Mar 2026); targeting $300–500M at ~$4B (May 7, 2026); SEC filing shows raising up to $500M (Axios, Aug 13). No close found. Founding team: Jerry Tworek, Rohan Anil, Joanne Jang, Anmol Gulati, Julia Villagra.

Sources: theinformation.com · businessinsider.com · the-decoder.com · theinformation.com

Why this call: Was a classic neolab: DeepMind co-founder, trained Inflection-2 on ~22K H100s, raised $1.3B with little revenue, before Microsoft hired its founders and most staff (Mar 2024).

Known for: Microsoft pseudo-acquihire; Pi assistant

In more detail: Sells customizable AI models that companies can run on their own infrastructure, built on its Inflection 3.0 models. It earlier built Pi, a consumer chatbot that is still available.

Revenue: Unknown ('generating revenue', no figure)

Backers: Microsoft, Nvidia, Eric Schmidt, Bill Gates (Series B leads), Greylock, Dragoneer

Valuation notes: Stale: $1.3B Series B at $4B post (Jun 2023). Suleyman and Simonyan left for Microsoft AI in Mar 2024. No 2026 financing found.

Sources: inflection.ai

Why this call: Researcher founders, pre-revenue, compute-heavy and RL-based; end market is chips.

Known for: AlphaChip, the RL chip-floorplanning method used in four TPU generations

In more detail: Builds AI that designs computer chips and gets better with each design, aiming to speed up how AI chips are made. No product has been publicly released yet.

Revenue: No

Backers: Lightspeed (Series A lead), Sequoia and Striker (seed leads), DST Global, NVentures, Felicis, Radical Ventures

Valuation notes: Solid: $300M Series A at $4B post (Jan 26, 2026; TechCrunch, PitchBook). Seed: $35M at $750M (Dec 2025). ~$335M raised.

Sources: techcrunch.com

Why this call: Video product revenue, but moving into world models and physical AI.

Known for: Dream Machine; Ray3 reasoning video model; Saudi HUMAIN mega-deal

In more detail: Builds AI models that generate video and images from text, and is extending them into world models and robotics. Its Ray3 video model powers its Dream Machine app and is offered in Adobe Firefly.

Revenue: Yes

Backers: HUMAIN/PIF (lead), a16z, Amplify, Matrix, AMD Ventures, Amazon

Valuation notes: $900M Series C at $4B (Nov 19, 2025), HUMAIN-led. No 2026 round found. Jun 1, 2026: launched an 'Open Physical AI Lab' (robotics) extending its world-model work.

Sources: semafor.com

Why this call: Compiler and infrastructure tooling rather than a model lab.

Known for: Mojo language + MAX engine; Chris Lattner (LLVM/Swift creator); CUDA challenger

In more detail: Builds software that lets AI models run fast on chips from many vendors without rewriting code, including the Mojo programming language and the MAX inference engine. Now part of Qualcomm.

Revenue: Unknown

Backers: US Innovative Technology Fund (lead), GV, General Catalyst, Greylock, DFJ Growth

Valuation notes: $250M Series C at $1.6B (Sep 2025). Qualcomm announced acquisition Jun 24, 2026 for ~$3.9B all-stock (19.2M shares at $204.13); closed Jul 29, 2026. Mojo, MAX and Modular Cloud continue as brands.

Sources: bloomberg.com · quasa.io

Why this call: Researcher-founded, models-first, compute-heavy, raised a lot early.

Known for: Ex-OpenAI and Meta FAIR founders; Chai-2 and Chai-3 de novo antibody design models

In more detail: Designs antibodies and other drug molecules from scratch with AI models. Its Chai-2 model designs antibodies with high lab success rates, and partners include Eli Lilly.

Revenue: Unknown (pharma deals with Lilly, Pfizer, Novartis, argenx; terms undisclosed)

Backers: Index Ventures (Series C lead), Oak HC/FT and General Catalyst (Series B leads), Menlo Ventures and Anthology Fund (Series A leads), Thrive, OpenAI, Kleiner Perkins, Sequoia

Valuation notes: Solid: $400M Series C at $3.8B post (Jul 14, 2026; The Pharma Letter, Dealroom). Sheet's $3.4B was the June talks figure. ~$630M raised.

Sources: thepharmaletter.com · dealroom.co

Why this call: Top researchers, no revenue, compute-heavy world models, software only, $700M seed.

Known for: Ex-Google DeepMind world-model team; Parker-Holder was a principal researcher on Genie 3

In more detail: Building world models that simulate how physical and virtual environments behave, for training robots and AI agents and for games. It is still in stealth with no public product.

Revenue: No

Backers: Index Ventures (co-lead), Lightspeed (co-lead), Creandum

Valuation notes: Advanced talks for a seed of up to $700M (Bloomberg, Sep 17, 2026). FT put it at $3.7B pre-money; Bloomberg at ~$3B pre; PitchBook logs it as in progress at $3B pre / $3.7B post. Not confirmed closed.

Sources: bloomberg.com · thenextweb.com · thecaptable.co.uk

Why this call: Researcher-founded robotics foundation model lab, pre-revenue; robotics rather than pure software.

Known for: French ex-Mistral / Skild robotics-model team

In more detail: Builds robot foundation models and a dexterous robot hand so robots can handle objects with human-like skill. Its GENE-26.5 model launched in 2026, and it also makes the open-source Genesis physics simulator.

Revenue: No

Backers: Eclipse and Khosla (seed co-leads), Bpifrance, Xavier Niel, Eric Schmidt; Premji Invest (in talks to lead), HSG and Northzone (weighing)

Valuation notes: Seed: $105M (co-led by Eclipse and Khosla). Bloomberg (Jul 23, 2026): talks to raise ~$500M at ~$3B pre-money, Premji leading; not confirmed closed. Sheet's earlier ~$500M figure was stale.

Sources: reuters.com · bloomberg.com · thenextweb.com

Why this call: Huge early raise, but an operator-led corporate spinout building hardware and deployments, not a researcher lab.

Known for: Rivian spinout ('Project Synapse') that raised >$1B in ~6 months

In more detail: Builds AI models and robot hardware for factory work, trained on real manufacturing data. Its first robots work in Rivian's Illinois plant, and it plans to sell to other manufacturers.

Revenue: Unknown (not disclosed)

Backers: Kleiner Perkins (A1 lead), Accel and a16z (Series A leads), Eclipse (seed lead), Greenoaks, Prysm, Bain Capital Ventures, Meritech, Redpoint, Rivian (~1/3 owner)

Valuation notes: Solid: $400M Series A1 at $3.4B post (May 13, 2026) confirmed by PitchBook, CNA/Reuters, TechFundingNews. Earlier: $115M seed (Nov 2025), $500M Series A at ~$2B (Mar 2026). No CEO named.

Sources: channelnewsasia.com · techcrunch.com · techfundingnews.com

Why this call: Ex-OpenAI researchers and heavy compute, but ~$50M annualized revenue and a services model fail the low-revenue test.

Known for: Three young ex-OpenAI researchers (post-training infra, early Codex, reasoning); valuation went $100M → $3.25B in ~15 months

In more detail: Trains custom AI models on a company's own data with reinforcement learning, then runs them on its cloud platform, AC2. Customers include DoorDash, Cognition and Harvey.

Revenue: Yes (~$14M FY2025 per PitchBook; ~$50M annualized by Aug 2026 per The Information)

Backers: Elad Gil (lead), Kleiner Perkins, Benchmark, Sequoia, Lux Capital, Shopify Ventures, QuantumLight

Valuation notes: PitchBook records $350M at $3.25B post (Sep 10, 2026) as closed; some press still calls it talks. The Information first reported ~$3B talks (Aug 10). Prior: $80M at $1.3B (Apr 2026), $60M Series A at $500M (Jun 2025).

Sources: theinformation.com · forbesargentina.com · aiweekly.co

Why this call: Preclinical wet-lab drug developer heading to trials, not a software/model company.

Known for: Coinbase CEO's longevity biotech; valuation roughly tripled in 2026; first liver trial slated for 2027

In more detail: Develops medicines that reprogram aging cells to act younger, using AI and genetic screens to find the right gene combinations. Its first therapy targets the liver, with human trials planned for 2027.

Revenue: No

Backers: Founders Fund (lead), Thrive Capital, Lilly Ventures, Valor, 8VC, Kleiner Perkins, Greenoaks, Nat Friedman, Daniel Gross

Valuation notes: Solid: $435M Series C at $3.1B post (Jun 2, 2026); STAT quotes CEO, PitchBook records as actual. ~$682M raised in total.

Sources: statnews.com · qz.com

Why this call: Applied healthcare workflow product with ~$50M revenue.

Known for: Prescription-access automation; valuation went $1B (May 2026) → $3B (Sep 2026)

In more detail: Builds AI agents that handle prior authorizations, insurance appeals and pharmacy routing so patients can get their prescriptions. It is free for doctors and earns money from drug companies.

Revenue: Yes (~$50M FY2026 per PitchBook)

Backers: Bain Capital Ventures (Series C lead), Accel (Series B lead), Thrive Capital, General Catalyst, Redpoint, BoxGroup, SV Angel

Valuation notes: Solid: $150M Series C at $3B post (Sep 8, 2026), Bloomberg, MedCity, Dealroom; PitchBook 'actual'. Series B: $160M (incl. $28M debt) at $1B (May 2026). Sheet's 'AI drug discovery' label corrected.

Sources: dealroom.co · advisorperspectives.com

Why this call: Researcher-founded, minimal revenue, compute- and data-heavy, model-first, $740M in under two years.

Known for: Florence and Zeng are ex-Google DeepMind robotics researchers (PaLM-E, RT-2 era); Barry ex-Boston Dynamics; GEN-series models

In more detail: Trains robot foundation models on hundreds of thousands of hours of real-world data so ordinary robot arms can do delicate hand tasks. Its models are GEN-0 and GEN-1.

Revenue: Unknown ('a handful' of customers)

Backers: Radical Ventures, 8VC (Series B leads), NVentures, Spark Capital, Boldstart, NFDG, Union Square Ventures, Bezos Expeditions, Norwest; angels Fei-Fei Li, Naval Ravikant

Valuation notes: Solid: Series B totaled $600M, $400M at ~$2B (Jun 2026, Bloomberg) plus ~$200M extension led by 8VC at $3B post (TechCrunch, Aug 25, 2026, confirmed by filing; PitchBook 'actual'). ~$740M raised in total.

Sources: techcrunch.com · bloomberg.com

Why this call: Researcher-led model lab with little revenue and heavy compute; also building an enterprise/government product business.

Known for: India's flagship sovereign AI lab; open-source Sarvam 105B and 30B MoE models; government GPU access (4,096 H100s) in exchange for equity

In more detail: Builds language, speech and translation models for Indian languages, offered by API and on-premises. Models include Sarvam-105B, Sarvam-30B and the Bulbul voice model, and clients include Aadhaar agency UIDAI.

Revenue: No (~$5.1M revenue, FY to Mar 2026, per PitchBook)

Backers: HCLTech ($150M strategic), Nvidia, Bessemer, Amazon, Khosla Ventures, Peak XV, Lightspeed, Government of India (equity for GPUs)

Valuation notes: Series B: $234M first close Jun 15, 2026 ($300M by Aug per PitchBook) at $1.5B post, solid (Bloomberg, AlphaSignal, PitchBook). Sources differ on lead (HCLTech vs. Nvidia/HCLTech/Bessemer co-lead). Oct 13, 2026: Economic Times reports talks for $250–300M at ~$3B, unconfirmed.

Sources: alphasignal.ai · bloomberg.com · analyticsindiamag.com

Why this call: Researcher-led and heavily funded early, but wet-lab and drug-pipeline heavy.

Known for: $1B committed launch; ARCH/Foresite-incubated

In more detail: Discovers and develops drugs by pairing AI models with its own large wet labs. It has released X-Cell, a model that predicts how cells respond to genetic changes, and an antibody design platform called X-Design.

Revenue: Unknown

Backers: ARCH, Foresite, Sequoia, Lightspeed, NEA, Menlo, Lux, others

Valuation notes: Soft: PitchBook estimates an ~$800M Series A at ~$3.55B post (Aug 2025); not a disclosed price. ~$1.3B raised. 2026: launched X-Cell virtual cell model (Mar 17), revealed antibody platform and first pipeline look (Oct).

Sources: fiercebiotech.com · xaira.com

Why this call: Researcher-founded model lab, now pushing hard into products and enterprise sales in Japan.

Known for: Japan-focused unicorn; AI for science results

In more detail: Builds nature-inspired AI and automated research tools, selling mainly to Japanese businesses such as MUFG. Products include The AI Scientist, the Marlin research agent and Fugu, which routes tasks across top models.

Revenue: Unknown ('generating revenue', no figure)

Backers: MUFG, Khosla, Macquarie, NEA, Lux, IQT, others

Valuation notes: Solid: ¥20B (~$200M) Series B at ¥400B (~$2.65B) post (Nov 17, 2025). No 2026 round; 2026 launches include Sakana Chat, Marlin, Fugu and a Datadog partnership. ~$444M raised.

Sources: sakana.ai

Why this call: Scientist-led model company with no disclosed revenue and a big raise; tied to lab partners rather than pure software.

Known for: Max Welling (ex-Microsoft, Qualcomm ML pioneer); Yann LeCun and Geoffrey Hinton are advisers; AI Materials Foundry with Nvidia, Meta, Samsung, Hyundai

In more detail: Uses generative AI and physics simulation to design new materials, for uses like carbon capture and chips. Partners include Meta, Hyundai and chemicals maker Kemira.

Revenue: Unknown (customers incl. ASML, Meta)

Backers: Kleiner Perkins and NEA (co-leads), Bezos Expeditions, UK Sovereign AI Fund, Lux, AMD Ventures, Temasek, Northzone, Hoxton, Hyundai, Samsung

Valuation notes: Solid: ~$450M (€393M) Series B at $2.6B (€2.2B) post, Jul 20, 2026 (Tech.eu, EU-Startups, Dealroom). Up from $520M at the $100M round (Sep 2025).

Sources: tech.eu · eu-startups.com

Why this call: Researcher-led, no revenue, huge early raise; runs a wet neuroscience lab and may build chips, so not pure software.

Known for: Reardon created Internet Explorer and co-founded CTRL-labs (bought by Meta); DeepMind's Greg Wayne joined

In more detail: Studies how the brain's cortex computes, using its own neuroscience lab, to build AI that keeps learning and runs on far less power. No models have been released yet.

Revenue: No

Backers: Jeff Bezos (~1/5 of round), GV, Lux Capital, Catalio Capital

Valuation notes: Solid: Wired reported $500M first round at $2.5B post (Jun 2026); PitchBook records it as completed ($2.0B pre). Crunchbase's 'Cortex AI' label did not appear in other sources.

Sources: pulse2.com · siliconangle.com · news.crunchbase.com

Why this call: MIT researcher lab with a novel architecture and big early raise, now selling to enterprises.

Known for: MIT spinoff working on efficient / on-device models

In more detail: Builds small, efficient foundation models that run on phones, laptops and cars instead of the cloud. Its models are the LFM2 family, and partners include AMD, Qualcomm and Mercedes-Benz.

Revenue: Unknown

Backers: AMD (Series A lead), Samsung, Shopify, G42, OSS Capital, YXS Capital (2026)

Valuation notes: $250M Series A led by AMD at ~$2.35B post (Dec 2024; PitchBook; Harmonic $2.4B). A further round from YXS Capital closed Aug 28, 2026, terms undisclosed, current value may be higher.

Sources: my.pitchbook.com · startuphub.ai

Why this call: Began as a DeepMind-researcher lab, but every founder has left; now an agent product company.

Known for: ex-DeepMind French AI model company

In more detail: Builds AI agents and computer-use models that operate software by seeing and clicking through screens, automating office work. Models include the Holo family, and products include Runner H and Surfer H.

Revenue: Unknown

Backers: Accel, Creandum, Visionaries, Bpifrance, Eurazeo, Amazon, Samsung, UiPath, Eric Schmidt, Xavier Niel

Valuation notes: Sheet's $2B kept per your rule but unsupported: PitchBook's last valuation is ~$370M post (~$220M seed, 2024). Three co-founders left Aug 2024; CTO Laurent Sifre, the last original founder, stepped down ~May 2026 and is reportedly raising $30M for a new startup.

Sources: sifted.eu · the-decoder.com

Why this call: Researcher-founded and raised big early, but product-led with enterprise customers and modest compute.

Known for: Stanford 'generative agents' lineage

In more detail: Builds AI simulations of real people, trained on interviews and behavioral data, so companies can test how customers will react before acting. Clients include CVS Health and Gallup.

Revenue: Unknown ('5x revenue growth in 5 months'; customers CVS, Wealthfront, Deloitte, Gallup)

Backers: Greenoaks and Index Ventures (Series B), Index (Series A lead), Bain Capital Ventures, A*, CVS Health Ventures, Fei-Fei Li, Andrej Karpathy, Adam D'Angelo

Valuation notes: Solid on $2B: ~$200–210M Series B at $1.8B pre / $2B post, announced Jul 2026, closed Sep 22, 2026 (Pulse2, PitchBook). Series A: $100M led by Index (Feb 2026).

Sources: pulse2.com

Why this call: Provisional: researcher-founded stealth lab seeking a very large first round; team partly confirmed.

Known for: Adi is a Hebrew University professor and ex-Meta FAIR senior research scientist (speech/audio language models)

In more detail: Stealth AI research lab spread across the US, France and Israel. Its focus is not public, though a co-founder is a speech and audio AI researcher.

Revenue: No

Backers: Not disclosed

Valuation notes: Weak: Sifted exclusive (Jun 30, 2026) says it is raising ~$400M (one aggregator: initial target $175M, expandable). No close or valuation reported; ~$2B imputed at ~20% dilution.

Sources: sifted.eu · inforcapital.com

Why this call: Robot-intelligence model company with a strong research team, but it deploys and plans its own hardware.

Known for: $450M Series A; FutureVision robot-intelligence platform trained on internet-scale video

In more detail: Trains robot models on internet video that predict what will happen next and turn that into actions for factory and warehouse robots. Its platform is called FutureVision.

Revenue: Unknown

Backers: Premji Invest (lead), Khosla Ventures, Temasek, Mayfield, Capricorn Investment Group, John Doerr, others

Valuation notes: Reuters and the company announcement reported a $450M Series A at a $1.7B valuation; company said it launched after 18 months in stealth.

Sources: reuters.com · businesswire.com

Why this call: Researcher-led formal-math lab, pre-revenue.

Known for: Formal mathematics / verified-AI lab; claimed 12/12 Putnam-style problem solving

In more detail: Builds AI that writes math proofs checked by the Lean proof language, aiming to verify both math and code. Its AxiomProver solved all 12 Putnam 2025 problems and scored perfectly at IMO 2026.

Revenue: No

Backers: Menlo Ventures (lead), Greycroft, Madrona, B Capital, Toyota Ventures, others

Valuation notes: Menlo announced a $200M Series A at $1.6B post-money; Axiom described the round as a $1.6B+ valuation.

Sources: menlovc.com · linkedin.com · pulse2.com

Why this call: Pure research lab with no meaningful revenue, compute-heavy, ~$1B raised early; very quiet.

Known for: Long-context frontier coding from Europe

In more detail: Builds AI coding models that can read huge codebases at once, aiming to automate software engineering. Its LTM-2-mini model handles about 10 million lines of code in a single prompt.

Revenue: No

Backers: Eric Schmidt, CapitalG, Atlassian, Jane Street, NFDG, Elad Gil, Nat Friedman/Daniel Gross, Sequoia, others

Valuation notes: Soft: PitchBook estimates a ~$550M Series D at ~$3.8B post (Jul 2025); Harmonic says $2.9B; no press confirmation. ~$1B raised in total. Sheet's $1.5B dated from Aug 2024. No 2026 news.

Sources: techcrunch.com

Why this call: Research collective training models with modest revenue; crypto-adjacent and smaller on compute.

Known for: Hermes open models; Psyche decentralized training network; Hermes Agent

In more detail: Builds open-weight language models and Hermes Agent, a popular open-source AI agent. Its models are the Hermes series, and it runs Psyche, a network for training AI across scattered GPUs.

Revenue: Unknown (no figures)

Backers: Robot Ventures (Series B lead), Paradigm, Nvidia, Union Square Ventures, Menlo Ventures, Samsung, M12, Distributed Global, North Island Ventures

Valuation notes: Solid: $90M Series B at $1.5B post (Oct 7, 2026; WSJ via CryptoRank, The SaaS News). The Block earlier reported talks for ~$75M at $1.5B.

Sources: thesaasnews.com · theblock.co

Why this call: Research pedigree and big early raise, but sells vertical application software with a customer in place.

Known for: Bob McGrew, ex-OpenAI Chief Research Officer; Augustus Odena co-founded Adept (ex-Google Brain)

In more detail: Builds an AI agent that reads product design files and plans factory assembly, writing work instructions and station layouts. Anduril is its first customer, and robots for assembly lines are next.

Revenue: Unknown ('generating revenue'; Anduril first customer)

Backers: Altimeter, Accel, Khosla (Series A co-leads), Founders Fund (seed co-lead), 8VC, Felicis, SV Angel

Valuation notes: Solid: $160M Series A plus a previously undisclosed $72M seed ($232M total) at ~$1.5B (Oct 20, 2026; CNBC, TNW, PitchBook, Harmonic). PitchBook says pre-money; press says post.

Sources: cnbc.com · thenextweb.com

Why this call: Researcher lab with no revenue, compute-focused, big early raise.

Known for: Aristotle, which produces Lean-verified proofs; gold-medal IMO-level results

In more detail: Builds AI that solves math problems and proves its answers correct using the Lean proof language. Its model, Aristotle, reached gold-medal level at the 2025 Math Olympiad and is offered as an app.

Revenue: No (Aristotle API free beta as of Jan 2026)

Backers: Ribbit (Series C lead), Sequoia, Kleiner Perkins (Series B leads), Index, Emerson Collective, NVentures, Paradigm

Valuation notes: Solid: $120M Series C at $1.45B post (Nov 25, 2025); NVentures joined (Jan 15, 2026, Axios). ~$295M raised; most new money goes to compute.

Sources: axios.com · pulse2.com

Why this call: Research lab, pre- or barely-revenue, very compute-heavy, model-first.

Known for: Oliver Cameron: ex-Voyage CEO, ex-Cruise VP Product; real-time, action-conditioned interactive video world models

In more detail: Builds world models that generate interactive video in real time, so users can move through and act in a scene as it is created. Models include Odyssey-2 and Odyssey-2 Max.

Revenue: Unknown (early API product)

Backers: Natural Capital (lead), GV, Amazon/AWS, AMD Ventures, NVentures, Samsung Next, In-Q-Tel, EQT Ventures, Air Street, DCVC; angels Jeff Dean, Elad Gil

Valuation notes: Solid: $310M Series B at $1.45B post ($1.14B pre), Jun 17–18, 2026 (Crunchbase News, PitchBook). ~$337M raised in total.

Sources: news.crunchbase.com

Why this call: Researcher team with a model at the core and a large early raise, but includes gripper hardware.

Known for: Agrawal is an MIT professor (Improbable AI Lab); Haarnoja is ex-DeepMind and author of Soft Actor-Critic

In more detail: Builds warehouse picking robots that pair a force-sensing gripper with a vision-force-action AI model, so they can feel how firmly they hold things. It claims about 1,500 picks an hour.

Revenue: Unknown

Backers: a16z and F-Prime (co-leads), Obvious Ventures

Valuation notes: Solid: $150M Series A at $1.45B post (Oct 20, 2026) per TechCrunch and PitchBook. Seed: $11M at $48M (May 2025).

Sources: techcrunch.com · aiweekly.co

Why this call: Eight years old with ~$50M revenue; a product company more than a research lab.

Known for: Israeli LLM company; Stanford professor founder

In more detail: Builds enterprise AI, now centered on Maestro, a system that plans and runs AI agents for businesses. It also made the Jamba open language models, and Nebius is a licensing partner.

Revenue: Yes (~$50M TTM 2025 per PitchBook)

Backers: Nvidia, Google, Zoom Ventures, Walden Catalyst, Pitango

Valuation notes: Last priced: $300M Series D at $1.4B post (May 2025). Nvidia talks at up to $3B (Dec 2025/Jan 2026) fell through; AI21 denied them Jan 12, 2026. Nebius talks (Apr 2026) cancelled (May). Globes reports it is for sale.

Sources: en.globes.co.il · finance.yahoo.com

Why this call: Research-led foundation model and heavy early raise, but already has enterprise contracts and commercial go-to-market.

Known for: NEXUS, the first 'large tabular model', deterministic, non-transformer, handles billions of rows; DeepMind alumni co-founders

In more detail: Builds large tabular models that make predictions from spreadsheet-style business data, such as demand forecasts, prices and churn. Its first model, NEXUS, is sold to enterprises and through AWS.

Revenue: Unknown ('several seven-figure contracts' with Fortune 100 clients)

Backers: Oak HC/FT, Salesforce Ventures (Series A leads), Valor, Battery, Headline (seed), Hetz, Capital One Ventures, MUFG; angels Aravind Srinivas, Olivier Pomel

Valuation notes: Solid: $255M raised ($225M Series A) at $1.4B post (TechCrunch, Feb 5, 2026); PitchBook estimates $1.45B. Sources disagree on round leads.

Sources: techcrunch.com · venturebeat.com

Why this call: Sells agent products, raised little, needs little compute.

Known for: Ex-MIT professor Robert Yang; formerly Altera (Project Sid, 1,000-agent Minecraft simulation)

In more detail: Builds AI agents for office work. Its main product, Shortcut, is an agent that builds financial models and analysis inside Excel like a junior analyst.

Revenue: Unknown ('revenue-generating', no figure)

Backers: Prosus Ventures and MiH (Series A leads), First Spark, Patron, a16z Speedrun, Patrick Collison, Eric Schmidt

Valuation notes: $33M Series A ($31M equity + $2M debt) at $232M post (Aug 15, 2025) per PitchBook; Harmonic estimates lower ($165M). No 2026 round found. ~$44M raised in total.

Sources: theaiinsider.tech

Why this call: Researcher lab training its own models, but meaningful enterprise revenue and now acquiring companies.

Known for: Ex-DeepMind/Google Brain team; Snowflake considered buying it for $1B in 2024; acquired Moonvalley (May 2026)

In more detail: Builds multimodal AI models that understand text, images, video and audio, including Reka Core and Reka Flash. Its newest, Rho-1, also generates video and controls robots.

Revenue: Yes* (~$10.9M per Caplight, date unclear)

Backers: Nvidia and Snowflake (Series B co-leads), DST Global and Radical Ventures (Series A leads), New York Life Ventures, Nat Friedman

Valuation notes: Soft: Juniper tracker lists a $12M Series A-2 (Feb 4, 2026) at $1.3B post; PitchBook and Caplight still show ~$1B from the $110M Series B (Jul 2025). Bought Moonvalley in an all-stock deal (May 2026).

Sources: juniper.xyz · siliconangle.com

Why this call: Research-led, little revenue, big early raise; uses less compute than a pretraining lab.

Known for: Tom McGrath co-founded Google DeepMind's interpretability team; says interpretability-informed training halved hallucinations; found new Alzheimer's biomarkers by reverse-engineering a bio model

In more detail: Builds interpretability tools that look inside AI models to understand, debug and edit how they work. Its platform is Ember, and partners include Arc Institute.

Revenue: Unknown (no figures; PitchBook 'generating revenue')

Backers: B Capital (Series B lead), Menlo Ventures (Series A lead), Lightspeed, Salesforce Ventures, DFJ Growth, Anthropic, Wing, South Park Commons, Work-Bench, Eric Schmidt

Valuation notes: Solid: $150M Series B at $1.25B post ($1.1B pre), announced Dec 17, 2025, closed Feb 5, 2026 (PitchBook 'actual'). No newer round found. Launched agent monitors Oct 2026; joined Base Labs open-weight safety partnership.

Sources: work-bench.com · pulse2.com

Why this call: Research-founded SSM model builder, but now a usage-based API business with many customers.

Known for: State Space Models; ultra-low-latency voice (Sonic); Albert Gu (Mamba) pedigree

In more detail: Builds real-time voice AI on a fast state space architecture, including Sonic for text-to-speech and Ink for speech-to-text, plus a platform for phone voice agents. Customers include ServiceNow, EliseAI and Quora.

Revenue: Unknown (10,000+ customers; ARR not disclosed)

Backers: Kleiner Perkins, Index Ventures, Lightspeed, Nvidia, A*, Dell Tech Capital, Samsung Ventures

Valuation notes: Weak: last round $100M (Oct 2025, Kleiner-led), no valuation disclosed; last disclosed valuation $362M post (Mar 2025 Series A). ~$191M raised. No 2026 round found; sheet's '$1B+' kept but unsupported by sources (imputed ~$650M from last round). Sonic-3.5 ranked #1 on Artificial Analysis Speech Arena (May 2026).

Sources: sacra.com

Why this call: Now trains its own open models and is compute-hungry, but started as an enterprise SLM vendor; small team (~32).

Known for: Created mergekit; Trinity Large 400B open-weight model (early 2026); says it trained four models for ~$20M

In more detail: Builds open-weight American language models and helps companies run them privately. Its models are the Trinity family, led by the 400-billion-parameter Trinity Large.

Revenue: Unknown (PitchBook: $2M FY2023; nothing newer)

Backers: Vista Equity Partners, Prosperity7, Emergence Capital (Series B co-leads), Cambium, M12, Wipro Ventures, Hitachi Ventures, Samsung Next, Long Journey, Flybridge, WndrCo

Valuation notes: Solid: $150M Series B at $1B pre / $1.15B post (closed Sep 10, 2026 per PitchBook; Dealroom, AI Weekly agree). A KuCoin headline saying '$10B' is wrong.

Sources: dealroom.co · aiweekly.co

Why this call: Research team training its own models, little disclosed revenue, heavy compute, big raises.

Known for: Zamba/Zamba2 hybrid SSM models; trains on AMD GPUs via IBM Cloud partnership

In more detail: Builds open-weight language models designed to be small and efficient, trained on AMD chips. Its releases include the Zamba hybrid models and ZAYA1, a reasoning model with under a billion active parameters.

Revenue: Unknown

Backers: Jaan Tallinn (Series A lead per Nextomoro), AMD, Bison Ventures, Future Ventures, Metaplanet, IBM (partner)

Valuation notes: Last closed: $100M Series A at $1B post (Jun 2025). Forbes (May 19, 2026, via 36Kr): raising ~$500M at ≥$5B with AMD participating, talks, no close confirmed.

Sources: eu.36kr.com · futurumgroup.com

Why this call: Pure researcher lab before acquisition.

Known for: ESM3 protein language model; team from Meta FAIR's ESM group; Rives is now Biohub's head of science

In more detail: Built protein language models that read and design proteins, led by ESM3 and ESM C. Now part of the Chan Zuckerberg Biohub.

Revenue: Unknown

Backers: Lux Capital (seed lead), Nat Friedman, Daniel Gross, Amazon/AWS, NVentures

Valuation notes: Last priced: $142M seed at ~$1B post (Jun 2024; PitchBook 'actual'). Nov 6, 2025: team joined Chan Zuckerberg Initiative's Biohub in a 'strategic transaction', PitchBook records a 100% acquisition, price undisclosed. No longer an independent lab.

Sources: gunder.com

Why this call: Application company with enterprise customers and modest compute.

Known for: Headline $1B Series A

In more detail: Builds AI simulations of whole populations so companies can test products, prices and messages on virtual customers before launch. Customers include EY, Accenture and Interpublic Group.

Revenue: No (ARR 'still below $10M' per TechCrunch, Dec 2025)

Backers: Redpoint (lead), Felicis, General Catalyst, K5 Global, QuantumLight, Accenture Ventures, Hanwha

Valuation notes: Series A at a $1B headline valuation (Dec 5, 2025; TechCrunch) using a multi-tier price, so the blended valuation is lower; ~$88M raised (PitchBook). No 2026 round. Joined Comcast LIFT Labs (spring 2026).

Sources: techcrunch.com

Why this call: Pure researcher lab, no revenue, frontier thesis; smaller than peers on capital and compute.

Known for: Keras founder and Zapier founder; program synthesis thesis

In more detail: Research lab that combines deep learning with program synthesis, where AI searches for small programs that explain data, aiming for systems that learn and reason more like people. No models released yet.

Revenue: No

Backers: Coatue, Factorial Capital, Menlo Ventures, Quiet Capital, Y Combinator (W26)

Valuation notes: Weak: ~$43M seed (2025) at ~$215M post are PitchBook estimates, never announced. Joined YC W26 (Jan 2026, $500K SAFE). No other 2026 raise. ~15 staff.

Sources: my.pitchbook.com

Why this call: Started as a classic neolab, now ships developer products and no longer trains frontier models.

Known for: Early AI-lab unicorn; trained its own 70B model, then pivoted to agent and coding tools (Sculptor, Vet, Bouncer)

In more detail: Builds open tools for developers working with AI coding agents, including Sculptor for running several agents in parallel and Vet for catching their mistakes. It earlier trained its own 70B language model.

Revenue: Unknown

Backers: Astera Institute, Nvidia, The Navigation Fund (Series B co-leads), Amazon Alexa Fund, Dell, Eric Schmidt

Valuation notes: Stale: $212M Series B at ~$1B post (Oct 2023; PitchBook 'estimated'), later topped up by Amazon Alexa Fund and Eric Schmidt. No 2026 round, talks or exit found.

Sources: imbue.com · forbes.com.au

Why this call: Researcher-founded, pre-revenue, compute-heavy, raised $200M early.

Known for: Mobileye/AI21 founder's new lab; spent ~half its money on 1,000 Nvidia B300s

In more detail: Builds AI meant to match top human experts in narrow technical fields. Its first model, Maxwell, rewrites software while it runs so it uses CPUs and GPUs more efficiently, cutting compute bills.

Revenue: No (Maxwell in early access)

Backers: Lightspeed and Bessemer (co-leads), Nvidia, Dell Technologies Capital, Michael Dell, Pitango, Greenfield Partners, BRM

Valuation notes: Medium: $200M Series A at '>$1B' (Dealroom, SiliconANGLE, Oct 19, 2026; Globes reported the same round at ~$1B in Nov 2025, one round reported twice). ~$220M raised incl. $21M seed. Legal name AA-I Technologies; brand doubleAI.

Sources: dealroom.co · en.globes.co.il · siliconangle.com

Why this call: Passes research/no-revenue/model tests, but a nonprofit with no venture equity.

Known for: Moshi, a real-time open-source voice model; Unmute; spun out Gradium

In more detail: Nonprofit Paris lab that releases open-source AI models, mostly for speech. Its models include Moshi, a real-time voice assistant, Hibiki for live speech translation, and Pocket TTS. Its spinout Gradium sells voice AI.

Revenue: No (nonprofit)

Backers: Xavier Niel, Rodolphe Saadé, Eric Schmidt

Valuation notes: Nonprofit, no equity valuation; launched with ~€300M committed funding (2023). Sheet's '>$1B' has no source. Commercial arm is Gradium.

Sources: sciencebusiness.net

Why this call: ~$100M run-rate fails the low-revenue test; much of the business is compute and infrastructure.

Known for: Distributed model training, inference, and RL

In more detail: Builds an open platform for training and serving AI models, offering GPUs, a hub of reinforcement learning environments and inference. It trains its own open models such as INTELLECT-3. Users include Goodfire and Extropic.

Revenue: Yes (~$100M annualized run-rate per TechCrunch)

Backers: Radical Ventures (Series A lead), Founders Fund, Menlo Ventures, NVentures, Intel Capital, Dell Technologies Capital, Andrej Karpathy, John Schulman

Valuation notes: Solid: $130M Series A at $1B post (Jul 8, 2026; TechCrunch, Cooley). Acquiring a startup to speed up inference (Oct 2026).

Sources: cooley.com · letsdatascience.com

Why this call: Transformer co-authors pretraining their own open models, pre-revenue, with a big early raise; Nvidia acqui-hired the team in Jun 2026.

Known for: Transformer-paper co-authors; Rnj-1 open-weight models

In more detail: Built open-source language models from scratch, focused on coding and reasoning, and released the Rnj-1 models in late 2025. Most of its team joined Nvidia's Nemotron group in 2026.

Revenue: No

Backers: March Capital, Thrive, AMD, Nvidia, Google, others

Valuation notes: Correction: Nvidia acqui-hired the team (Vaswani and others joined Nvidia's Nemotron group, Jun 2026; reported by Sharon Goldman and Axios), it did not buy the company; terms undisclosed. Last round: $16.5M SAFE (Jun 2024); ~$76–121M raised. An '$8.6B' rumor was retracted by its own source (Oct 6, 2026); a claimed $175M at $1B is unverified.

Sources: groundlevel-ai.com · valueaddvc.com

Why this call: Pre-revenue research lab with a frontier thesis and a large seed.

Known for: Sara Hooker was Cohere's VP of research (ex-Google Brain); Sudip Roy a Cohere inference exec

In more detail: Builds AI systems that keep learning and adapting after deployment instead of needing costly retraining. Its platform improves datasets and trains custom models, starting with a tool called AutoScientist.

Revenue: No

Backers: Emergence Capital (lead), Threshold Ventures, Mozilla Ventures, Fifty Years, Alpha Intelligence Capital, E14 Fund, Neo

Valuation notes: $50M seed (Feb 4, 2026) confirmed by Wilson Sonsini and Seedtable. ~$1B valuation is 'reported' (ai2.work; Juniper marks it closed), not confirmed by the company.

Sources: seedtable.com · ai2.work

Why this call: Revenue-generating creative-tools turnaround.

Known for: Stable Diffusion; post-2024 recapitalization + leadership overhaul

In more detail: Makes generative AI models for images, video, audio and 3D, best known for Stable Diffusion and Stable Audio. It now focuses on entertainment, partnering with Warner Music, Universal Music and EA.

Revenue: Yes

Backers: Coatue, Lightspeed, Greycroft, Sound Ventures, Sean Parker (exec chair), Eric Schmidt, WPP; Sony Music, Universal Music, Warner Music, Electronic Arts, AMD Ventures

Valuation notes: $76M Series B (Aug 25, 2026), valuation undisclosed; total funding ~$232M. Last known valuation still ~$1B (2024 recap). Repositioned around music, gaming and entertainment; moved to LA; Sean Parker and James Cameron on board.

Sources: tech.eu · stability.ai

Why this call: Professor-founded model lab, early revenue, new architecture that needs compute.

Known for: Stanford/UCLA/Cornell professors who pioneered diffusion models; Mercury, the first commercial diffusion LLM

In more detail: Builds diffusion language models that write whole passages at once instead of word by word, making them much faster and cheaper. Its Mercury models, now on Mercury 3, handle chat and code.

Revenue: Unknown (paid API, early customers)

Backers: Menlo (lead), Mayfield, Innovation Endeavors, NVentures, M12, Snowflake, Databricks, angels incl. Andrew Ng and Andrej Karpathy

Valuation notes: Medium: $50M seed (Nov 2025, Menlo-led) at a reported ~$500M, not confirmed by PitchBook or Harmonic. No 2026 round found.

Sources: techfundingnews.com · unite.ai

Why this call: Raised big and early, but wet-lab/robotics infrastructure led by a product executive.

Known for: Kevin Weil: former OpenAI CPO and VP of OpenAI for Science; earlier Instagram, Twitter

In more detail: Runs chemistry labs powered by AI and robots that make molecules for drug discovery teams, working as a contract research service rather than developing its own drugs. Each experiment feeds back to improve its AI.

Revenue: No

Backers: Lightspeed (lead), IVP, Premji Invest, Lux Capital, Menlo Ventures, Conviction, Mayfield

Valuation notes: Solid: $215M at $865M on exiting stealth (Oct 22, 2026; Bloomberg via Startup Fortune; Harmonic). In Aug 2026 it was in talks for $150M at ≥$750M.

Sources: bloomberg.com · startupfortune.com · unite.ai

Why this call: BCI hardware and medical devices, not mainly software.

Known for: Altman-backed non-invasive BCI spinout from Forest Neurotech

In more detail: Developing brain-computer interfaces that read brain activity without surgery, using ultrasound and engineered molecules instead of implanted electrodes, plus AI to decode the signals. Butterfly Network supplies its ultrasound chips.

Revenue: No

Backers: OpenAI, Bain Capital, Gabe Newell, others

Valuation notes: TechCrunch/Wired reported $250M-$252M seed at ~$850M valuation; early research-stage.

Sources: techcrunch.com · wired.com

Why this call: Trains its own big bio models, but it's a 2019 company with wet-lab and therapeutics work that raised gradually, not heavily early.

Known for: EDEN biological foundation models trained on its 'Trillion Gene Atlas' evolutionary dataset; Jan 2026 preprint reported a 63.2% functional hit rate

In more detail: Trains biological AI models, called EDEN, on a huge library of genetic data collected from nature across dozens of countries. It uses them to design gene-editing tools, enzymes and its own therapies.

Revenue: Unknown (PitchBook: 'generating revenue', no figure)

Backers: S32 (lead), Anthropic, NVentures, Menlo Ventures (Anthology Fund), NATO Innovation Fund, UK Sovereign AI, Rockefeller Foundation, Redalpine, True Ventures, Catalio, André Hoffmann

Valuation notes: $140M Series C (Sep 23, 2026) confirmed by GEN, Tech.eu and PitchBook. Company did not disclose valuation; ~$800M post comes only from secondary aggregators citing an MSN-syndicated report. PitchBook's last recorded valuation is $177M (Oct 2024). ~$224M raised in total.

Sources: genengnews.com · tech.eu · hedgeco.net

Why this call: Research-founded and trains its own models, but a vertical CX product with enterprise revenue.

Known for: Roth and Klein co-founded Semantic Machines (acquired by Microsoft); Klein is a Berkeley NLP professor; APT (Agentic Pretrained Transformer)

In more detail: Builds AI models designed to follow business rules reliably without making things up, for customer service agents. Its model APT-1 powers virtual agents on Genesys Cloud through a partnership with Genesys.

Revenue: Unknown (Fortune 500 customers; PitchBook 'generating revenue')

Backers: Khosla Ventures (lead), Genesys, Lerer Hippeau, SilverCircle, Owl Ventures

Valuation notes: Solid: $100M Series A led by Khosla at $750M post ($650M pre), Jun 25, 2026 (PitchBook, AI Weekly). Earlier ~$21M at $96M (2023).

Sources: aiweekly.co · letsdatascience.com

Why this call: Agent-orchestration software for finance; smaller raise and less compute-heavy than a model lab.

Known for: OpenAI-backed agent-swarm company; 2,000-agent demos

In more detail: Builds software that coordinates thousands of AI agents working together on hard analysis and forecasting problems, starting with investment firms. An early demo used about 2,000 agents to forecast the price of gold.

Revenue: No

Backers: OpenAI, Stanley Druckenmiller, Michael Ovitz, Amity Ventures, others

Valuation notes: WSJ-reported / company-said $94M raise at $650M valuation; replacing the prompt's ~$1B rumor with a public figure.

Sources: wsj.com · chch.ox.ac.uk

Why this call: Researcher-founded, pre-revenue, training-heavy, model-first, big first round.

Known for: Stanford/Berkeley/Cornell team incl. Chris Ré (Together AI, SambaNova); claims 100-token answers where frontier models need 100k

In more detail: Builds AI memory that bakes a company's documents into model weights, so agents answer from learned knowledge instead of rereading files and use far fewer tokens. Partners include Microsoft, Notion and Harvey.

Revenue: No (early partnerships with Notion, Microsoft, Harvey)

Backers: General Catalyst, Kleiner Perkins, Sequoia, Amplify Partners, Factory, Neo; angels Andrej Karpathy, Assaf Rappaport, Pieter Abbeel

Valuation notes: Medium: $98M at a reported $600M post on emerging from stealth (Jun 25, 2026; Dealroom, neolabs.fyi). No primary press release with valuation.

Sources: dealroom.co · vktr.com

Why this call: Elite researchers, pre-revenue, compute-heavy world models, software first, $100M+ seed.

Known for: Hafner created the Dreamer world-model agents (DreamerV3); both ex-Google DeepMind

In more detail: Builds world models that let robots imagine and rehearse actions before doing them, extending the Dreamer line of research. Still in stealth and testing on humanoid robots, with no product yet.

Revenue: No

Backers: Andreessen Horowitz (reported lead), Khosla, DST Global, Striker; Sunflower Capital (per PitchBook)

Valuation notes: Weak: The Information (Mar 5, 2026) reported a $100M+ seed in talks led by a16z, no valuation. PitchBook shows a closed but undisclosed round naming Sunflower (~Oct 2026). Valuation imputed from $100M at ~17.5% seed dilution.

Sources: theinformation.com · my.pitchbook.com

Why this call: Researcher-led and pre-revenue, but a vertical industrial-robotics play that may build hardware; round not closed.

Known for: Ex-Tesla Optimus AI lead; UC Berkeley PhD in reinforcement learning; hired several ex-Optimus engineers

In more detail: Uses reinforcement learning to teach robots skilled factory and aerospace work, focusing on non-humanoid machines. Still in stealth and software-first for now, with no product announced.

Revenue: No

Backers: Not disclosed

Valuation notes: Weak: one paywalled report (The Information, Oct 7, 2026) says the seed 'could end up around $100M'; not closed, no investors named. Valuation imputed from $100M at ~17.5% dilution.

Sources: theinformation.com · dealroom.co

Why this call: Researcher-founded, model-only (no hardware), big seed; already selling an API, so slightly product-led.

Known for: Ex-Nvidia GR00T researchers (DreamZero, World Action Models)

In more detail: Trains robot AI on videos of people doing everyday tasks. Its first model, Dream-1, works across robots from many makers and is sold as an API that turns camera images and instructions into robot motions.

Revenue: No (Dream-1 API in paid beta since Oct 2026)

Backers: B Capital (lead), Kindred Ventures (co-lead), A.Capital, Unusual Ventures, 500 Global, SV Angel, Wing VC, Scribble Ventures

Valuation notes: $90M seed closed Jun 30, 2026 (PitchBook), announced Oct 19, 2026 (Robot Report, The AI Insider). No valuation disclosed; imputed from $90M at ~17.5% dilution.

Sources: therobotreport.com · theaiinsider.tech

Why this call: Elite researchers, pre-revenue, funds go to compute and research hires, software only, big seed.

Known for: Fidler was Nvidia VP of AI Research and founded its Toronto lab; U of Toronto professor; Vector Institute co-founder. Among Canada's largest seeds

In more detail: Builds world models that simulate the physical world so robots and other embodied AI can train in realistic virtual environments. No model released yet.

Revenue: No

Backers: Khosla Ventures (co-lead), Radical Ventures (co-lead), Shopify Ventures, Innovation Endeavors, OMERS Ventures, Scribble Ventures

Valuation notes: $90M+ seed closed Aug 19, 2026 (The Logic via Unite.AI; PitchBook). No valuation disclosed; imputed from $90M at ~17.5% dilution.

Sources: unite.ai · siliconangle.com

Why this call: Pure researcher lab, pre-revenue, trains its own models, big seed.

Known for: ex-DeepMind execs working on vision reasoning

In more detail: Builds AI models that reason with images and video, zooming in and re-checking visual details as they think. Its first model, Glimpse, launched in October 2026 as an API preview.

Revenue: No (Glimpse research preview, Oct 16, 2026)

Backers: Striker Venture Partners, Menlo Ventures, Altimeter (co-leads), Nvidia, 500 Global, Sunflower, AGI House, Jeff Dean

Valuation notes: $55M seed at $300M post (TechCrunch, Jul 16, 2026; Harmonic shows $80.4M across two rounds). Sheet's ~$500M kept per your rule, but the reported figure is $300M. First model Glimpse launched Oct 16, 2026.

Sources: techcrunch.com · vktr.com

Why this call: Research lab training computer-use foundation models from raw video.

Known for: FDM-1; video-first computer-use / computer-action model; six-person lab

In more detail: Trains AI models that learn to use computers by watching screen recordings, then act through clicks and keystrokes. Its model FDM-1 learned from 11 million hours of video.

Revenue: No

Backers: Sequoia Capital, Spark Capital; angels incl. Andrej Karpathy, Milan Kovac, Stanley Druckenmiller (reported)

Valuation notes: Raised $75M Series A at $425M pre-money / ~$500M post-money; round led by Sequoia and Spark; Standard/Sequoia describe FDM-1 trained on an 11M-hour computer-use video dataset.

Sources: sequoiacap.com · sequoiacap.com · si.inc · news.crunchbase.com

Why this call: Research team wrapping other companies' models rather than training big ones; low compute.

Known for: Ex-Google DeepMind team; ARC-AGI-2 record

In more detail: Builds a self-improving system that wraps existing language models like GPT, Claude or Gemini to make them reason better at lower cost. It set records on the ARC-AGI-2 reasoning benchmark.

Revenue: No (estimated)

Backers: FYRFLY Venture Partners and Surface Ventures (co-leads), Y Combinator, 468 Capital, Operator Collective

Valuation notes: $45.8M seed (Jan 30, 2026; SiliconANGLE); no valuation disclosed, ~$229M imputed by Harmonic at 20% dilution.

Sources: siliconangle.com

Why this call: Trains its own action models, researcher founder, but small raise and product/consumer oriented.

Known for: Div Garg: ex-MultiOn founder, Stanford PhD (paused); AGI-0 action models and REAL Bench benchmark

In more detail: Builds small on-device AI agents that read a phone or computer screen and tap through apps for you. Its model AGI-0 powers an Android app, and partners include Lenovo and Qualcomm.

Revenue: No

Backers: Menlo Ventures, General Catalyst, Anthropic, Point72 Ventures, Visa, Streamlined Ventures, AGI House, Berkeley Frontier Fund

Valuation notes: Last closed: $20M SAFE (Sep 29, 2025), no valuation. Forbes (Nov 2025, re-reported Jan 2026): talks to raise ~$50M at $500M, not confirmed closed. A '$1.92B' figure on an aggregator belongs to Brazil's Banco Agibank (also 'AGI Inc'), not this company.

Sources: forbesargentina.com · ai-market-watch.com

Why this call: Pure research team calling itself 'a lab first', but end use is proprietary trading and compute needs are modest.

Known for: Ex-DeepMind trio behind DeepStack, the first AI to beat professionals at no-limit poker; Richard Sutton is an investor

In more detail: Trains reinforcement learning agents that trade US stocks, building on the team's poker-beating AI research. Its agents trade in partnership with quant firm Tower Research Capital.

Revenue: Unknown (trades billions daily via Tower Research partnership)

Backers: Creandum (Series A lead), Blossom Capital (seed lead), Credo Ventures, RockawayX, K5 Global; Richard Sutton

Valuation notes: Series A led by Creandum at ~$500M post (€438M), PitchBook dates it Feb 12, 2026 ($32M); publicly reported Jun 30–Jul 3, 2026 (Dealroom, The Recursive). Seed: $10M at ~$140M.

Sources: dealroom.co · therecursive.com

Why this call: Consumer video product; no 2026 round found.

Known for: "Reality is optional"; Stanford AI PhDs; consumer video effects

In more detail: Makes a consumer app for creating and editing AI videos, images and audio. It mixes its own models, like Pika 2.5 and Pikaframes, with outside models from Google, ByteDance and others.

Revenue: Yes

Backers: Spark Capital, Lightspeed, a16z, Nat Friedman, others

Valuation notes: $135M raised total; $470M (June 2024, $80M Series B). No 2026 round found, below the $500M bar and pre-window. Consumer product, not a neolab.

Why this call: Pure research team, no revenue, model-first, large seed; compute appetite not yet proven.

Known for: Collins and Hughes did cooperative-AI research at DeepMind (Collins also White House AI policy); Kirsch ex-DeepMind meta-learning; Aleksiev ex-Reka/Microsoft

In more detail: Builds Faraday, an AI system of self-improving agents that works alongside human scientists on long-running, unsolved research problems.

Revenue: No

Backers: Index Ventures, Radical Ventures (co-leads), NVentures, Ex/Ante, Metaplanet; angels Matt Clifford, Thomas Wolf, David Luan, Dwarkesh Patel

Valuation notes: $50M seed closed May 27–30, 2026 (TNW, PitchBook). Valuation not disclosed; ~$250M is imputed (~17–20% dilution).

Sources: thenextweb.com · tech.eu

Why this call: Known researchers, no revenue, compute-heavy training, software only, large seed.

Known for: Co-creators of Evo and Evo 2 DNA models; three founders built core tech at Liquid AI

In more detail: Trains AI models that read DNA, RNA and proteins to understand whole biological systems, for drug discovery and biodefense. Its preview model, Omnii, pinpoints which DNA changes affect how genes are controlled.

Revenue: No

Backers: Emergence Capital (lead), Obvious Ventures, Triatomic, Factory, First Spark, Patrick Collison (pre-seed), AGI House

Valuation notes: Fortune broke the $50M seed (Jun 15, 2026). PitchBook records $245M post ($195M pre) as actual. Earlier ~$4.4M pre-seed at ~$54M (Mar 2026).

Sources: fortune.com

Why this call: SaaS/data-infrastructure vendor with customers, not a compute-heavy model lab.

Known for: UC Berkeley founders; OmicsBench agent benchmark

In more detail: Builds cloud software and AI agents that analyze biology lab data such as sequencing and imaging results. It also publishes benchmarks that test how well AI models handle biology tasks.

Revenue: Unknown ('generating revenue'; 300+ teams claimed)

Backers: Coatue and Lux (Series A leads), General Catalyst, Caffeinated Capital, Fifty Years, Hummingbird

Valuation notes: Weak: only confirmed priced round is a $28M Series A at ~$150M post (Jun 2022). A $130M Series B (Apr 2026) appears on one aggregator only; unverified.

Sources: venturebeat.com · bouncewatch.com

Why this call: Model-centric and pre-revenue, but not a researcher team and a small raise.

Known for: SubQ sparse-attention LLM claiming 12M-token context and ~1,000x attention compute savings; researchers have publicly asked for independent proof

In more detail: Builds language models with a new attention design that can read up to 12 million tokens at once at far lower cost. Its model SubQ is offered through an API and a coding agent, SubQ Code.

Revenue: No

Backers: HTwenty Capital, Streamlined Ventures, Coalition Operators, Javier Villamizar, Justin Mateen (JAM Fund), Gaingels

Valuation notes: $29M seed (May 5, 2026) confirmed by SiliconANGLE and PitchBook. No valuation disclosed; ~$145M is imputed at ~20% dilution (Harmonic's estimate).

Sources: siliconangle.com · runtimewire.com

Why this call: Research pedigree and own models, but an API product company with modest funding and compute.

Known for: Kyutai spinout commercializing the tech behind the Moshi speech-to-speech model; Zeghidour ex-Google Brain/DeepMind/FAIR

In more detail: Builds fast voice AI for real-time agents, covering text-to-speech, speech recognition and live translation. Products include Gradium Translate and Phonon, and customers include Renault.

Revenue: Unknown ('generating revenue'; Renault a named client)

Backers: FirstMark, Eurazeo (seed leads), Nvidia (extension lead), DST Global, Eric Schmidt, Xavier Niel, Rodolphe Saadé

Valuation notes: $70M seed (Dec 2025, FirstMark/Eurazeo) plus a ~$30M Nvidia-led extension (Jul 8, 2026) for ~$100M total. No valuation disclosed publicly; PitchBook estimates ~$357M post (€211M pre). Sources conflict on the July round's size.

Sources: sifted.eu · sifted.eu · firstmark.com

Why this call: Trains its own models with RL, but it's a vertical security vendor already selling to enterprises.

Known for: Out of stealth with a $60M Sequoia-led seed; team of ex-Google/DeepMind researchers and Unit 8200 alumni

In more detail: Trains a cybersecurity AI model on logs and network data to power defensive agents that learn a company's systems and catch and respond to attacks. It says it works with Fortune 500 firms.

Revenue: Unknown (claims Fortune 100/500 deployments)

Backers: Sequoia Capital (lead), Khosla Ventures, Coatue

Valuation notes: $60M seed announced Aug 10, 2026 (closed early 2026 per CEO), confirmed by Fortune, Calcalist, PitchBook. No valuation reported; ~$350M imputed at ~15–20% dilution (range ~$300–400M).

Sources: fortune.com · calcalistech.com · dealroom.co

Why this call: Researcher-led, new architecture, pre-revenue, compute-heavy pretraining, large early raise.

Known for: ~10 years at DeepMind (generative audio/image), then VP of AI at Microsoft AI; ICML and ICLR 2016 best papers

In more detail: Builds diffusion language models, which write text by refining a whole passage at once rather than word by word like today's chatbots. No model has been released yet.

Revenue: No

Backers: Accel (lead), GV (reported), NEA (reported)

Valuation notes: Bloomberg (Sep 25, 2026) reported talks for $100M+ led by Accel; Sifted (Oct 15) reported Accel leading a $60M first tranche. No valuation reported; imputed from $60M at ~17.5% dilution (~$570M if $100M fills). Close not confirmed by a primary source.

Sources: sifted.eu · finance.yahoo.com · startupresearcher.com

Why this call: Researcher-founded forecasting-model lab; small raise and unclear compute need.

Known for: Former Meta researchers building probabilistic forecasting models

In more detail: Trains forecasting models that give calibrated odds for future events in geopolitics and markets, with a timeline of what drives each estimate. It targets finance, defense, insurance and real estate.

Revenue: No

Backers: Felicis Ventures (lead), Yann LeCun, Jeff Dean; Andrew Bosworth advisor

Valuation notes: Bloomberg reported ~$50M raise at ~$335M post-money; founder LinkedIn post names broader founding team.

Sources: bloomberg.com · moneycontrol.com · linkedin.com

Why this call: Research-first, no revenue, software/agents; first product is data-center optimization and the seed is modest.

Known for: 'AI-native physics lab staffed by virtual physicists' (Emmy platform); open-sourcing 'Get Physics Done'

In more detail: Builds Emmy, a platform of AI virtual physicists that design physical systems and test ideas in simulation. Its first commercial use is optimizing power and cooling in AI data centers.

Revenue: No

Backers: Breakthrough Energy Ventures (lead), Variant (co-lead per PitchBook), Dragon Global, Robot Ventures, SV Angel, Balaji Srinivasan

Valuation notes: $58M seed (Sep 1, 2026) confirmed by press release, Unite.AI and PitchBook. No valuation reported; ~$300M imputed from $58M at ~15–20% seed dilution (range ~$290–390M).

Sources: prnewswire.com · unite.ai

Why this call: Scientist-founded lab, early revenue, model and agent focus, large seed.

Known for: For-profit spinout of nonprofit FutureHouse; Kosmos 'AI scientist'; Glyph open models for chemical patents (2026)

In more detail: Builds AI scientists that read research papers, analyze data and propose experiments. Its flagship, Kosmos, is used by Incyte, and it released Glyph open models for reading chemical patents.

Revenue: Unknown (sells Kosmos commercially)

Backers: Triatomic Capital, Spark Capital (leads), NVentures, Pillar VC, Susa Ventures, Striker, Hawktail; angels Jeff Dean, Dmitri Alperovitch

Valuation notes: Solid: $70M seed at $250M post (Dec 18, 2025; PitchBook $255M). No 2026 round found.

Sources: caproasia.com · trysignalbase.com