Revolution Labs

Revolution Labs is an artificial intelligence startup founded in 2026 by Nando de Freitas, a machine learning researcher who worked at Google DeepMind for about a decade before a period at Microsoft.^[1] According to press reports, the company is building diffusion models for text generation.^[1][2] Diffusion language models generate text by refining a noisy or partly masked draft over several steps, producing blocks of tokens at once, instead of predicting one token at a time as conventional large language models do.^[3][4]

In September 2026 Bloomberg reported that de Freitas was in talks to raise at least $100 million initially for the company, with venture capital funds including Accel in talks to lead the financing.^[1] In October 2026 Sifted reported, citing sources, that he had raised $60 million for Revolution Labs in a round that Accel was set to lead.^[2] Both reports relied on unnamed sources, and no valuation was reported. The speaker list for KHIPU 2027, an AI meeting scheduled for April 2027 at the University of São Paulo, describes de Freitas as the chief executive of Revolution Labs.^[5]

Overview

Founder

Academic career

Nando de Freitas studied electrical engineering at the University of the Witwatersrand in South Africa, where he first trained neural networks as a final-year undergraduate project.^[7] He received a PhD from Trinity College, Cambridge, in 2000 for work on Bayesian methods for neural networks, was a postdoctoral fellow in Stuart Russell's AI group at UC Berkeley from 1999 to 2001, and was a professor at the University of British Columbia from 2001 to 2014.^[8] He later became a machine learning professor at the University of Oxford and a fellow of the Canadian Institute for Advanced Research (CIFAR).^[8] His doctoral thesis applied Markov chain Monte Carlo and sequential Monte Carlo methods, including particle filters, to the training of neural networks.^[9]

Google DeepMind

In October 2014 Google added two Oxford University spin-off companies to DeepMind. One of them, Dark Blue Labs, worked on deep learning for natural language understanding, and de Freitas was one of the four researchers named as coming from it; he also stayed on at Oxford as a part-time lecturer.^[10] His Oxford profile lists Dark Blue Labs among a few companies he spun off.^[8]

According to his speaker biography for KHIPU 2027, de Freitas served on DeepMind's research board for ten years and co-led Google DeepMind's GenMedia team, which delivered the Veo, Imagen and Lyria generative models. The same biography lists his contributions to projects including learning to learn, neural programmer-interpreters, reinforced self-training, AlphaCode, Genie and Gato.^[5] Bloomberg reported that he left DeepMind in 2024.^[1]

Selected papers co-authored by de Freitas:

Microsoft AI

A Microsoft profile published in January 2025 described de Freitas as a vice president at Microsoft AI and a member of its multimodal team, which works on perception and generation.^[7] On 1 August 2026 he wrote that it was his last day at Microsoft AI and that he was looking forward to his next AI project after a break with his family in the Argentinian Andes.^[19]

Founding and corporate details

A private company limited by shares named Revolution Labs Ltd was incorporated in England and Wales on 16 September 2026, nine days before Bloomberg's report. Its registered activities are research and experimental development on natural sciences and engineering, and business and domestic software development, and its registered office is in Altrincham, Cheshire.^[6] The incorporation filing names four directors, who were also the initial shareholders: Joao Ferdinando Gomes de Freitas, Caglar Gulcehre, Brendan Timothy O'Donoghue and Zi Yu Wang. De Freitas, Gulcehre and O'Donoghue each held 2,857,143 of the 10,000,000 ordinary shares in issue at incorporation, and Wang held 1,428,571.^[20] De Freitas's Cambridge doctoral thesis gives his name as João F. G. de Freitas.^[9]

Wang resigned as a director on 30 September 2026.^[21] The register records de Freitas, Gulcehre and O'Donoghue as persons with significant control, each holding more than 25% but not more than 50% of the shares.^[22] Bloomberg's report named only de Freitas in connection with the company.^[1]

Written resolutions passed on 30 September 2026 and filed on 19 October 2026 authorised the directors to allot new shares, disapplied pre-emption rights for those allotments, re-designated the existing 10,000,000 ordinary shares as "ordinary 1" shares, and adopted new articles of association.^[20]

Technical focus

Press reports describe the company's plans only in general terms. Bloomberg's sources said Revolution Labs would focus on AI diffusion models, which Bloomberg described as an alternative to the transformer technology that underpins large language models.^[1] Sifted said the company was building a type of diffusion model for text generation,^[2] and Dealroom's summary of the reporting said it would extend the diffusion approach, already used for images and audio, to text.^[23] None of these reports described a released model or product.

Conventional large language models are autoregressive: they generate text one token at a time.^[3] Google DeepMind describes diffusion models for text as learning to generate outputs by refining noise step by step, which lets them produce whole blocks of tokens at once and correct errors during generation.^[3] The approach changes how text is generated rather than necessarily replacing the transformer network itself. LLaDA, an 8-billion-parameter diffusion language model trained from scratch and described in a February 2025 paper, uses a Transformer to predict masked tokens, and its authors reported performance comparable to their own autoregressive baselines.^[4]

Other developers working on the approach include Google DeepMind, whose Gemini Diffusion is offered as an experimental demo,^[3] and Inception, which describes its Mercury models as diffusion LLMs.^[24] The diffusion language models article covers the wider field.

Funding

Accel declined to comment to Bloomberg, and de Freitas did not respond to its requests for comment.^[1] The same Bloomberg article said funding rounds for so-called neolabs were increasingly being split into tranches, with well-known investors coming in early at a lower valuation and later investors paying a higher price that raises the headline valuation.^[1] Sifted's $60 million figure appeared under a headline saying Accel was "set to lead" the round, while its article text said the money had been raised.^[2] No valuation for Revolution Labs was reported in this coverage.

Context

Bloomberg's September 2026 report placed de Freitas among a group of Google DeepMind researchers who had left to raise money for their own companies, many of them working on approaches other than large language models.^[1] It cited a June analysis by Radical Ventures finding that Google and DeepMind had produced more than 40 founders of so-called neolabs, more than OpenAI, Anthropic and Meta combined.^[1] Other examples in the article were David Silver's Ineffable Intelligence, which had raised around $1 billion; Thore Graepel's reasoning-model startup Metis Reasoning; and Emulate, founded by three former DeepMind researchers including Jack Parker-Holder, which was in discussions to raise $700 million.^[1] Demis Hassabis had dismissed concerns about a brain drain, saying at an event in Cannes that "there's a lot of talent movement between all the leading labs and we win our fair share of the top talent."^[1]

Debarghya Das's "The Ultimate Neolab List", an online list of AI research startups, includes Revolution Labs in its frontier-lab group under the label "Diffusion LLMs".^[25]

Timeline

See also

• List of neolabs
• Neolab
• Diffusion language models
• Gemini Diffusion
• Inception Labs
• Ineffable Intelligence
• Google DeepMind

References

1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17Yazhou Sun, Mark Bergen and Shona Ghosh. "Google DeepMind Exodus Sparks VC Frenzy for AI's Next Big Thing." Bloomberg, syndicated by Yahoo Finance (archived copy), 25 September 2026.
2. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10Maya Dharampal-Hornby and Freya Pratty. "Exclusive: Accel set to lead DeepMind alum's $60m round." Sifted, 15 October 2026.
3. ^1 ^2 ^3 ^4Google DeepMind. "Gemini Diffusion." Model page, accessed 27 October 2026.
4. ^1 ^2Shen Nie et al. "Large Language Diffusion Models." arXiv:2502.09992, February 2025.
5. ^1 ^2 ^3KHIPU. "Confirmed Speakers 2027." KHIPU 2027 speaker list and biographies, accessed 27 October 2026.
6. ^1 ^2 ^3 ^4Companies House. "REVOLUTION LABS LTD (17463119): company overview." Accessed 27 October 2026.
7. ^1 ^2 ^3Microsoft AI. "Nando de Freitas: A day in the life." 29 January 2025.
8. ^1 ^2 ^3 ^4University of Oxford, Department of Computer Science. "Nando de Freitas." Staff profile, accessed 27 October 2026.
9. ^1 ^2João F. G. de Freitas. "Bayesian Methods for Neural Networks." PhD dissertation, Trinity College, University of Cambridge.
10. ^1 ^2Samuel Gibbs. "Google buys two more UK artificial intelligence startups." The Guardian, 23 October 2014.
11. ^Scott Reed and Nando de Freitas. "Neural Programmer-Interpreters." arXiv:1511.06279, November 2015 (ICLR 2016).
12. ^Google Research. "Research at Google and ICLR 2016." Google Research blog, 1 May 2016.
13. ^University of Oxford, Department of Computer Science. "ICLR Deep Learning Conference - Best Paper Award." 21 April 2016.
14. ^Ziyu Wang, Tom Schaul, Matteo Hessel, Hado van Hasselt, Marc Lanctot and Nando de Freitas. "Dueling Network Architectures for Deep Reinforcement Learning." arXiv:1511.06581, November 2015 (ICML 2016).
15. ^TWIML. "Dueling Neural Networks at ICML, Plus Training a Robotic Housekeeper." This Week in Machine Learning & AI, 24 June 2016.
16. ^Marcin Andrychowicz et al. "Learning to learn by gradient descent by gradient descent." arXiv:1606.04474, June 2016.
17. ^Scott Reed et al. "A Generalist Agent." arXiv:2205.06175, May 2022.
18. ^Caglar Gulcehre et al. "Reinforced Self-Training (ReST) for Language Modeling." arXiv:2308.08998, August 2023.
19. ^1 ^2Nando de Freitas (@NandoDF). "It's my last day at Microsoft AI." Post on X, 1 August 2026.
20. ^1 ^2 ^3Companies House. "REVOLUTION LABS LTD (17463119): filing history." Includes the incorporation application (16 September 2026) and written resolutions passed 30 September 2026 (filed 19 October 2026). Accessed 27 October 2026.
21. ^1 ^2Companies House. "REVOLUTION LABS LTD (17463119): officers." Accessed 27 October 2026.
22. ^Companies House. "REVOLUTION LABS LTD (17463119): persons with significant control." Accessed 27 October 2026.
23. ^1 ^2Dealroom News. "Accel to lead $60M round for DeepMind alum's post-LLM startup." 16 October 2026.
24. ^Inception. "Inception: When Every Millisecond Matters." Company website, accessed 27 October 2026.
25. ^Debarghya Das. "The Ultimate Neolab List." Accessed 27 October 2026.