Emulate, a British artificial-intelligence startup developing world models, is in advanced talks to raise up to $700 million in a seed round, The Next Web reported, citing separate accounts from the Financial Times and Bloomberg. The company is only about a month old and remains in stealth, without a publicly described commercial product.
Former Google DeepMind researchers Jack Parker-Holder, Matthew McGill and Philip Ball founded Emulate in August. All three contributed to Genie 3, DeepMind’s world-model project. That pedigree matters because it is currently the clearest evidence of what Emulate may build; the startup has not disclosed its own technical architecture, target customers or pricing.
World models do something different from a chatbot. Given a description of an environment, a system generates a space that changes as a user or AI agent moves through it. DeepMind says Genie 3 processes each new action against the trajectory that came before, producing a navigable scene in real time rather than a fixed video. An agent can practice tasks inside the simulation while the model predicts what happens next.
That could make the technology useful for training robots and other autonomous systems without conducting every trial in the physical world. DeepMind says Genie 3 can run at 24 frames per second for a few minutes, but it also lists constrained action space and short sessions among the limitations. Those details describe the founders’ previous work, not a demonstrated Emulate product.
The financing would put an unusually large amount of capital behind a research team before buyers or a revenue model are public. TNW said Index Ventures and Lightspeed Venture Partners are expected to lead, with Creandum participating. The reported valuation is unsettled: TNW said the Financial Times put it at $3.7 billion before the new money, while Bloomberg reported $3 billion on the same basis. The terms have not been agreed and could change, and Emulate and the expected lead investors did not comment to Bloomberg.