New and recently funded AI Startups
1
Country: UK | Funding: $2.5B
Wayve is developing an end-to-end autonomous driving software Wayve AI Driver based on a proprietary AI model. This approach replaces the modular "sense-plan-act" architecture of the traditional AV1.0 approach with a single neural network trained on diverse data to transform raw sensor inputs into safe outputs. AV2.0 learns driving skills from raw, unlabeled data using self-training, eliminating the need for costly and time-consuming curation of labeled datasets. Wayve also uses generative models of the world for training, creating rich and realistic synthetic scenarios. Wayve's technology does not require HD maps, allowing it to be easily scaled to new roads and cities. Wayve AI Driver is sensor- and hardware-independent and compatible with any type of vehicle.
Wayve is developing an end-to-end autonomous driving software Wayve AI Driver based on a proprietary AI model. This approach replaces the modular "sense-plan-act" architecture of the traditional AV1.0 approach with a single neural network trained on diverse data to transform raw sensor inputs into safe outputs. AV2.0 learns driving skills from raw, unlabeled data using self-training, eliminating the need for costly and time-consuming curation of labeled datasets. Wayve also uses generative models of the world for training, creating rich and realistic synthetic scenarios. Wayve's technology does not require HD maps, allowing it to be easily scaled to new roads and cities. Wayve AI Driver is sensor- and hardware-independent and compatible with any type of vehicle.
2
Country: USA | Funding: $625M
MatX is an AI chip startup that designs chips that support large language models. The company’s goal is to make its processors 10 times better at training LLMs and delivering results than Nvidia’s GPUs.
MatX is an AI chip startup that designs chips that support large language models. The company’s goal is to make its processors 10 times better at training LLMs and delivering results than Nvidia’s GPUs.
3
Country: USA | Funding: $47M
Nimble is developing a platform that uses AI agents to search the web in real time, verify and validate results, and structure the information into tables that can then be queried like a database. The platform integrates with enterprise data warehouses and data lakes offered by companies like Databricks and Snowflake. This means its AI agents can connect to a company's massive data set, using it to create context and shape the structure and display of search results. These integrations also allow Nimble's software to learn constraints - for example, how a search should be performed or which data sources to use. This is particularly useful for applications such as competitor analysis, pricing research, know-your-customer processes, brand monitoring, in-depth research, and financial analysis.
Nimble is developing a platform that uses AI agents to search the web in real time, verify and validate results, and structure the information into tables that can then be queried like a database. The platform integrates with enterprise data warehouses and data lakes offered by companies like Databricks and Snowflake. This means its AI agents can connect to a company's massive data set, using it to create context and shape the structure and display of search results. These integrations also allow Nimble's software to learn constraints - for example, how a search should be performed or which data sources to use. This is particularly useful for applications such as competitor analysis, pricing research, know-your-customer processes, brand monitoring, in-depth research, and financial analysis.
4
Country: USA | Funding: $126M
Freeform is AI-native manufacturing-as-a-service that uses laser equipment for high-precision metal parts 3D printing. Its system consists of production equipment, computing power, sensors, control systems, software, machine learning stack and physical AI model. The system learns, predicts and controls the metal forming process in real time. Freeform produces for example rocket engine parts and Formula 1 racing cars.
Freeform is AI-native manufacturing-as-a-service that uses laser equipment for high-precision metal parts 3D printing. Its system consists of production equipment, computing power, sensors, control systems, software, machine learning stack and physical AI model. The system learns, predicts and controls the metal forming process in real time. Freeform produces for example rocket engine parts and Formula 1 racing cars.
5
Country: USA | Funding: $10M
Mirai allows to deploy and run models of any architecture directly on user devices. It developed fastest inference engine built from scratch for Apple devices.
Mirai allows to deploy and run models of any architecture directly on user devices. It developed fastest inference engine built from scratch for Apple devices.
6
Country: USA | Funding: $37M
Didero develops software that uses AI-powered purchasing agent to automate supply chain processes for manufacturers and distributors. It runs on top of an existing ERP system, acting as a coordinator, reading incoming messages and automatically performing necessary updates and tasks. The system analyzes natural-language communication with suppliers via email, WeChat, phone calls, as well as purchase orders and packing slips and automates a significant portion of the procurement process. The platform is designed for global companies that need to procure raw materials and components necessary for the production or sale of their products.
Didero develops software that uses AI-powered purchasing agent to automate supply chain processes for manufacturers and distributors. It runs on top of an existing ERP system, acting as a coordinator, reading incoming messages and automatically performing necessary updates and tasks. The system analyzes natural-language communication with suppliers via email, WeChat, phone calls, as well as purchase orders and packing slips and automates a significant portion of the procurement process. The platform is designed for global companies that need to procure raw materials and components necessary for the production or sale of their products.
7
Country: USA | Funding: $255M
Fundamental develops AI models to extract useful insights from the massive volumes of structured data generated by large enterprises. It combines legacy predictive AI systems with more modern tools. Unlike traditional LLMs, Fundamental's large table model, called Nexus, is deterministic - meaning it produces the same answer every time it's asked a given question, and does not use the Transformer architecture. Because Transformer-based AI models can only process data within their context window, they often struggle to analyze extremely large datasets, for example, a spreadsheet with billions of rows. However, such huge structured datasets are common in large enterprises, creating significant opportunities for models capable of handling such scale.
Fundamental develops AI models to extract useful insights from the massive volumes of structured data generated by large enterprises. It combines legacy predictive AI systems with more modern tools. Unlike traditional LLMs, Fundamental's large table model, called Nexus, is deterministic - meaning it produces the same answer every time it's asked a given question, and does not use the Transformer architecture. Because Transformer-based AI models can only process data within their context window, they often struggle to analyze extremely large datasets, for example, a spreadsheet with billions of rows. However, such huge structured datasets are common in large enterprises, creating significant opportunities for models capable of handling such scale.
9
Country: USA | Funding: $1.1B
Physical Intelligence develops general-purpose ML models for robots and other physical devices. The company collects data from robots in its lab and from real-world robots in other locations - warehouses, shops, homes - and uses this data to train universal base robot models. The goal is to create a pre-trained model (similar to GPT). The idea is that if someone creates a new hardware platform, they won't have to start collecting data from scratch—they can transfer all the knowledge the model already has. The company already works with a small number of companies in various industries - logistics, grocery stores and chocolate makers. When researchers train a new model, it comes back to stations like these for evaluation.
Physical Intelligence develops general-purpose ML models for robots and other physical devices. The company collects data from robots in its lab and from real-world robots in other locations - warehouses, shops, homes - and uses this data to train universal base robot models. The goal is to create a pre-trained model (similar to GPT). The idea is that if someone creates a new hardware platform, they won't have to start collecting data from scratch—they can transfer all the knowledge the model already has. The company already works with a small number of companies in various industries - logistics, grocery stores and chocolate makers. When researchers train a new model, it comes back to stations like these for evaluation.
10
Country: USA | Funding: $481M
Decagon provides a conversational AI platform for automating customer support across multiple channels.
Decagon provides a conversational AI platform for automating customer support across multiple channels.
