FlexAI, a French start-up created in July 2023, announced on April 24 a seed fundraising of 28.5 million euros. It is developing a virtual infrastructure solution for training AI models. Aimed at developers who don't have access to large amounts of compute, it aims to simplify the training process to make it as fast as opening a cloud instance.
No need to be an expert in GPU networking or memory allocation management, the customer will just need to specify their needs (model of such size with such priority) and FlexAI will take care of finding them. adequate resources, with usage-based billing rather than hourly GPU rental.
Advertisement
A heterogeneous infrastructure virtualized in multi-cloud mode
Similar solutions already exist (we think of CoreWeave or Lambda Labs), but they are limited to Nvidia GPUs. FlexAI stands out by offering to deploy the models on heterogeneous architectures, which can include Intel or AMD GPUs as well as Gaudi or Cerebras accelerators. Its software layer abstracts it away, allowing developers to concentrate on the essential: the creation of models and use cases.
Dali Kilani, co-founder and CTO of FlexAI, declared in a press release: “By making computing resources more accessible, we hope to spur an unprecedented wave of innovation, enabling businesses and developers to devote more of their budget and energy to innovation and research, rather than infrastructure spending.”
The funding round was led by the Luxembourg fund Alpha Intelligence Capital, the French venture capitalist Elaia Partners and the Danish Heartcore capital, with the participation of First Capital, Motier Ventures, Partech and Karim Beguir, CEO of InstaDeep. If FlexAI has achieved a very good seed round, it is because its founders Brijesh Tripathi (CEO) and Dali Kilani (CTO) did not come out of nowhere. Together they worked at Apple, Intel, NvidiaTesla and Zoox.
Launch planned this year
Their close links with the industry give them a legitimacy which should also allow them to be competitive on prices, in particular by taking advantage of the “non-Nvidian” resources of the market, which have struggled to attract until now. The near-perpetual supply shortage of Nvidia's AI-optimized GPUs will undoubtedly make things easier. Note also that Jim Keller, a major industry figure and CEO of Tenstorrent (which is developing an AI accelerator for data centers), is a member of the board of directors of FlexAI.
Advertisement
The FlexAI solution is currently in closed beta with a few hand-picked customers. It will go on sale “later this year”. The start-up, which has 40 employees including 25 in France, targets two categories of customers: on the one hand, European developers of large language models, who have access to sufficient computing resources but wish to increase efficiency and make turn their models faster. On the other hand, companies that do not have access to a sufficiently developed computing infrastructure.
FlexAI also plans to partner with large cloud providers, so that they can integrate its solution directly into their platforms. It notably cites Amazon Web Services and Google Cloud among its partners. It is also considering developing its own computing capacities, with its own data centers. To do this, it intends to launch a round of debt financing by the end of the year.
Do you want to stay up to date on the latest news in the artificial intelligence sector? Register for free to the IA Insider newsletter.
Selected for you