French startup FlexAI exits stealth with M to ease access to AI compute | TechCrunch

French startup FlexAI exits stealth with $30M to ease access to AI compute | TechCrunch

A French start Huge seed investments have been made to “re-architect compute infrastructure” for developers looking to build and train AI applications more efficiently.

FlexAIThe company, as it's called, has been operating in stealth since October 2023, but the Paris-based company is officially launching Wednesday with €28.5 million ($30 million) in funding, while teasing its first product. Has been: An on-demand cloud service for AI training.

It's a small change for seed round, which usually means real coffee founder pedigree – and that's the case here. Co-founder and CEO of FlexAI Brijesh Tripathi He was previously a senior design engineer at GPU giant. Now AI Darling Before landing various senior engineering and architecting roles at Nvidia, Apple; Tesla (working directly under Elon Musk); Zouks (first Acquired by Amazon. autonomous driving startup) and, most recently, Tripathi was VP of Intel's AI and supercompute platform offshoot, AXG.

Co-founder and CTO of FlexAI Dali Kalyani Also has an impressive CV, serving in various technical roles at companies including Nvidia and Zynga, while most recently taking on the role of CTO. French startup Lifenwhich develops digital infrastructure for the healthcare industry.

The seed round was led by Alpha Intelligence Capital (AIC), Elia Partners and Heartcore Capital, with participation from First Capital, Motor Ventures, Partic and InstaDep CEO Karim Beguier.

The FlexAI team in Paris

The FlexAI team in Paris

The puzzle of computing

To understand what Tripathi and Keilani are trying to do with FlexAI, it's worth first understanding what developers and AI practitioners are up against in terms of access to “compute.” It refers to the processing power, infrastructure, and resources required to perform computational tasks such as processing data, running algorithms, and executing machine learning models.

Deploying any infrastructure in the AI ​​space is complex. It's not for the faint of heart, and it's not for the inexperienced,” Tripathi told TechCrunch. “You need to know a lot about building infrastructure before you can use it. “

In contrast, the public cloud ecosystem that has evolved over the past two decades serves as an excellent example of how an industry has emerged from the need for developers to build applications without of concern

“If you're a small developer and want to write an application, you don't need to know where it's going to run, or what the backend is – you just need an EC2 (Amazon Elastic Compute CloudFor example and you're done,” Tripathi said. “You can't do that with AI compute today.”

In the AI ​​realm, developers must figure out how many GPUs (graphics processing units) need to be connected to what kind of network, which is managed by a software ecosystem to configure. They are fully responsible. If a GPU or network fails, or if something goes wrong in the chain, it's up to the developer to fix it.

“We want to bring the AI ​​compute infrastructure to the same level of simplicity that the general-purpose cloud has achieved — after 20 years, yes, but there's no reason why AI compute can't have the same benefits,” Tripathi said. could see,” Tripathi said. “We want to get to the point where you don't need to be data center experts to run AI workloads.”

With a handful of beta users in the current iteration of its product, FlexAI will launch its first commercial product later this year. It's essentially a cloud service that connects developers to “virtual heterogeneous compute,” meaning they can run their workloads and deploy AI models across multiple architectures, using GPUs for dollars per hour. Instead of renting on a per-use basis, you can pay.

GPUs are important cogs in AI development, working to train and run large language models (LLMs), for example. Nvidia is one of the leading players in the GPU space, and one of the main beneficiaries of the AI ​​revolution. OpenAI and ChatGPT. In the 12 months since OpenAI Launched an API for ChatGPT in March 2023.allowing developers to bake ChatGPT functionality into their own apps, Nvidia's shares soared to nearly $500 billion. Over $2 trillion.

LLMs are moving out of the technology industry.The demand for GPUs is increasing rapidly. But running GPUs is expensive, and for small jobs or ad hoc use cases renting from a cloud provider doesn't always make sense and can be prohibitively expensive. Why is this? AWS is working on limited-time rentals for small AI projects.. But rent is still rent, which is why FlexAI wants to remove the underlying complexity and give users access to AI compute as needed.

“Multicloud for AI”

FlexAI's starting point is that most developers don't. Really Consider for the most part which GPUs or chips they use, whether it's Nvidia, AMD, Intel, Graphcore or Cerebras. Their main concern is developing their AI and developing applications within their budget constraints.

This is where FlexAI's concept of “universal AI compute” comes in, where FlexAI meets the user's needs and understands whatever architecture is needed for that particular task, taking care of all the necessary conversions across different platforms. I come allocates it. Intel's Gaudy Infrastructure, AMD's Rocm or Nvidia's CUDA.

“This means that the developer is only focused on building, training and using the model,” Tripathi said. “We take care of everything down there. Failures, recovery, reliability, are all under our control, and you pay for what you use.”

In many ways, FlexAI is setting out to fast-track for AI what's already happening in the cloud, which means more than replicating the pay-per-use model: it means leveraging tech on a variety of benefits. Ability to go “multi-cloud” by deploying different GPU and chip infrastructures.

For example, FlexAI will customize customer-specific workloads depending on what their priorities are. If a company has a limited budget for training and fine-tuning their AI models, they can set it up to get the most compute bang for their buck within the FlexAI platform. This might mean going with Intel for a cheaper (but slower) compute, but if a developer has a small race that requires the fastest output, they'll go with Nvidia instead. Can be transferred through

Under the hood, FlexAI is essentially a “demand aggregator,” renting hardware from traditional sources and using its “strong connections” with people at Intel and AMD to get preferential pricing. Saves what it spreads across its customer base. That doesn't necessarily mean sidestepping kingpin Nvidia, but it likely means a big one. Intel and AMD are fighting for GPU scrap. Left in Nvidia's perspective — there's a huge incentive for them to play ball with aggregators like FlexAI.

“If I can make it work for customers and get hundreds of customers on their infrastructure, they (Intel and AMD) will be very happy,” Tripathi said.

It sits in contrast to similar GPU cloud players in the space. Like a well-funded cover-view And Lambda Labswhich are focused entirely on Nvidia hardware.

“I want to push AI compute to the point where the current general purpose is cloud computing,” notes Tripathi. “You can't multi-cloud AI. You have to have specific hardware, number of GPUs, infrastructure, connectivity, and then maintain it yourself. Today, that's really the only way to get AI compute.

Asked who the exact launch partners were, Tripathi said he was unable to name them all due to the lack of “formal commitments” from some of them.

“Intel is a strong partner, they're certainly providing the infrastructure, and AMD is a partner that's providing the infrastructure,” he said. “But there's another layer of partnership that's happening with Nvidia and some other silicon companies that we're not ready to share yet, but they're all in the mix and MOUs (memorandums of understanding) have yet to be signed. have been. “

The Elon Effect

Tripathi is more than equipped to tackle the challenges ahead after working in one of the world's biggest tech companies.

“I know a lot about GPUs; I used to build GPUs,” Tripathi said of his seven-year tenure at Nvidia, which ended in 2007 when he jumped ship for Apple as it was. . Launching the first iPhone. “At Apple, I became focused on solving real customer problems. I was there when Apple started building its first SoCs (system on chips) for phones.

Tripathi also spent two years as the hardware engineering lead at Tesla from 2016 to 2018, where he worked directly under Elon Musk for his final six months after the two people above him abruptly left the company.

“At Tesla, what I've learned and what I'm taking to my startup is that there are no barriers other than science and physics,” he said. “The way things are done today is not how it should or needs to be done. You have to go after what is the right thing to do from first principles, and to do that, remove every black box. .

Tripathi was involved. Tesla's shift to making its own chipsa move that has since Copy from GM And Hyundaiamong other automakers.

“One of the first things I did at Tesla was figure out how many microcontrollers were in a car, and to do that, we literally had to set up these big black boxes with metal shielding. And there were casings around it to find really tiny microcontrollers,” Tripathi said. “And we put it on a table, held it out and said, 'Elon, there's 50 microcontrollers in a car. And we sometimes pay 1,000 times the margin on them because they're shielded and protected in a big metal case. .' And he's like, 'Let's make our own.' And we did.”

GPUs as collateral

Looking further into the future, FlexAI aspires to build its own infrastructure, including data centers. Tripathi said it will be financed with debt financing, building on a recent trend that has seen competitors in the space. including CoreWeave And Lambda Labs uses Nvidia chips. As collateral to secure loans – rather than giving more equity.

“Bankers now know how to use GPUs,” Tripathi said. “Why give up equity? Until we become a true compute provider, our company value is not enough to get us the hundreds of millions of dollars we need to invest in building data centers. If we just did equity, When the money is gone we disappear. But if we really bank it on GPUs, they can take the GPUs and put it in another data center.”

About the Author

Leave a Reply