Skip to content
AI

FlexAI: Pioneering Universal AI Compute Infrastructure

Modern AI workloads were being forced to run on computing architectures designed decades ago. FlexAI wants to create a faster, more flexible system that enables rapid scaling.

Compute infrastructure has emerged as a critical bottleneck in the rapidly evolving AI landscape. While innovation in AI models continues at breakneck speed, the hardware and infrastructure needed to train and run these models remain constrained, expensive, and complex.

A few years ago, over cappuccinos in Paris, tech industry veterans Brijesh Tripathi and Dali Kilani began a conversation about this dilemma that led to the founding of FlexAI in 2023. The company emerged from stealth mode in 2024 with $30 million in seed funding, establishing co-headquarters in Paris, Bangalore, and the United States.

Their observation was simple but fundamental: Modern AI workloads were being forced to run on computing architectures designed decades ago.

"At FlexAI, we are aiming to build the first universal AI compute infrastructure," Tripathi said.

In France, the company's arrival was hailed as another validation of the country's AI ambitions. For Tripathi, the French connection was driven by what he views as some key strategic advantages offered by the role the government plays and the nation's strength in areas such as nuclear power and alternative energy to help fulfill the company's long-term ambition: a data center in a box.

Fulfilling this goal will require more than just technological innovation. Tripathi said the industry needs to rethink the way scaleups and infrastructure are funded. Having the right policies and a holistic view of solving the energy puzzle are essential.

"It needs to be supported by the government and financing has to be done differently," he said.

Origin Story

FlexAI co-founders Dali Kilani (left) and Brijesh Tripathi

Tripathi's career took him to Intel, where he served as General Manager and Chief Architect for AI and Super Compute Platforms, and before that as Vice President and CTO of the Client Computing Group. Kilani spent time at NVIDIA, gaming company Zynga, and health tech firm Lifen.

Since the creation of the company, the co-founders have been working to develop the first version of FlexAI's offering: a cloud platform that provides Workload as a Service (WaaS).

"We are building an AI cloud," Tripathi said. "You come on our platform. You bring your model and you bring your data, and we take care of the rest of the infrastructure maintenance."

The goal is to dramatically simplify AI infrastructure for customers.

"We set it up for you. Set up the whole software stack for you. We get it to a point where you can just run a single command line training," Tripathi explained.

Behind the scenes, FlexAI manages extraordinary flexibility in infrastructure. This flexibility gives customers options based on their priorities.

"If they say, 'Hey, I'm willing to pay more, but make it faster,' they always run on Nvidia. If they say, 'Hey, I'm willing to go slow, but make it cheap,' a micro run on AMD, and if they say, 'We don't care,' then it will find the most cost-effective for us and them."

The system also ensures reliability with self-healing capabilities. "If a GPU fails in the long run, the user has to go back and restart. They have to go back and restart from the previous checkpoint. We take care of that in our environment. The user never has to go back into anything else except to collect their output at the end," he said.

The Infrastructure Challenge

According to Tripathi, "the biggest challenge for enterprises to enter into AI or use AI is actually understanding what AI infrastructure they need to build. It's a complex thing. Today, if you want to build an AI system, you have to go figure out which GPUs you want to use, how big your cluster is going to be, and actually become experts at infrastructure."

This creates significant barriers for organizations. "One of the challenges that people are facing right now is that they don't have the talent. They don't have the expertise in building this infrastructure before they even get started with what AI tools to build," he said.

Tripathi notes how this affects even specialized teams. "If you take a look at Mistral, or OpenAI or whatever, their large language model teams, the majority of them are actually data center operators. Why? Because they have to figure out how to get this large AI cluster to operate and function better."

FlexAI's approach simplifies this: "Our entire job is to reduce that complexity. We take care of building the infrastructure, managing the infrastructure, and giving access to compute. Instead of thinking about GPUs and thinking about the number of GPUs, thinking about how to maintain, manage, and handle failures."

The European AI Infrastructure Opportunity

FlexAI's decision to base its operations in Paris reflects strategic thinking about the European AI landscape.

"One of the reasons why we are in Paris is because I could be building the same solution, same hardware and storage, same service plans in US, and I would still be competing with five or 10 other companies," Tripathi said. "I'm probably the only one doing it in Europe, so instead of trying to compete with 10 others, the customers look at us and say, 'You're my European provider.'"

The company has structured its global operations strategically. "We have focused mainly on software in the Paris area, and even in the Paris office, we have people from all over Europe... and in Bangalore, we are focused mainly on the system side. So, eventually, our long-term plan is to build our own systems. So we started building that capability in Bangalore and getting prepared for our long-term plans."

This European focus addresses sovereignty concerns: "100% of the workload, AI training, fine tuning, inference runs on American GPUs, 100%. 90% of the trainings for Europeans and modern companies, including the big ones in Paris and Germany, and in US."

Growth and Timeline

FlexAI has been growing rapidly and has clear milestones ahead:

"By the end of the year, we want to get to public beta. So this is a private beta. We wanted to have hand-selected customers only, so we looked at their workloads, we looked at what they needed, and kind of did a white cloud service. By the end of the year, we will actually open it up for general availability."

The company's seed funding is focused on its cloud service. "That only allows us to build our cloud service. That's it. That was the plan. That was very clearly said to our investors. We have to do a bit of analysis on what is going to happen in the future, so a little bit of modeling work and architecture work on the large data set and the box thing. But a lot of our investment is going to go towards building the cloud service."

Building a Partnership Ecosystem

Rather than viewing traditional European cloud providers as competitors, FlexAI is pursuing partnerships:

"Scaleway is potentially a partner for us, instead of competition, because there are going to be people, very many customers or players, who are happy renting GPUs at the raw hardware, but there are many, many customers that come to them and say, 'Hey, I don't want to deal with all this complexity. Do you have a service or managed service version of this?'" he said.

Tripathi describes their approach to partnerships: "We have partners in the US who also, they're also trying to bring your service onto our data centers, and our customers can access our hardware through your service. The whole idea is we are maturing a service, what hardware it runs on, and what hardware the customer uses, completely abstracted."

The Future: Data Centers in a Box

Looking beyond their current cloud service, Tripathi shared a more ambitious vision

"I want to build a data center in a box," Tripathi explains. "It's a shipping container-sized box that has racks of computers, power delivery, and cooling. It's like a refrigerator that you can install anywhere in the world, right next to power."

This approach could dramatically reduce deployment time. "Because it's a fully contained system, all you have to think about is a Starlink to the rest of the network, and you're done. Nuclear reactor feeding it, Starlink connecting it to the rest of the world, you have your data center up and running without asking anyone for any permits or any new designs or five-year cycles."

Tripathi also emphasizes energy efficiency: "One of the things that we are focused on is our energy efficiency for things that we do is going to be orders of magnitude better than what happens today... for the same amount of work, we consume up to two orders of magnitude less energy."

Comments

Latest