FOUR QUESTIONS TO VSORA'S CEO, KHALED MAALEJ

Khaled Maalej co-founded Vsora in 2015, after two decades in the semiconductor industry. With Vsora, he has set out to design a new generation of high performance silicon chips capable of meeting the exponential computational demands of Generative AI and Level 4-5 autonomous driving applications.

Today, Vsora stands as Europe's only alternative to the glants in high-performance chip design.

Vsora is a French company, with three centers of R&D in Vélisy, Grenoble and Nantes. The company's chips are designed to integrate seamlessly into next-generation AI servers and high-density data centers, supporting the broader ecosystem that combines renewable energy, grid resiliency, and computing efficiency.

Q What challenges does Vsora solve ?

KM: Artificial intelligence has created an unprecedented demand for computing power, which existing chip architectures can no longer sustain. Even the most advanced GPUs today only use about 10 to 20% of their theoretical capacity. This inefficiency makes large-scale AI applications prohibitively expensive to operate. Vsora's architecture overcomes this bottleneck - known as the 'memory wall' - and raises actual compute utilization to 50 or 60%. By doing so, we drastically reduce the cost per query and make AI economically scalable.

Q Can you tell us more about the memory wall ?

KM: Lowering the cost per inference is essential to building efficient data centers and enabling scalable AI deployment. It is often assumed that the cost of AI processing is driven by energy consumption. In reality, energy accounts for only 10 to 15% of the total cost of a query. The true driver is the amortization of the silicon Infrastructure - how efficiently each chip can be used over its lifetime. For example, every time OpenAI gains a client with ChatGPT, they lose money because of the cost of processing the inference. They cannot scale up, even with huge investments.

To make this business model work and be able to deploy AI massively, the price of each query has to go down. In Silicon Valley, you often hear people say that we have to lower the queries to 0.2 cents of US$ so that ads can pay for processing. Our goal is to maximize that efficiency.

Q How does this new architecture impact the energy equation ?

KM: Energy is critical for the engineering of data centers. Today, when one looks to develop a data center in a specific area, the first question one asks is: What is the energy capacity of this location - and what is therefore the compute capacity that I can host with this amount of power?

The other question one has to think about is: How far away am I from the end user? It would be impossible to concentrate data centers all in one location. It would simply create too much traffic. That is why developers are looking to be close to the end user, and finding solutions near cities, with empty offices for example.

Ultimately, by improving the ratio between energy and performance, we enable data centers to host more computation within the same electrical capacity. Energy thus becomes a capacity constraint rather than a cost constraint, and more efficient chips translate directly into more computing power per megawatt.

Q As a founder, how important is it for you to be developing a European solution? What are the next steps for Vsora?

KM: We are seeing a structural shift: more and more players are looking for alternatives to American solutions. The fact that we are European, not American, and that we offer an alternative to Nvidia and AMD, is very favorable for Vsora. It gives us a unique position — and not just in Europe. We are also seeing this demand coming from other parts of the world.

Clients want to diversify their options; they do not want to rely entirely on the same ecosystem. They tell themselves: “At least part of my infrastructure is not American — and that makes me feel more secure in the face of uncertainty."

Finally, the European vision is twofold. First, it is economic: Europe is investing massively in data centers, yet this capital today mainly feeds the American economy. That is a real issue. And second, it is about sovereignty: there is a growing need in Europe for technological independence. Both dimensions are essential.

Vsora’s immediate goal is to bring its first high performance chip to market and benchmark it on MLPerf, the global reference for AI processors. We expect to demonstrate that Vsora ranks among the top three worldwide in raw computing capacity—and number one in efficiency. To accelerate, we are preparing a new funding round in 2026, involving both European and international investors. Beyond financing, our ambition is clear: to build a profitable and sustainable model for AI computation—one where performance and energy efficiency go hand in hand.

First published in Scope Winter 2025