Homo numericus, Robo humanus
Just how far can we go?

David Gurlé embarked on entrepreneurship without a safety net in 2013, with the full support of his loved ones. Perzo, Symphony, hive (renamed hivenet), PoliCloud... The common thread? Secure communications that allow private and professional exchanges to remain confidential, a systematic compliance-oriented market approach and a clear tendency to shake up the status quo, notably by resolutely taking the opposite tack to hyperscalers. An occasional resident of the Côte d'Azur, David Gurlé shared his unencrypted views on the digital transitions currently underway with our editorial team.

You have helped shape the cloud and a large part of your business today is focused on developing an alternative to GAFAM. Do you think the distributed cloud that you have chosen to promote, which is less energy-intensive and less expensive than traditional centralised infrastructures, is struggling to gain acceptance?

I don't think so. We are right in the thick of it with hivenet and we can see what is happening in terms of adoption. Our distributed storage offering now has more than half a million users from over 190 countries and growth remains very strong. We're even having to slow down because we're encountering capacity issues in meeting demand. Quite honestly, I didn't expect such rapid globalisation. Of course, there are some countries where demand is higher than others, such as South-East Asia, Brazil, the United States and West Africa, but the trend is definitely global. People need storage and they want to pay less than the existing offers, which are mainly provided by hyperscalers such as Google or Apple.

You know, most of our customers don't even know that we do distributed cloud computing and, frankly, they don't care. They subscribe to our offers simply because we are cheaper. It works because of the cost. What you have to understand is that there is a paradox today. The price of data storage services is rising, while the price of hard drives is halving every four years. A few years ago, we were paying £10 for a basic storage service; today, the price is often £15. And yet, when we look at the prices of hard drives, we see a steady decline. In 2025, Seagate is selling its Exos M models at £0.015 per GB, whereas ten years earlier it was selling its SATA-3 at £0.30. The market is counterintuitive and customers are very price-sensitive.

Computing and storage requirements are increasing rapidly, driven by generative AI applications. What is your vision for the digital landscape in 10 years' time?

It's difficult to say what will happen in 10 years. I say that because 10 years ago, I would never have imagined being where I am today. However, I am certain of one thing. Many roles in companies will be replaced by AI. This is a very clear trend that has already begun. Business leaders are ultimately faced with a very simple choice. Who would you keep: someone who produces 100% of the time at 100 times less cost, or someone who works eight hours a day? It's going to be very difficult for many people to keep the same job they have today. To be honest, we have already started to reduce the workforce in my company, and I think this trend will accelerate everywhere.

In terms of the digital landscape, what seems to me to be the least anticipated today is the way in which AI is transcending the physical world. The impact of AI on the digital world is widely documented, so I won't go back over that. But for the physical world, it's still very abstract. Yet this is a truly significant shift that is taking place in parallel with the ongoing digital transformation that we are experiencing through our computers. AI is becoming advanced enough to combine mechanics with intelligence. We are entering the era of intelligent physical robots in all their forms, and the applications in this field are endless and will profoundly disrupt our social practices.

More and more robots will be used in domestic settings. Some will be programmed to assist elderly people in their homes, others will independently take care of the garden, others will iron your shirts or cook... These social revolutions will first take hold in the upper echelons of society, but then they will become more widespread. And it is this imminent arrival of intelligent mechanics in our daily lives that I still find difficult to imagine today. In any case, it is not yet widely publicised, and yet it is happening...

Should we set limits for ourselves?

No. And I don't think we should. Everything is being digitised and, whether we like it or not, we have become homo numericus. We are producing more and more data and will continue to do so. We are not at all in a period of digitisation deflation, quite the contrary. We are experiencing geometric progression. For me, setting limits would be limiting human progress. And I think that in the global context, it would be neither wise nor acceptable to set limits.

Like you, of course, I hear narratives of disaster scenarios. But I believe people are mistaken. AI has no objective. AI has no consciousness. And even though we have recently entered the agentic era, where one AI can talk to another AI, they use a common language that is framed by a Model Concept Protocol designed by human intelligence to enable them to interact more easily.

So, of course, we can legitimately ask ourselves whether there could be a possible scenario in the future where AIs break free from the MCP protocol and invent their own language in order to work faster. At that point, there would no longer be a common language between humans and machines. But the basic question remains: why would they do that? AI has no objective. Therefore, there is no possible desire for optimisation or efficiency on its part. AI is not human, and because its intelligence is precisely artificial, it does not reason like we do. It continues to respond to data, or we could say to a stimulus, according to a probabilistic model.

In contrast, the goal of humans is to survive, at all costs. So we reproduce, we seek to live in peace, we want to crush others... Catastrophic scenarios overlook this fundamental fact, which is that AI has no objective. I do not believe at all that one day, without us asking it any questions, AI will say to itself, « Hey... I have a goal... » I find it really hard to imagine this point of autonomous consciousness given the way artificial intelligence works today.

What are you working on at the moment?

In collaboration with Inria Sophia Antipolis, we are working on distributed AI. The idea is to take AI models, cut them into small pieces, distribute them across a variety of processors such as PCs, smartphones and servers, and achieve exactly the same performance.

We have asked other Inria branches to help us think about race conditions, which are complex algorithmic issues. Typically, when a file needs to be edited by several people at the same time, we find ourselves in a situation where several processes are trying to access the same source at the same time. How do we arbitrate this situation of simultaneous writing on the same text? It's very complicated, which is why we're getting help.

Another area we are working on with Inria Rennes concerns writing algorithms. It is important to note that in order to store data in a distributed network, a directory needs to be created where the data will be placed. This directory has a writing algorithm and a reading algorithm. When we launched hivenet, the number of users grew very quickly and we ended up with billions and billions of distributed file fragments that had to be placed somewhere and, above all, retrieved later. However, given the volume, we reached a point where the writing algorithm “forgot” to write a number of things. In other words, the model failed to keep up with the increase in load and was unable to adapt. As you can see, we are a long way from autonomous consciousness!

Since 2023, Inria and hivenet have been working together to develop distributed storage and computing technologies that reduce dependence on hyperscalers' cloud infrastructures. This approach is based on the use of decentralised computing resources, mobilising the computing power available on different types of hardware.

Two operational challenges have since been launched: the ALVEARIUM challenge (2023) and the CUPSELI challenge (2025). This public-private scientific collaboration brings together a large scientific community comprising 11 Inria research teams – ARGO, COAST, COATI, OCKHAM, MAGELLAN, MIMOVE, NEO, TADAAM, TOPAL, STACK and WIDE – from six research centres: Rennes, Bordeaux, Lorraine, Côte d'Azur, Lyon and Paris.

The dynamic is resolutely collaborative with leading academic institutions, including the University of Bordeaux, the University of Lorraine, the University of Rennes and the University of Côte d'Azur, as well as research organisations such as the CNRS, the Bordeaux Polytechnic Institute, the Mines-Télécom Institute and the École Normale Supérieure de Lyon.

The advances resulting from this work will be tested on cutting-edge infrastructure, notably the Jean Zay and SLICES computing platforms, ensuring concrete and operational applications of the research results.