AI
MISTRAL SETS UP A PARTNERSHIP WITH THE US AI PLATFORMS SUPPLIER CLOUDERA
(Tribute to L’USINE DIGITALE for this very comprehensive discussion)
Mistral AI is continuing its push into the business world with a new partnership described as “strategic” with Cloudera. The American data platform company will integrate the French firm’s models and tools into its hybrid environment, giving its customers the option to run Mistral models on their own infrastructure, in the cloud, on-site, or in completely isolated environments. With Mistral, CLOUDERA’s clients will be able to use French-language models directly within their own data environment, without necessarily sending their information to an external AI API. The goal is especially to allow inference in a private setting, but also to enable customization and training of models using proprietary data.
This is where Mistral highlights its concept of ‘sovereign AI.’ The company explains that data can stay within client-defined boundaries, models can be adapted and kept as open weights, and both training and inference can be done on the infrastructure and within the jurisdictions chosen by the company.
And there starts the debate: Mistral can provide the models, open weights, and some deployment guarantees; the data management layer, however, remains in the hands of an American provider. The nuance is important. There’s a difference between operational sovereignty and technological sovereignty. In the first case, a company can actually control where its data is hosted, where inference is run, and who can access it. In the second, you also need to have control over critical technological components, infrastructures, providers, and the jurisdictions they fall under. The partnership with Cloudera fairly well addresses the first goal. It’s not enough to meet the second.
ECA comment: in a sense, it’s the same debate as the so-called “hybrid” cloud services as BLUE (Orange, Cap Gemini, based on Microsoft Azure technology) and S3NS (THALES, based on Google Cloud technology). Sovereignty means control over data, true. But it’s also a matter of technological autonomous capability …
AND ANOTHER ONE WITH NUMSPOT
The two companies propose a joint offer, “unique because of its level of integration”. It brings together, on a single platform hosted in France, the AI models and tools from Mistral AI, a sovereign cloud infrastructure, as well as all the services needed to use them in production, including the Numspot platform and its ready-to-use PaaS services, orchestration, storage, monitoring and security, with end-to-end operations handled by Numspot and governance suited to the most demanding environments.
In this way, Numspot and Mistral want to give their clients access to the entire technical chain, with the added benefit of having a single point of contact.
AND ONE WITH TOTAL ENERGIES
(Tribute to LE MONDE)
The French oil and gas giant TotalEnergies announced on Tuesday, September 15, a three-year partnership with the French artificial intelligence company Mistral, representing an investment of more than 100 million euros. The goal is “to generate multiple scenarios,” both “for developing exploration opportunities” and “for optimizing and extending the life of existing projects.” TotalEnergies and Mistral specify that a joint scientific laboratory will be set up. “Exploration and reservoir engineering are among the areas where artificial intelligence can create the most value for our activities”, said Patrick Pouyanné, CEO of TotalEnergies. “By combining a century of geoscience data, the expertise of our teams, and Mistral’s capabilities, we want to develop a new generation of tools that can help our experts analyze the most complex data and make decisions“, he explained, adding that “this partnership also shows [the company’s] desire to contribute to the development of a high-performing European digital ecosystem to support our competitiveness.“
𝗙ORTAEGIS RAISES $50M SERIE A TO ACCELERATE SECURE COMPUTE’s DEPLOYMENT
Fortaegis, a Dutch computing manufacturer, has raised an oversubscribed $50 million Series A, led by Serendipity Capital, to accelerate the commercial production and global deployment of our Secure Compute architecture. According to Fortaegis, “ AI, quantum computing and autonomous systems are fundamentally reshaping digital infrastructure. As compute becomes more distributed, intelligent and autonomous, security and trust can no longer be an overlay: they need to be built into the foundations of computing itself. That is what Fortaegis is building. Delivered through our unified Fortaegis Silicon Platform, the full-stack architecture spans silicon, firmware, cryptography and software, rooting security in the physical properties of silicon and establishing trust at machine speed and scale. The investment will accelerate deployment across the United States, Europe, Singapore and Japan, as we move from technology validation into commercial production, and scale our engineering and commercial capabilities globally.”
THALES DEVELOPS AI–BASED INTELLIGENCE FOR DEFENSE
Thales’ ‘HexaForce’ solution is meant to let the military bring together data from radars, drones, satellites, cyber software, or existing software all in one place. The company adds AI features to sort, cross-check, and translate this information to help operators prepare deployments, analyze terrain, or handle targeting objectives. HexaForce is a “C4l” system, standing for “Command, Control, Communications, Computers and Intelligence”, a software architecture designed to connect the data and applications used by headquarters. Thales says it was designed to work from the national headquarters all the way down to the battalion level.
A military operation produces and uses information from multiple sources (radars, drones, satellites, cyber systems…). These components may have been developed by different companies, use different formats, or be deployed on separate networks. HexaForce therefore provides a common environment where this information can be used.Moreover, as communication on the battlefield can be difficult, Thales has adopted a ‘data mesh’ type of architecture, with data access distributed across several nodes. The goal is to allow decentralized access to data, especially when communications with other sites are degraded.
AXELERA AI SPEEDS UP AI CHIPS DEPLOYMENT
Axelera AI, the Dutch chips manufacturer, launches ‘Europa,’ a new architecture designed for more demanding inference tasks in businesses and data centers. Unlike training, inference is using an already trained model to produce a result. This chip is meant to run language models, multimodal models, agent systems, as well as image and video analysis models. The company, which announces it has deployed products in more than 600 customers worldwide and has a book of 1,5 billion €, targets EU AI factories.It emphasizes the energy efficiency of its solution, saying that its Edge 232p card based on Europa delivers up to 6 times more tokens per second per watt than the competing platform used as a reference. Axelera is targeting use cases where models need to be run frequently or continuously, and where energy costs become a big deal.
CLOUD SERVICES
OVH CLOUD GETS THE SECNUMCLOUD CERTIFICATION
SNC Cloud Platform, the new platform by OVH Cloud, has received the SecNumCloud certification issued by ANSSI. It’s a recognition that meets the expectations of our most demanding clients, who are looking for a managed cloud with one of the highest levels of security on the market, without giving up agility.
QUANTUM COMPUTING
ALICE & BOB, a French start up,attended the recent HPC User Forum at Argonne National Laboratory, one of the biggest national labs in the US. They report: Quantum and AI used to be separate conversations at HPC events. Not anymore. The good news: nobody’s asking “does quantum belong in HPC” anymore. The report became a starting point for real discussion on what should (and shouldn’t) be standardized across the stack, how to evaluate AI’s role in these hybrid workflows, and which mix of expertise needs to be in the room to design the first useful HPC-Quantum-AI workflows.
AND PARTNER WITH CEA
Happy to reproduce a post by ALICE & BOB:
Alice & Bob’s quantum computing software just found a home inside HPC centres! We’ve partnered with the CEA (France’s leading scientific research body) to make this happen.
The goal is to make our cat qubit tech usable exactly where the real computational work already happens, not stuck in a lab.
Here’s what that means in practice:
Our software stack now plugs into Bull Qaptiva, so developers write code once and run it on quantum backends, on-premises or on cloud.
A joint program with the CEA on many-body physics, building algorithms for early fault-tolerant systems around 2030, aimed at real problems in materials, chemistry, and industrial simulation.
POST QUANTUM CRYPTOGRAPHY: TIME TO ACT, THE ECA IS FULLY ENGAGED
The French conversation on post-quantum migration is rich in method and thin on position. Practitioners explain how to build a cryptographic inventory, how to prioritise by data lifetime, how to hybridise. Almost nobody says where Europe actually stands. CYBERUN, an online cybersecurity publication, focuses its latest edition on the post quantum necessary transition. However, the eleventh CESIN barometer, published this year from 397 security leaders of large French companies, makes no mention of the cryptographic transition in its published results. Shadow AI is their first risk behaviour and digital sovereignty rises eleven points, while post-quantum migration sits outside the frame.
Is the situation so bleak ? Other figures counterbalance the first impression. A weekly measurement of the public surfaces of about 9,900 large companies and organisations across 145 markets, as of 13 September 2026, reaches about 4,000 European ones and finds about 1,600 of them negotiating a hybrid post-quantum key exchange. Four in five of those do it behind a content delivery or security platform, and France shows the same pattern, with 142 of the 365 reached ready and 114 of them behind a platform. What is visible is therefore less a decision taken in those companies than a maturity inherited from the infrastructure they buy. The number travels with its perimeter, as every number of this kind should: index-level scanning sees the exposed TLS surface, but not internal PKI, code signing, embedded systems or interbank links, which is precisely much of where harvest-now-decrypt-later actually lives.
So there is a gap between what is assumed and what is measured, and that’s exactly what a European observatory is for. It will deliver four outputs, none of them available today:
An annual readiness state of play, external measurement complemented by a short self-assessment on a shared maturity model, so the demand side appears at its real level.
A vendor-neutral map of the European supply side, read by the functions actors deliver rather than by the labels they carry.
Short interpretation notes when an episode moves the field, a cryptanalytic result or a supervisory letter, saying what changes and for whom.
And the questions the market has not claimed, signatures as their own problem, and the agent layer where credentials live for seconds and receipts must outlast the algorithm that signed them.
The method is as much the point: perimeters published, aggregation by default, a right of reply for every mapped actor, and the European doctrine stated where ANSSI and BSI diverge from NIST on hybrid mechanisms. No vendor can publish such a map about itself and be believed, and a consultancy produces it one client at a time. An alliance can hold it in the open.
The preparation committee, research, vendors, services and users, meets this month to settle what is covered and how the work runs, and what comes of it belongs to the group. So, keep posted !And, if interested, get in touch with Amin Hasbini, who carries the post-quantum and crypto-agility roadmap of the ECA cybersecurity focus group: ma@mahasbini.org
