The use of AI as we know it today is only the tip of the iceberg. Tomorrow, users will not only request AIs in the cloud, but also on their company’s servers if the data is confidential, or even directly on their own workstation if there is no need to call a large language model. Frédéric Oster, Partner Account Manager at Microsoft, draws an analogy with hybrid cloud architectures that have emerged in recent years: “Software architectures are increasingly hybrid. Some processing is done in the Cloud, in very large data centers. These are notably the processes that handle enormous amounts of data. In parallel, machines are deployed as close as possible to the company’s processes, which is what we call Edge Computing. This movement foreshadows what AI and agentics will bring.”
The hybrid approach will prevail in AI
Some agents will operate in the back office, in the data centers, but others could be deployed closer to users for quicker responses and greater efficiency. “This approach also aims to move toward a more frugal AI, more aligned with the CSR commitments of companies,” adds the expert. “The energy consumption of AI can be contained through more optimized architectures that can leverage the computing capabilities of client devices.”
If generating a text summary is possible locally with a Small Language Model (SLM) executable on a Copilot+ PC, it is unnecessary to transmit the original document to the Cloud and run an expensive LLM to achieve the same result… This approach will put less strain on the network, consume less computing power by avoiding the use of a powerful GPU, and reduce the carbon footprint of the processing. An SLM will deliver an instant response, without needing to be connected. The user can always call on a more powerful LLM if they deem the result insufficient. This latter one could access various data sources and provide a more elaborate answer.
The agentic AI will accelerate the evolution of software architectures
In a very near future, a number of tasks will be processed locally, while others will need to be handled in the Cloud. An orchestrator will distribute tasks based on the complexity of the processing or the criticality of the data. All the intelligence of the architecture will reside in these orchestrators, which will know how to make the most of it. The role of the CxO (or CIO) will be to decide which foundational models (the LLMs) to allow access to, but also where the models will be deployed—with a public cloud provider, on sovereign infrastructure, internally in the company’s own data center, and now also on users’ devices. “The future won’t be about SSLs or LLMs, but about both,” adds Frédéric Oster.
This emerging evolution, aimed at supporting chatbot-type agents, will be accelerated further with the advent of agentic AI. The automation of major business processes will be carried out with AIs running in the Cloud, but also models localized on production sites, down to employees’ PCs, as close to the field as possible. “Local AI won’t do everything. Hybrid remains, in my view, the most energy-efficient option,” the expert notes.
The Copilot+ PC, a step toward the ultra-personal computer
The future lies in model hybridization, but businesses must be equipped to deploy models where they deem it most relevant to their strategy. “To implement agentic capabilities at the local level, it is essential to ensure there is enough computing power and, above all, memory on the machine to run such tasks,” recalls Gaby Mougeolle, IT Product & Business Management at Lenovo, “that’s why you should outfit yourself with a Copilot+ PC today to have the choice tomorrow.” A Copilot+-labeled machine delivers both the AI power required for today’s productivity tools and the guarantee to handle the rapid evolutions of AI. “Lenovo’s deep conviction about the future of AI is reflected in our logo: Smarter AI for All, with AI integrated into all our PC-IA lines. Qira is the best illustration of this strategy. Our assistant is available on all our terminals and our aim is to put ourselves in the user’s shoes and offer seamless AI across all these devices.”
Lenovo had already imagined this kind of seamless interactions back in 2024 with the launch of the Aura Edition PCs with Microsoft and Intel. They already featured the SmartShare function, which allowed easy transfer of files from one device to another. With Qira, Lenovo truly moves to the next stage—everything is now managed transparently, regardless of data format, whether it’s a text, a task, or a calendar entry. “AI will shape the interface according to its habits, its tasks,” explains the manager. At each work session, at every interaction, the machine will increasingly understand the user and adapt to their way of working. The computer will completely reconfigure itself depending on who logs in. “Today, whether it’s ChatGPT, Gemini or Copilot, AI has a persistent memory of its exchanges with the user. AI ends up knowing your writing style and adapts the generated text to what you have asked so far,” adds Frédéric Oster. “It knows who my N+1 is and will automatically insert their name when I refer to my boss. This approach will deeply reshape how we interact with our PC.”
Today, the personal computer is a fairly standardized machine that is very quick to replace with a new one updated with data stored in the cloud. The trend is that the PC will become more personal, somewhat like the smartphone, tightly tied to the user’s identity and activated by their fingerprint or facial image. As Lenovo CEO Yuanqing Yang explained at CES 2026, we are moving from the era of the Personal Computer to the Personalized Computer.
