The facts: At its Build 2026 conference, Microsoft announced MAI-Code-1-Flash, its first proprietary model dedicated to generating code from natural-language descriptions. It is the first time since 2019 that Microsoft has introduced an in-house frontier model rather than relying exclusively on OpenAI. At the same time, the S-1 filed by SpaceX with the SEC revealed a colossal contract between Elon Musk and Anthropic : 300 megawatts of compute capacity at $1.25 billion a month through May 2029 — a total commitment of $45 billion. And OpenAI is preparing its own confidential IPO filing (Goldman Sachs and Morgan Stanley), targeting September 2026.

The Codito take: For three years, the enterprise AI market ran on a simple logic : you picked your model (Claude, GPT, Gemini) and plugged it into your stack. What changes this week is that Microsoft is durably refusing that dependence and would rather build its own models in-house. The practical consequence : if you are embedded in the Microsoft 365 ecosystem (Outlook, Teams, Word, Excel, SharePoint), you will no longer choose your model in absolute terms. By default, you will use the models deployed where you already work. And those models will increasingly be built in-house — Microsoft at Microsoft, Google at Google Workspace, Anthropic through its enterprise partnerships. Fragmentation by ecosystem has just become the new normal.

What it means for you: For a French SME — the vast majority of which run on Microsoft 365 — the strategic question to ask this week is no longer “ which model should I buy ? ”. It is : “ which models are coming to the tools I already use, and how do I want my teams to make the most of them ? ”. The right move is not to buy a separate Claude Enterprise subscription for 10 users if Microsoft Copilot with MAI-Code-1-Flash already covers 70 % of your use cases. Conversely, if your priorities justify specific models (research, business agents, advanced security), decide so knowingly — not out of habit.

Number of the week $45 billion. That is the total commitment revealed this week by SpaceX to supply Anthropic with 300 megawatts of compute capacity, at $1.25 billion a month through May 2029. For comparison : that is the equivalent of Renault's annual revenue. And it is what a single AI player will spend on servers over four years. AI is no longer a software story — it has become a question of energy infrastructure.

The Build 2026 conference confirmed something many leaders sense without daring to say it out loud : for most French SMEs, the best AI tool may well be the one they already have — or close to it. With Microsoft Copilot built into Outlook, Teams, Word, Excel and SharePoint, the strategic calculus has shifted. Here are three use cases that deserve your attention this week.

1. Your office staff (knowledge workers)

If your teams spend six to eight hours a day in Outlook, Word, Excel and Teams, Copilot becomes formidable. It drafts minutes from Teams conversations, suggests replies to your emails, turns a Word note into a PowerPoint deck, and produces Excel analyses without formulas. The ROI is immediate because adoption is immediate. Cost : between €22 and €30 per user per month depending on the licence. The opportunity calculation : if your staff save 45 minutes a day on document work, the tool pays for itself in the first month.

2. Your technical teams (developers, integrators)

MAI-Code-1-Flash, for its part, targets code specifically. It integrates with GitHub Copilot, Visual Studio Code and, in time, Microsoft's internal tools. For SMEs that develop in-house or do software integration, it is a genuine productivity shift — without replacing Claude Code or Cursor, which remain stronger on complex reasoning. Recommended approach : let your developers test the tools that best fit their workflows. Do not decide for them.

3. Your specific business cases (RAG, agents, sensitive data)

This is where Microsoft falls short — and it is important to acknowledge it. For business agents that need to reason over your proprietary data (bespoke quotes, lead qualification, case file analysis), and for sensitive deployments (law firms, investment funds, notaries), dedicated frontier models (Claude Enterprise, Mistral for sovereignty, GPT-5 Enterprise) still hold a clear edge. The trap would be to believe Copilot solves everything. It solves 70 % — and the rest calls for models chosen and deployed deliberately.

“ The right AI is not the most powerful one. It is the one that shows up where your people already work. ”

So the trade-off is not “ Microsoft or Anthropic ” — it is “ which tools for which uses, and in what order ”. The simple rule : start with what covers 70 % of your people without changing their habits, then invest specifically in the 30 % that justify a step up. Do it the other way round and you miss both.

Book a strategy call

Discover Codito Support

To close this edition, a fresh look behind the scenes — one that ties directly back to the two previous sections.

On Tuesday 9 June, we attended the Annual General Meeting of Hub France IA, the leading association of the French AI ecosystem, working closely with the Directorate General for Enterprise. Three topics came up insistently in the discussions : technological sovereignty, the cybersecurity of deployments, and training within companies. And behind each of them, the same shared observation — one it would be dishonest to play down : this technology frightens people.

What we see every day in the field — and what left us, as we walked out of the AGM, more convinced than ever — tells a different story. AI used well does not destroy jobs. It takes away the hours lost to data entry, copying and hunting for information, and gives people back the heart of their work and a better working life.

The difference between these two scenarios — the one that worries and the one that liberates — does not lie in the technology itself. It lies in how it is introduced : train the teams before equipping them, explain before deploying. That is what we structure every month with our clients through Codito Support : adopting it with your teams, not against them. The organisations that grasp this early will not only have saved time — they will have earned the trust of the people who work there.

Technology or teaching ? Beyond the technical topics raised, what really played out at the AGM was the shared realisation that the condition for a successful AI rollout is not its engine, but its teaching. That is exactly the stance we have taken from day one at Codito, and the one we structure month after month with our clients.