AI Infrastructure Week: OpenAI, Microsoft, Google, Slack
AI just got more infrastructural this week
If you run a B2B company in Europe, the signal from the last seven days is pretty clean: the AI market is shifting from model novelty to operating system territory.
That matters more than the usual hype cycle. We are now watching the major players fight over four layers at the same time: capital, compute, model ownership, and workflow control. That is where advantage gets built. Not in who has the flashiest demo on X. In who owns the stack that businesses actually run on.
This week gave us four big moves worth paying attention to.
1. OpenAI is no longer talking like a lab. It is talking like core infrastructure.
OpenAI closed a $122 billion funding round at an $852 billion valuation and framed the whole announcement around scale, distribution, and compute rather than pure research ambition. According to OpenAI, the company is now generating $2 billion in revenue per month, has more than 900 million weekly active users, and sees enterprise revenue above 40% of total revenue. TechCrunch called the announcement “less like a typical blog post than a draft of an S-1,” which is exactly right. This was positioning, not just fundraising. (OpenAI, TechCrunch)
The line that stood out most was OpenAI’s push toward a “unified AI superapp” that combines ChatGPT, Codex, browsing, and agentic workflows into one surface.
Here’s what works strategically about that move: once an AI vendor owns the front door, it can pull consumer usage into the enterprise, then upsell APIs, workflow automation, and infrastructure behind the scenes. That is a brutal flywheel if you can finance the compute.
For operators, the takeaway is simple. Stop evaluating OpenAI as “just another model provider.” It is trying to become the interface layer for how work gets done.
2. Microsoft is building more of its own model stack, even while staying tied to OpenAI
Microsoft AI launched three new foundational models this week: MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2. The move matters because it shows Microsoft pushing further into first-party multimodal capability instead of acting only as OpenAI’s distribution partner. TechCrunch reported that the models were built by the MAI Superintelligence team led by Mustafa Suleyman, with pricing designed to undercut rivals. (TechCrunch, Microsoft AI)
This is the important nuance: Microsoft is not replacing OpenAI. It is reducing dependency risk.
That is smart infrastructure behavior. If you sell AI into enterprises, you do not want your entire product roadmap, cost base, or negotiation leverage sitting inside one upstream relationship. You want optionality.
I’ve seen this pattern before in hosting and infrastructure. The winners are usually not the companies with the single best component. They are the companies that control enough of the stack to manage margin, resilience, and customer experience.
That is what Microsoft is doing here.
3. Google is pushing open-weight AI as a sovereignty play, not just an open-source flex
Google released Gemma 4 on April 2 and put real emphasis on enterprise deployment, sovereign environments, and agentic workflows. The official Google Cloud announcement highlighted Apache 2.0 licensing, 256K context windows, native vision and audio processing, and deployment options across Vertex AI, Cloud Run, GKE, TPUs, and Sovereign Cloud. (Google Cloud, Google Developers Blog)
That wording is not accidental.
Google understands the next buyer wave is not only startups looking for cheap inference. It is enterprises and public-sector organizations that want capable models without handing over control of data, infrastructure, or regional compliance posture.
For Swiss and European operators, that is the real story. Open weights plus sovereign deployment plus serious model quality is a much stronger commercial proposition than “here’s another model.” It maps directly to the ownership-over-convenience argument that more CIOs and boards are starting to make.
If your AI roadmap still assumes every serious workflow must run through a closed US-hosted black box, this launch is a reminder that the architecture choices are widening fast.
4. Salesforce wants Slack to become an agentic work surface, not a chat tool
Salesforce announced 30 new AI features for Slack, with the biggest change centered around a much more capable Slackbot. According to TechCrunch, Slackbot can now use reusable AI skills, act as an MCP client, transcribe and summarize meetings, coordinate across connected apps, and even monitor desktop context to suggest or draft follow-ups. (TechCrunch)
Read that again and strip out the marketing language. What Salesforce is really doing is trying to turn Slack into a lightweight operating layer for work orchestration.
This is where a lot of AI strategy gets confused. Companies think they need one giant moonshot. Usually they do not. What they need is a better interface for repetitive coordination work: summarizing, routing, drafting, scheduling, chasing, updating, and triggering the next step.
That is why this Slack move matters. Not because every feature will land perfectly, but because enterprise software vendors are racing to own the workflow surface where agents actually get used.
The bigger pattern: AI vendors are fighting for the stack, not the prompt
Put these four stories together and the pattern sharpens.
- OpenAI is buying time and compute to become the AI superapp.
- Microsoft is building first-party model leverage inside a broader enterprise platform.
- Google is making the sovereignty and deployment-control argument with open weights.
- Salesforce is embedding agents into day-to-day workflow surfaces.
That is the market now.
The conversation is no longer “Which model is smartest on a benchmark?” The real question is: Which vendor helps you own the workflow, the data path, and the economic upside?
For mid-market B2B teams, this is good news if you stay disciplined. The tools are getting stronger, but the strategy still needs to stay simple.
4 takeaways for operators
1. Treat AI like infrastructure procurement, not software experimentation
If the vendor becomes deeply embedded, switching costs will come from workflows and data movement, not just API calls. Plan accordingly.
2. Model quality matters less than stack fit
The best model on paper can still be the wrong choice if it creates compliance friction, cost volatility, or integration drag.
3. Sovereignty is moving from edge case to board-level requirement
European buyers increasingly want control, auditability, and optionality. Google knows it. Microsoft knows it. You should too.
4. The fastest route to proof is workflow-level deployment
Do not start with “enterprise AI transformation.” Start with one revenue, service, or operations workflow and get to measurable proof in 30 days.
The companies that win this cycle will not be the ones with the loudest AI narrative. They will be the ones that pick the right layer of the stack, wire it into a real business process, and compound from there.
