Microsoft ends OpenAI exclusivity — the AI vendor era is shifting
Microsoft, Google, and open-source rewrote the AI market in one week — single-vendor strategy is no longer safe.
Summary
- Microsoft launched 7 of its own AI models and ended OpenAI exclusivity — OpenAI is now also available through Amazon and Google.
- Google released Gemma 4 12B — a full multimodal model that runs on a 16 GB laptop, under an open Apache 2.0 license.
- Uber capped AI tool spending at $1,500/month per engineer per tool — the first public signal of what AI actually costs a company at scale.
The underlying read: single-vendor AI strategy is no longer safe. Three events in one week show that price, vendor, and deployment location are now choices — and you need to know your fallback.
1. Microsoft and OpenAI are no longer one
What changed. At Microsoft Build 2026, the company unveiled 7 in-house AI models, including the MAI-Thinking-1 reasoning model — direct competition with the OpenAI products into which Microsoft has invested $13 billion. In parallel, OpenAI began selling its API through Amazon and Google cloud services, ending a five-year exclusivity with Microsoft.
Why it matters. For most business users, “AI” meant “OpenAI through Microsoft” — and price and availability were not negotiable. Now Microsoft is trying to reduce its dependence and cut costs, while OpenAI is diversifying revenue ahead of a public offering. The result: both sides will compete for the same customer — yours — for years. That means two conflicting signals: downward pricing pressure (good news) and two divergent product roadmaps (bad news if you’re locked into one).
What to do this month.
- Compile a list of which business processes currently use an AI model — and specify WHICH model.
- Check whether your integration (or your vendor’s solution) is built so the model can be swapped in 2 weeks without code rewrites.
- Before signing any new AI contract over 12 months, demand a multi-vendor clause and data exit terms.
What I expect.
- Within 60 days — smaller, “MAI”-based variants in Microsoft Copilot pricing.
- Within 90 days — several European SaaS vendors announcing “model choice” as a feature.
- Within 180 days — first migration stories from Microsoft AI to Google or Anthropic on price.
2. AI on your laptop — Google opened the gate
What changed. Google released Gemma 4 12B — a multimodal model that handles text, images, audio, and video, running on a standard laptop with 16 GB of memory. The model is available under an Apache 2.0 open-source license — meaning a company can deploy it on its own hardware without sending data to a third party.
Why it matters. Until now, “use AI” practically meant “send data to a US cloud.” Compliance, customer contract clauses, and GDPR were the main blockers for many businesses. Gemma 4 is not the best model in the world, but it is the first to solve a simple set of tasks — document processing, support letters, office audio notes — inside your office, without data leaving. That opens a specific category of work that legally could not use AI at all before.
What to do this month.
- List the workloads where AI is not used today purely for confidentiality reasons (legal documents, customer data, finance).
- Ask your IT team (or partner) to pilot Gemma 4 for one concrete use case — a proof of concept in a week.
- Review vendor contracts where the AI clause says “cloud only” — express openness to a local option.
What I expect.
- Within 30 days — first compliance specialists will recommend Gemma 4 as an option for legal and medical firms.
- Within 90 days — Microsoft and OpenAI will respond with smaller, locally deployable models.
- Within 180 days — open-source models will take 30%+ of new enterprise AI projects.
3. Uber shows what AI engineers actually cost
What changed. Uber capped AI tool spending at $1,500/month per engineer per tool (Cursor, Claude Code, etc.) — after the company burned through its full annual AI budget in 4 months. 95% of Uber engineers use AI tools, and roughly 10% of code is now written by AI.
Why it matters. For the first time we see publicly what AI actually costs a company where everyone uses it. $1,500/month × engineer is roughly $18,000/year — until now, many leadership teams missed this because AI bills were spread across departments, credit cards, and experiments. The Uber number is a reference point — if you have 10 developers, AI tools alone cost $180,000/year at full use.
What to do this month.
- Ask finance for a complete list — every AI tool payment in the last 6 months, broken down by tool and department.
- Set caps per tool per user (Uber model: $1,500/month) with override permission requiring manager approval.
- Understand what productivity is being bought — how many hours of work AI is replacing, compared to the bill.
What I expect.
- Within 30 days — several European tech companies will publish similar internal caps.
- Within 60 days — first AI tool vendors will offer “team plans” with fixed pricing to stop runaway bills.
- Within 90 days — productivity metrics (code volume, document volume) will become standard talking points in AI contracts.
Today’s picture
One week — three events that all say the same thing: the choice of AI vendor has gone from forced to strategic. Microsoft and OpenAI are no longer one company. Google has put a full model on your laptop. And Uber has publicly shown what AI actually costs when everyone uses it. For three years AI was “plug it in and use it” — now begins act two, where you have to choose which vendor, on which deployment, at what price.
| Event | What it means for you |
|---|---|
| Microsoft–OpenAI exclusivity ends | Your AI integration is not immovable — demand exit clauses |
| Gemma 4 open-source on a laptop | Local AI is now real for confidential data |
| Uber $1,500/month cap | Find out what you already spend — and how much productivity you get |
Three questions for the leadership table:
- Which business processes today depend on a single AI model, and how quickly could you switch?
- Which work in your business does not use AI purely for confidentiality reasons — and which of those could run locally?
- How much do you currently spend on AI tools per month, and what productivity gain comes with that bill?