Proprietary AI Models: The Strategic Risks for Businesses

The context: why Nadella's warning changes the game

On July 13, 2026, Satya Nadella made remarks that are spreading quickly across tech and management circles. According to TechCrunch, the Microsoft CEO raised an issue that many in Silicon Valley have long discussed in whispers. The major AI labs that sell proprietary models could act as a Trojan horse for their business customers.

So the issue isn't just the technical security of the models. It's about bargaining power, data sovereignty and the strategic freedom of organizations. Still, observers haven't missed the irony: Nadella runs Microsoft, a company that has invested billions in OpenAI. His words should therefore be read with a critical eye, without dismissing their substance.

For marketing and digital leads at Italian companies, this scenario has immediate implications. Many SMEs and mid-market businesses have already built third-party AI tools into their daily workflows. Understanding where the real risks lie is now a strategic priority.

The numbers that matter: AI vendor dependence in Europe

The available data paints a worrying picture. According to research by Gartner, by 2027 more than 70% of European companies adopting AI will rely mainly on solutions from a single vendor. This figure matters. It points to a concentration of dependence that reduces operational flexibility.

In addition, according to McKinsey analysis, organizations that don't diversify their AI stacks risk switching costs up to three times higher than those that took a multi-vendor approach from the start. And the cost isn't only financial: it includes retraining teams, migrating data and rewriting automated workflows.

For Italian SMEs, these numbers carry even more weight. Unlike large corporations, smaller organizations have fewer resources to manage complex technology transitions. The choices made today set the room for maneuver for years to come.

Anatomy of lock-in: how the proprietary trap works

Vendor lock-in in AI works differently from traditional software. First, proprietary models are trained on specific architectures. This makes it hard to port results to alternative systems.

Next, companies tend to build deep integrations with their chosen provider's APIs. Every automation, every AI-driven process becomes a node in the proprietary network. Over time, the perceived cost of change outweighs the potential benefits of diversifying.

There's also a less visible but equally critical dimension: data. Many contracts with major AI providers include clauses allowing company data to be used to improve their models. As a result, sensitive information about customers, processes and marketing strategies can indirectly become the vendor's asset.

Finally, there's reputational risk. If your AI provider suffers a breach, changes its policies or gets acquired, dependent companies are left exposed with no way to react quickly. AI risk governance is no longer a technical issue: it's a management issue.

Strategic reading: what Nadella's warning really means

Reading Nadella's statements takes two levels of analysis. On one hand, the Microsoft CEO clearly has an interest in promoting a more open, or at least more distributed, AI ecosystem. On the other, the substance of his warning reflects a real concern across the industry.

But it's important not to fall into the opposite trap. Avoiding proprietary models isn't a workable option for most companies. Open-source models bring deployment complexity, maintenance costs and performance gaps that not every organization can handle on its own.

The right approach, then, is a conscious hybrid strategy. Use proprietary models where performance justifies it, while keeping clear visibility into the data you share, the contract terms and the alternatives available. Likewise, invest in in-house skills that reduce dependence on any single outside partner.

For Italian digital marketing leads, this comes down to a concrete question: does my company know exactly what data it shares with its AI providers? And is there a plan B?

Work still in progress: AI governance in Italian SMEs

AI governance in Italian SMEs is, in most cases, still a work in progress. Many companies have adopted AI tools, for content production, for Google Ads campaigns, for data analysis, without defining clear internal policies.

Compliance is also tied to technology dependence. The European AI Act, which is being phased in, imposes transparency and traceability obligations on the use of AI systems. Companies with no visibility into their technology stacks risk running into trouble both strategically and from a regulatory standpoint.

At SHM Studio, we see this often in the digital marketing and SEO projects we run. The most advanced companies are starting to build an AI inventory: a register of the tools adopted, the data shared and the associated risks. Most, by contrast, still adopt tools ad hoc, with no strategic direction.

Operational implications: what to do now to reduce your exposure

Nadella's warning doesn't call for a radical response. It calls for a methodical one. Here are some concrete steps for marketing and digital leads at Italian companies.

• Map your current AI stack. Identify every AI tool in use, including those adopted by individual teams without IT oversight. This takes a cross-functional audit, not just a technical one.
• Review your contracts. Examine the data clauses in your agreements with AI providers. In particular, check who holds the rights to input and output data.
• Diversify gradually. You don't have to change everything at once. Still, adding at least a second provider for critical functions significantly reduces lock-in risk.
• Train your teams. Invest in your teams' understanding of AI models. That way, adoption decisions become more informed and less driven by vendor marketing.
• Integrate with your web and SEO strategy. Content generated with AI for SEO activities or your web presence must meet quality and traceability standards. Setting clear editorial guidelines is part of AI governance.

Also, if you run LinkedIn campaigns or other paid channels with AI support, check how the automation platforms you use handle user data.

Outlook for 2027-2028: toward a more regulated AI market

The European regulatory framework will keep getting tighter. Between 2027 and 2028, companies will have to show not only that they use AI ethically, but also that they have real control over the systems they adopt. So those who start building a structured governance framework today will gain a measurable competitive advantage.

According to analyses by Harvard Business Review, organizations that take an AI governance by design approach perform better in risk management and earn more trust from their stakeholders. What's more, pressure from end customers for transparency on how AI is used is growing across every industry.

In short, Nadella's warning is a useful wake-up call. Not because Microsoft is disinterested, but because the problem he describes is real. Italian companies that want to build a solid digital presence, through websites, SEO and digital marketing, need to put AI governance on their strategic agenda. Not as a bureaucratic exercise, but as a lever for competitiveness. To dive deeper into these topics, you can contact the SHM Studio team or explore the articles on our blog.