Proprietary AI models: strategic risks for businesses

The context: why Nadella's warning changes the game

On July 13, 2026, Satya Nadella made statements that are quickly circulating in tech and management circles. According to reports by TechCrunch , the Microsoft CEO brought up an issue that many in Silicon Valley have been whispering about for a long time. The big AI labs selling proprietary models might operate as Trojan horse towards client companies.

Therefore, the issue isn't just about the technical safety of the models. It's about bargaining power, data sovereignty, and organizations' strategic freedom. However, observers can't miss the irony: Nadella runs Microsoft, a company that has poured billions into OpenAI. Consequently, his words should be read with a critical eye, though without downplaying their substance.

For marketing and digital managers of Italian companies, this scenario has immediate implications. In fact, many SMEs and mid-market companies have already integrated third-party AI tools into their operational workflows. Therefore, understanding where the real risks lie is now a strategic priority.

The numbers that matter: AI vendor dependency in Europe

Available data outlines a worrying picture. According to research by Gartner , by 2027 over 70% of European companies adopting AI will mainly use solutions from a single vendor. This data is significant. It points to a concentration of dependency that reduces operational flexibility.

Furthermore, according to the analysis by McKinsey , organizations that do not diversify their AI stacks risk switching costs up to three times higher than those who adopted a multi-vendor approach from the start. In particular, the cost is not just financial: it includes team retraining, data migration, and the rewriting of automated workflows.

For Italian SMEs, these numbers carry even more weight. Unlike large corporations, smaller businesses have fewer resources to manage complex tech transitions. Therefore, the choices made today set the limits of what they can do for years to come.

Anatomy of lock-in: how the proprietary trap works

Vendor lock-in with AI works a bit differently than traditional software. First off, proprietary models are trained on specific setups. This makes it tough to move your results over to other systems.

Subsequently, companies tend to build deep integrations with the chosen provider's APIs. Every automation, every AI-driven process becomes a node of the proprietary network. Thus, over time, the perceived cost of change outweighs the potential benefits of diversification.

Besides this, there is a less visible but equally critical dimension: data. Many contracts with major AI providers include clauses that allow company data to be used to improve the models. Consequently, sensitive information about customers, processes, and marketing strategies can become the indirect property of the provider.

Finally, there is the reputational risk. If the AI provider suffers a breach, changes its policies, or gets acquired, dependent companies find themselves exposed without tools for a quick reaction. Therefore, AI risk governance is no longer a technical issue: it's a managerial one.

Strategic read: what Nadella's warning really means

Reading Nadella's statements requires a double level of analysis. On the one hand, Microsoft's CEO has a clear interest in promoting a more open AI ecosystem — or at least a more distributed one. On the other hand, the substance of his warning reflects a real concern spreading across the industry.

However, it is important not to fall into the opposite trap. Avoiding proprietary models is not a practical solution for most companies. In fact, open source models come with deployment complexities, maintenance costs, and performance gaps that not all organizations can handle on their own.

Therefore, the correct interpretation is that of a conscious hybrid strategy . Use proprietary models where performance justifies it, while maintaining clear visibility over shared data, contractual terms, and available alternatives. Similarly, invest in internal skills that reduce dependence on a single external partner.

For managers Digital marketing Italians, this translates into a concrete question: does my company know exactly what data it is sharing with its AI providers? And does it have a plan B?

The still open construction site: AI governance in Italian SMEs

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

Furthermore, the issue of compliance is intertwined with that of technological dependence. The European AI Act , which has rolled out gradually, introduces transparency and traceability rules for using AI systems. So, companies that don't know what's in their tech stack risk running into trouble both strategically and legally.

We at SHM Studio we observe this situation frequently in projects of Digital marketing and SEO that we follow. The most advanced companies are starting to build a AI inventory : a log of adopted tools, shared data, and associated risks. In contrast, the majority still proceed through spontaneous adoption, without strategic direction.

Operational implications: what to do now to reduce exposure

Nadella's warning doesn't call for a radical response. It calls for a methodical one. Below are some concrete operational directions for marketing and digital managers of Italian companies.

• Mapping of the current AI stack. Identify all AI tools in use — including those adopted by individual teams without IT supervision. Therefore, a cross-functional audit is necessary, not just a technical one.
• Contract review. Analyze the data-related clauses in contracts with AI providers. In particular, check who holds the rights to the input and output data.
• Gradual diversification. It is not necessary to change everything right away. However, introducing at least a second provider for critical functions significantly reduces the risk of lock-in.
• Internal training. Invest in helping teams understand AI models. This way, adoption decisions become more conscious and less driven by vendor marketing.
• Integration with web and SEO strategy. AI-generated content for SEO activities or for the web presence must comply with quality and traceability standards. Therefore, defining clear editorial guidelines is part of AI governance.

Moreover, for companies using LinkedIn campaigns or other paid channels with AI support, it is advisable to check how user data is handled by the automation platforms adopted.

2027-2028 outlook: towards a more regulated AI market

The European regulatory framework will gradually become stricter. Between 2027 and 2028, companies will have to prove not only that they use AI ethically, but also that they have real control over the systems they adopt. Therefore, anyone who starts building a structured governance today will have a measurable competitive advantage.

According to the analyses of Harvard Business Review , organizations that adopt an approach of AI governance by design they record better performance in risk management and stakeholder trust. Furthermore, end-customer pressure for transparency on AI use is growing across all sectors.

All in all, Nadella's warning is a useful catalyst. Not because Microsoft is uninterested, but because the problem he describes is real. Italian companies that want to build a solid digital presence — through Websites , SEO and Digital marketing — must integrate AI governance into their strategic agenda. Not as a bureaucratic exercise, but as a lever for competitiveness. To explore these topics further, you can contact the SHM Studio team or explore the articles of the our blog .