The integration of artificial intelligence (AI) into healthcare is transitioning from theoretical promise to operational reality. While diagnostic algorithms and administrative bots have made inroads, the next transformative wave is centered on Large Language Models (LLMs) and their potential to deeply integrate into clinical workflows, from patient matching to decision support. The global AI in healthcare market, projected to reach staggering values by 2030, underscores this accelerating adoption. However, the path forward is not merely technological; it is fundamentally shaped by critical debates around data sovereignty, clinical trust, and regulatory compliance.

To understand the practical trajectory for 2026, Drug and Device World spoke with Danny den Hamer, Product Manager at myTomorrows, a company specializing in using AI to improve patient access to clinical trials. With a global footprint and a base in Europe and the US, myTomorrows operates at the critical intersection of cutting-edge AI application and stringent data governance, offering a grounded view of the trends that will define healthcare’s near future.

The Infrastructure Dilemma

The core challenge for hospitals in 2026 is balancing the immense potential of frontier AI models with the non-negotiable imperative of patient data privacy. As den Hamer outlines, this is leading to a strategic fork in the road for healthcare institutions.

The Case for In-House Expertise

For larger, well-resourced institutions—particularly leading university and specialist hospitals—2026 may see significant investment in in-house GPU capacity to run LLMs locally. The driving force is “sovereign AI,” a concept gaining immense traction in regions like the European Union. It emphasizes keeping sensitive patient data within the physical and legal jurisdiction of the healthcare provider.
“This hardware is extremely expensive, so you need to know the deployment is worth the investment. Beyond that, you need a comprehensive evaluation framework—benchmark datasets and tasks—to rigorously test that the AI you’re deploying locally meets your performance standards,” den Hamer states, highlighting that performance validation remains crucial even with local deployment. The appeal is clear: complete control over data flow, reduced external compliance overhead, and the ability to tightly integrate AI into proprietary hospital systems.

den Hamer was quick to caution for robust safeguards. “You also need robust data governance to ensure that data processed by your local LLM doesn’t leak externally, which is a risk if the model has agent-like capabilities, such as connecting to external tools or databases,” he says. Adding, “The key is that, even if the model runs locally, any communication with the outside world must be with a trusted party.”

However, the barriers are steep: massive upfront capital for hardware, specialized AI operations talent, and the ongoing complexity of maintaining governance.

The Trusted Partnership Pathway for the Majority

For the vast majority of small and medium-sized hospitals, building a local AI fortress will be impractical. Their path forward lies in forming “trusted partnerships” with external AI providers. Here, the criteria extend far beyond basic functionality. “A trusted partner should be transparent about the models they use, their sub-processors, their data retention policies, and they should have clear benchmarks with proven performance over time. You need this transparency to build trust,” den Hamer explains.
He did caution that, even when using an external provider, data leakage risks persist, especially with AI agents that can call external tools. “If it can communicate with an untrusted party in the outside world, there might still be data that’s leaking,” den Hamer notes. True trusted partnerships require closed, verifiable loops. Adding that, furthermore, you have to consider evolving “sovereign AI” policies, especially in Europe, which may influence preferred partners.

A Diverging Transatlantic Landscape?

den Hamer notes that the US and EU may emphasize different paths. In the US, den Hamer observes a likely concentration around large, compliant cloud platforms like Azure, GCP, and AWS, which offer robust security frameworks familiar to health systems. In contrast, the EU’s push for digital sovereignty and stricter data localization may foster a formal or informal “healthcare trusted” label, potentially favoring European AI providers like Mistral.

“For a global company like ours, we comply with the strictest standards—for us, GDPR first,” he says. Adding, “currently, to get frontier-level AI performance compliantly in the EU, we ensure all data and processing stays within the EU through strict agreements with cloud providers. Over time, as European AI players reach frontier capabilities, we may see a preference shift. But in healthcare, performance and reliability are paramount, so using the best available compliant model is often the priority.”

Software-as-a-Medical Device Comes of Age

Beyond where the AI runs, 2026 will be defined by what the AI is legally permitted to do. The key differentiator will be between general AI software and certified Software-as-a-Medical Device (SaMD).

Automation vs. Assistance

The fundamental distinction, as drawn by den Hamer, is the level of automation in clinical decision-making. Non-medical device AI software, like the tools used at myTomorrows, is designed for search assistance and reviewability. “We operate with a physician-in-the-loop model,” he says. Its value lies in augmenting human efficiency—sorting through thousands of clinical trials to surface potential matches for a physician’s final review.
A certified SaMD, however, would imply a higher standard with the capability for automated clinical decision-making. “The key distinction for a SaMD is that it involves real, automated decision-making for a patient. It might still have a physician overseeing it, but the software itself is making clinical decisions,” den Hamer explains. This could mean autonomously flagging a critical lab trend, generating a differential diagnosis, or recommending a specific treatment pathway within defined parameters. The certification from a body like the FDA or an EU Notified Body is a formal attestation to its safety, efficacy, and robustness for that specific intended use.

The Trust Dividend of Certification

In a market flooded with AI demos, certification provides a powerful trust signal. “The trust that comes from something being certified as a medical device… that will be the key differentiator,” den Hamer states. For physicians burdened with liability and for hospital procurement teams navigating vendor claims, a regulatory stamp of approval cuts through the noise. It signals validated performance, rigorous quality management systems, and accountability—factors that are paramount in life-critical environments. This trust will increasingly translate into commercial advantage, with certified SaMD products beginning to displace non-certified “demo-friendly” tools in serious clinical settings.

Strategic Trade-offs: Breadth vs. Depth

The choice between developing a broad-assistance AI or a narrow-certified SaMD involves a strategic trade-off. “A non-medical device AI platform can aim for breadth. myTomorrows, for instance, serves a global patient population across diverse diseases from cancer to diabetes. We use broad benchmarks and keep the physician in the loop for review of the search results. This allows for wide applicability,” den Hamer notes.

Achieving SaMD certification, conversely, typically requires a narrow, deep focus. “You typically have to start narrow, focusing on one disease, one region, or even one hospital’s data format. You can deliver deep automation and convenience for that specific group, but it takes much longer to scale and help a broader population,” den Hamer says. The development process is longer and more costly, focused on exhaustive validation for a specific use case and often a specific data format.

So, it’s a trade-off: breadth with human oversight versus depth and automation with a narrower initial focus.

The Impact on Patient Access and Clinical Trials

These infrastructural and regulatory trends converge powerfully in the domain of patient access to innovative treatments, particularly clinical trials. Here, AI’s potential to alleviate systemic bottlenecks is immense. Den Hamer describes myTomorrows’ two-step AI process: structuring disparate patient data (notes, records) into a coherent medical profile, and then matching that profile to a global database of clinical trial criteria. The search results are then reviewed by the physician, who decides the best way forward for their patients.
The problem it solves is one of human limitation: “Our core use of AI is matching patient profiles to global clinical trials—a complex, time-consuming task most physicians can’t do comprehensively. These trends shape how we deliver this capability. Through trusted partnerships, hospitals, especially small and medium-sized ones, can integrate our platform, instantly expanding the trials they can consider for their patients.”

“For hospitals with local AI, we can collaborate to deploy our software within their walls or, crucially, provide our benchmarks so they can test their internal models to ensure performance meets the required standard,” he adds. Both paths lead to the same goal: more clinical trial options being considered for more patients, breaking down geographic and informational barriers to access.

Key Takeaways for Healthcare in 2026

As Danny den Hamer notes, the year ahead will be characterized by maturation and operationalization. The relentless improvement of AI models is a given; the differentiator will be the framework for their deployment.

1. Privacy and Sovereignty Are Paramount: The decision between local LLMs and trusted partners will be a top-strategic priority for hospital leadership, dictated by budget, scale, and regional regulations.
2. Trust is Built on Transparency and Evidence: For physicians to adopt AI, they require more than marketing. They demand transparency on models, data usage, and partners, backed by published, credible evidence of performance on relevant clinical tasks.
3. Certification Creates a New Trust Tier: Regulatory approval for SaMD will begin to segment the market, creating a trusted tier for applications capable of automated decision-making and shifting procurement preferences in hospitals.
4. The Goal is Integration, Not Just Innovation: The most valuable AI will be that fits seamlessly—and compliantly—into existing clinical workflows, whether through APIs into EHRs, local deployments, or trusted cloud platforms. The focus shifts from what AI can do, to what it should do safely and effectively within the complex ecosystem of care.

In conclusion, 2026 will not be the year of AI replacing physicians. It will be the year where the foundational structures for its responsible, scalable, and trustworthy use in hospitals become solidified. The trends towards localized infrastructure and certified software represent the healthcare sector’s pragmatic answer to a powerful technology: embracing its potential while rigorously engineering the guardrails it requires. The outcome, as envisioned by pioneers in the space, is a future where technology systematically lowers barriers—making the best available treatment options, wherever they exist in the world, discoverable and accessible to every patient who needs them.