Introduction
The integration of artificial intelligence (AI) technologies in healthcare has progressed beyond mere scientific experiments and pilot projects, reaching a stage of maturity and practical application.
Just two or three years ago, the use of AI technologies was primarily viewed as a regulatory requirement for the digital transformation of healthcare, mostly pursued by innovative developers. Now, at least in public healthcare, AI solutions have become integrated into medical information systems (MIS), making them widely accessible to practicing physicians. Over 80 regions of the Russian Federation are using various medical AI- products for analyzing medical images and electronic health records (EHR).
This achievement is the result of a comprehensive set of initiatives related to AI in medicine implemented by the government since 2018. Key measures include the development of regulatory frameworks, the establishment of national standards, and the enhancement of the state registration process for medical devices with AI. Another significant driver has been government procurements and the deployment of AI solutions across regions, which began in 2023, followed by Federal Incident No. 11 concerning the monitoring of AI medical devices, launched in the fall of 2024.
Despite this progress, we are still in the early stages concerning market adoption. There are many areas where AI products are either absent or available only in sparse versions. The utilization of AI by physicians has yet to reach 100%, with various medical specialties and conditions still lacking AI applications.
We anticipate that by 2030, the Russian healthcare AI market will grow significantly. This expansion will lead to structural changes in the players, products, and scenarios for AI application. AI solutions are poised to transition from being informal "advisors" to essential tools integral to nearly every process.
In this publication, we aim to highlight the most significant trends regarding the development of artificial intelligence in healthcare between 2026 and 2030.
Shift to Large Generative Models
- The deployment of generative AI technologies, particularly large language models (LLMs), allows new functionalities to be added to AI solutions much more quickly and cost-effectively than traditional machine learning models or expert systems.
- For certain standard tasks, such as creating symptom-checking services for preliminary diagnoses or systems for selecting medical prescriptions, the traditional expert-led approach required collaboration among hundreds of specialists over several years—an investment that seldom aligned with business expectations, relegating it to an academic endeavor rather than a commercial one.
- The shift to data-driven AI and machine learning technologies has reduced the resource requirements for developing AI systems to just a few domain specialists, with development timelines shrinking to just a few months, making it feasible for startups and sparking a boom in companies offering healthcare-focused AI solutions.
However, we believe this approach is nearing the end of its dominance and will evolve into a hybrid model where LLMs are combined with rule-based recommendation algorithms and traditional ML models for predictive tasks.
Currently, creating a prototype of an AI product based on an LLM can require just one LLM engineer and one clinician for quality assurance, with MVP versions potentially being developed in a matter of weeks—significantly faster and cheaper than those based on traditional ML methods.
With appropriate levels of standardization and integration of AI in not just product functionality but also in development and documentation processes, the time required to release an MVP could be shortened to as little as two weeks.
Previously, establishing healthcare AI systems necessitated creating specialized companies to accumulate unique, costly competencies; the time to market could extend to two to three years or more.
Now, thanks to accelerated and cost-effective processes for creating AI solutions, the number of varied developments is set to increase rapidly. However, due to the nature of LLM technology, the quality and safety of the produced AI products may proportionately decline.
As long as hype surrounds AI, the healthcare sector will remain a highly attractive field for developers, including those with dubious ethics, who may ignore the standards of evidence-based medicine, offering hastily constructed "quick-fix" products.
The high risk of potential health harm due to issues like "hallucinations" will underscore the necessity of developing specially fine-tuned industry-specific LLMs for healthcare. Such models should form the foundation of future AI solutions, possibly supported by legislative restrictions on the use of general-purpose LLMs in healthcare products, particularly in clinical decision support systems (CDSS) and patient digital assistants.
Developing these industry-specific LLMs will require a sophisticated fine-tuning process on Russian medical data and the establishment of knowledge bases through Retrieval-Augmented Generation (RAG). These measures can help mitigate the risks of incorrect or dangerous recommendations from generic LLMs.
Based on these forecasts, we believe that soon Russia will develop an industry-specific LLM model based on existing Russian commercial LLMs or open-source projects (e.g., DeepSeek, Llama, Qwen, etc.).
Regulatory changes will be necessary; on the one hand, regulations should allow the registration of LLM-based AI solutions as medical devices and include them under the definition of medical products intended for patients. On the other hand, regulations will need to govern the circulation and market access in Russia of solutions that utilize an approved industry-specific Russian LLM housed in a secure data center located in the country.
General-purpose LLM solutions, especially American ones, as well as commercial developments, are likely to face either restrictions or possibly outright bans for use in healthcare.
Transition to AI Agents
The associated trend of moving toward autonomous AI agents will further develop and eventually dominate the market.
AI agents, rapidly constructed within specialized frameworks using LLMs, will be capable of independently executing complex, multistep tasks, planning activities, and adapting without continuous human oversight. They will infiltrate all popular AI application scenarios, from chatbots to virtual colleagues integrated into business processes—such as conducting automatic background analyses of entries in electronic health records and alerting physicians to critical findings.
AI agents will also be actively utilized in personalized digital assistants for patients, including remote monitoring for those with chronic illnesses, helping them generate clear and appropriate recommendations for examinations, treatments, lifestyle changes, and more.
This technological trend could significantly revitalize certain segments of digital healthcare that have seemingly stalled in their development. Telemedicine, for instance, has yet to achieve substantial growth in usage and popularity, largely because current telemedicine services necessitate a consulting physician's involvement. Transforming this model to incorporate AI agents replacing, at least partially, telemedicine consultations between "Patient" and "Physician" could yield substantial market gains by reducing the costs of such consultations.
Transition to Cost-Efficient Autonomous AI
A primary challenge facing healthcare is the persistent financial and staffing deficits. Despite government social policies and regulatory efforts to address workforce issues, these challenges are likely to remain or even worsen.
As industry efforts successfully increase life expectancy, the proportion of elderly patients in the population will rise. These individuals will simultaneously require more medical attention, physician visits, medications, and monitoring, while facing significant limitations in accessing paid healthcare, leading them to rely more heavily on public medical organizations.
These macro trends will create a growing demand for free medical care, particularly within primary healthcare. Addressing the rising expectation for state expenditure and the increasing number of physicians and nurses will necessitate fundamentally new approaches to organizing healthcare, incorporating digital transformation and AI.
Consequently, the primary requirement of AI solutions, both now and in the foreseeable future, will be to reduce costs and staffing needs.
However, current AI products often do not meet these demands. In fact, they frequently exacerbate workforce issues and costs rather than alleviating them.
Historically, AI solutions in medicine have been developed based on technical capabilities and data availability for machine learning rather than on critical sector needs. Consequently, around 70-80% of all registered AI medical devices worldwide, including in Russia, focus exclusively on analyzing medical images, particularly radiology. This market is already "overheated," filled with numerous nearly indistinguishable products.
Developers offer these solutions to physicians as a "second opinion,” meaning no structural changes to costs or staffing occur. Medical care continues to be delivered based on traditional principles, with physicians maintaining responsibility for patient interactions, data analysis, diagnoses, and treatments. AI simply verifies the gathered data and offers recommendations without reducing the overall reliance on physicians or removing the responsibility for potential medical errors.
The existing emphasis on AI usage for image analysis and electronic health records only solidifies this problem, as contemporary AI systems identify high-risk patients or early-stage pathologies, ultimately requiring additional resources for evaluation and treatment.
Combined, these factors lead us to conclude that the current implementation of AI in diagnostic and therapeutic processes does not actually alleviate the burden on the healthcare system. Furthermore, integrating AI solutions incurs additional expenses for product procurement and ongoing technical support, integration with existing MIS, user training, and oversight of effectiveness and safety.
As a result, there currently exists a gap between the healthcare system's pressing need to reduce costs and staffing and the actual outcomes AI products can deliver.
This is a serious issue. In our view, it can only be resolved in one way: we must transfer some of the functions and responsibilities for delivering medical care from physicians (and nurses) to AI. This, in turn, implies that we need to start utilizing autonomous AI agents.
Only such a reassessment of AI application scenarios will offer substantive value to healthcare organizers, encouraging them to increase expenditures for AI solutions and adjust regulations to shift from restrictive measures to incentives.
We believe that the demand for financial and staffing savings, coupled with increased competition and technological advancements in AI, will compel developers to rethink how AI can be applied and reevaluate payment models for AI solutions. In the future, AI should not generate extra costs but rather reduce them—either by lowering the costs of medical diagnostic processes or by eliminating unnecessary or unjustified expenditures.
A similar shift must occur regarding staffing: AI products should assume some of the roles and workload of physicians and nurses, enabling a gradually increasing volume of medical services through autonomous AI, rather than relying entirely on healthcare workers. This will undoubtedly be hindered by the dilemma of liability for possible AI errors, yet the demand for such a transformation will remain high.
We foresee a transition to autonomous AI agents unfolding in several key areas:
- Diagnostic Reference Centers with AI-Generated Descriptions: When conducting an examination (e.g., radiology, ECG, etc.), the results would be sent to an autonomous AI agent rather than a physician. If the results fall within normal limits, the AI agent would independently generate and sign off on the examination report in the electronic health record. If pathology is detected, only this particular examination—along with the pre-prepared description—would be referred to a diagnostic physician. It is essential to move towards centralized models that place such physicians in reference centers. Ultimately, this digital transformation could drastically reduce diagnostic costs since the fees for AI agent interpretation should be significantly lower than those for a physician.
- Digital Preventive Care: A significant portion of check-ups and preventive medical examinations currently caters to essentially healthy patients. Data from these cases can be analyzed by autonomous AI agents that can currently perform assessments more accurately and, importantly, more personalized. If the AI agent is confident that the results of the preventive examination show no signs of pathological processes, it can independently produce an evaluation and recommendations without involving a physician. The cost of such analysis could be 2-3 times lower than that conducted by a human physician, while the interpretation time could be significantly reduced.
- Remote Monitoring and Long-term Observational Care: The boom in wearable devices, combined with their capacity for continuously capturing a wealth of health-related data, presents another promising avenue for autonomous AI applications. Continuous analysis of data streams, coupled with insights drawn from electronic health records and personalized patient profiling, allows for ongoing health assessments and predictions of potential complications, all without requiring direct involvement from healthcare workers. This monitoring could emerge as a more affordable and efficient alternative to traditional long-term observation, appealing not only to healthcare organizers but also to patients themselves.
Currently, most existing solutions are effectively niche products in terms of functionality. There are separate AI solutions for analyzing images, some focused solely on radiology, others for risk assessment, compliance with clinical guidelines, drug therapy selection, and so forth.
As LLM products and agent-based AI evolve, the number of solutions offered by developers is only set to increase. The industry will soon face a situation where buyers will no longer be able to purchase and utilize distinct niche products due to prohibitive integration costs and the time required to implement them within existing information systems.
In the nearest future, similarly to previous developments with MIS, a demand for integrating various modalities and standalone AI products into unified platform solutions will emerge.
This centralization trend will evolve on two levels.
- The first is infrastructure integration, as individual AI solutions will require mandatory exchange of both the received analytical data and the results produced. For example, AI agents analyzing images will need to extract data from electronic health records and interpret results accordingly from clinical decision support systems. Simultaneously, sharing findings from image analysis AI agents with clinical decision support systems and management analytics will provide added value to these products. Such secure and high-performance data exchange will only be feasible if all products are housed within a shared data center. If LLMs are utilized, a data center equipped with specialized chips will become essential.
- The second level of centralization will occur in products and their integration with MIS and patient applications. Developers of corresponding IT systems will struggle to ensure integration and data exchange with numerous disparate AI solutions. They will likely opt for centralization, securing integration with a major developer who, in turn, will encompass niche AI products under their umbrella.
As a result, market consolidation will take place. The significant player who invests in integrating existing AI services rather than constantly creating new specialized AI solutions will effectively gain control over the AI market.
Transitioning to Multimodal Solutions
Another important feature of current AI solutions is their monomodal nature. Some companies specialize in image analysis, while others focus on electronic health records, and so forth. However, it is becoming increasingly apparent that further improvements in the accuracy and value of AI products for clients and users will be restricted by this specialization.
It is impossible to infinitely enhance the accuracy of pathological detection in imaging without the AI model having access to patient history, prior diagnoses, and past medical consults—all of which reside outside of DICOM files and exist only in EHRs.
A similar issue applies to AI products for electronic health records; for instance, it is impossible to continuously improve predictive accuracy for cancer progression without knowledge of carcinogens or a patient's lifestyle. These data points are not found in medical documents; they can only be captured from patient applications or other unrelated AI systems, like corporate databases.
Such limitations create a clear strategic demand for the development of comprehensive multimodal products that integrate the analysis of various data sources.
The most promising approach involves combining medical image analysis with data extraction from EHRs, while also incorporating data from wearable devices and patient applications. Looking further ahead, integrating complex sources such as social media, patient shopping behaviors, lifestyle choices, and environmental conditions could enhance these solutions.
Shift in Solutions from Specialized Developers to Major Tech Corporations
The demand for comprehensive centralized and multimodal solutions, alongside the shift to generative AI—particularly industry-specific LLMs and autonomous AI agents—will require substantial financial investments for retraining models and operating them in specialized secure data centers. This, in turn, will necessitate a corresponding increase in electrical power capacities and costly GPUs for both retraining and running generative AI.
Such investments are beyond the realm of small technological startups, which currently supply around 80% of AI products for healthcare.
As a result, we will likely see a gradual transfer of AI solutions from startups to large technology corporations.
In the United States, organizations like OpenAI, Anthropic, Google, and even Apple—who has fallen behind in the AI race—are already developing specialized AI platforms for healthcare, pouring enormous infrastructure and financial resources into these efforts. It appears that the same trend will eventually occur in Russia, albeit with some delay.
Specialized startups are unlikely to win in the AI arms race. Developing inertia within the paradigm of a mono-product approach or occupying a narrow market niche, they will find themselves competing not against similar small developers, but against large tech giants. This dynamic puts them at a strategic disadvantage, eventually leading to their gradual exit from the market.