AI × Healthcare: The Next Trillion-Dollar Industrial Revolution

Executive Summary

Over the past decade, artificial intelligence has achieved its most significant breakthroughs in internet platforms, software, and information processing. The next phase — the one most likely to transform humanity's long-term productive capacity — may unfold in an industry that is simultaneously larger, more complex, more inefficient, and more directly consequential to every human life: healthcare.

Healthcare is home to the world's most complex data, its highest-value knowledge systems, its longest development cycles, and its highest costs of failure. Against this backdrop, global demographics are deteriorating rapidly: populations are aging, the burden of chronic disease and cancer is escalating, and the healthcare workforce shortage is becoming acute. The World Health Organization projects a global shortfall of approximately 11 million healthcare workers by 2030. Simultaneously, the global population aged 60 and above is projected to grow from approximately 1 billion in 2020 to 1.4 billion by 2030, and to 2.1 billion by 2050.

The question for the healthcare industry is therefore not whether demand exists. The real question is whether the traditional healthcare system possesses sufficient human capital, time, and operational efficiency to meet that demand. AI may be emerging as a critical technology to resolve this contradiction — and this time, AI is not simply helping doctors write reports. It is progressively penetrating the entire value chain: data → diagnosis → drug discovery → clinical trials → treatment → gene editing → surgery → outcomes monitoring. As of the end of 2025, the FDA had authorized 1,357 AI/machine-learning medical devices across 693 companies and 17 clinical specialty areas. The cumulative total surpassed 1,000 authorizations in 2024. This is no longer a laboratory concept — it is an industrial force entering real clinical settings.

I. Structural Demand Surge: Why Healthcare's Long-Term Opportunity Is Compelling

Healthcare's greatest long-term opportunity does not originate primarily from AI. It originates from a more fundamental reality: human longevity is increasing, and disease is becoming progressively more complex.

1. Global Population Aging Is Reshaping the Structure of Medical Demand

WHO data shows that the global population aged 60 and above stood at approximately 1 billion in 2020 and is projected to reach 1.4 billion by 2030 and 2.1 billion by 2050. In other words, from 2020 to 2050, the global elderly population will more than double. Aging does not simply generate demand for more hospitals; it generates sustained, escalating demand for long-term management of the following chronic conditions:

• Cancer
• Cardiovascular disease
• Diabetes
• Neurodegenerative diseases
• Renal disease
• Chronic respiratory disease
• Orthopaedic conditions

2. Non-Communicable Diseases Are the World's Largest Healthcare Burden

WHO data shows that non-communicable diseases caused at least 43 million deaths in 2021, representing approximately 75% of all non-pandemic-related global deaths. Of this total, cardiovascular diseases accounted for at least 19 million deaths, cancers approximately 10 million, chronic respiratory diseases approximately 4 million, and diabetes and related kidney diseases over 2 million.

This means the largest future healthcare market is not the treatment of a single disease — it is the continuous understanding of the human body, the early detection of disease, the precise selection of treatment pathways, and the long-term monitoring of therapeutic outcomes. This is precisely where AI has one of its greatest potential advantages: processing vast datasets, identifying complex patterns, and generating predictions.

3. Healthcare Supply Cannot Keep Pace With Demand: Workforce Gap Is the Core Bottleneck

One of the most acute bottlenecks in the global healthcare system is human capital. Physicians, nurses, radiologists, pathologists, drug researchers, and clinical trial personnel all require years of specialist training. WHO currently projects a global healthcare workforce shortage of approximately 11 million by 2030. AI will not produce more doctors in the near term, but it can materially improve the productivity of existing physicians and healthcare systems. The genuine value of AI in medicine is therefore more likely to manifest as enabling one physician to handle the workload that previously required several, rather than replacing physicians outright.

II. AI Enters Healthcare: The Opportunity Is a Complete Value Chain, Not a Single Product

The traditional healthcare value chain has been highly fragmented across its nodes: patient → testing → physician diagnosis → drug discovery → clinical trials → treatment → surgery → long-term monitoring. AI is progressively integrating these segments, forming a closed loop from data to treatment. The most important competitive dynamic in healthcare AI will not necessarily be which company possesses the most powerful model, but which company can command the complete data–experiment–treatment feedback loop.

III. FDA Data Confirms: Healthcare AI Has Entered Real Clinical Systems

One of the most important methods of assessing whether a technology delivers genuine value is to observe whether regulators have begun to approve it. The US FDA's AI medical device database provides a critical observation window. As of the end of 2025, the FDA had authorized 1,357 AI/machine-learning medical devices across 693 companies and 17 clinical specialty areas. The cumulative total surpassed 1,000 authorizations in 2024, with 258 new authorizations granted in the first three quarters of 2025 alone. AI is actively entering real clinical settings in radiology, cardiovascular medicine, neurology, ophthalmology, surgery, pathology, ultrasound, and oncological diagnosis. This is no longer a laboratory concept.

IV. The First Frontier: Medical Imaging and Diagnosis

Medical imaging is fundamentally a large-scale pattern recognition problem. CT, MRI, X-Ray, ultrasound, and pathology slides are all instances of the same process: data → image → pattern → judgment. This is one of deep learning's most natural domains of competence.

A study published in JAMA Network Open found that as of June 2024, the FDA had authorized 950 AI/machine-learning medical devices, of which 723 (76%) belonged to radiology. A separate study published in npj Digital Medicine analysed 1,016 FDA authorizations as of 2024 and found that images remained the predominant AI input type, accounting for 84.4% of cases.

This establishes an important point: the first phase of AI's entry into healthcare is likely not the 'AI physician' — it is AI becoming the physician's second pair of eyes. Stanford's 2025 AI Index summarised a broad body of medical AI research, with some studies finding that advanced AI models can match or exceed physician performance on complex clinical diagnostic tasks. Crucially, a growing body of evidence suggests that AI working in conjunction with physicians typically outperforms either AI alone or physicians alone. This means the more realistic future competitive dynamic is not AI replacing physicians, but physicians who are most proficient with AI gradually displacing those who are not.

The more realistic competitive dynamic is not AI replacing physicians, but physicians who are most proficient with AI gradually displacing those who are not.

V. Beyond Imaging: AI Begins to Process Biology Itself

Medical imaging represents only the first phase. The truly transformative industrial opportunity lies in AI beginning to process biology itself. The human body encompasses DNA, RNA, proteins, cells, metabolism, the immune system, microbiome, pharmaceuticals, environment, and clinical manifestations — interacting with extraordinary complexity. Scientists have historically struggled to process so many variables simultaneously. AI offers this capability for the first time.

AlphaFold: AI Opens the Data Architecture of Life Sciences

In 2024, Google DeepMind's AlphaFold project was awarded the Nobel Prize in Chemistry. This is not an ordinary AI milestone — it represents AI beginning to solve biological problems that traditional computational methods have long been unable to address efficiently. AlphaFold has predicted over 200 million protein structures, with its database used by over 3 million researchers across more than 190 countries.

AlphaFold 3 further attempts to predict interactions among proteins, DNA, RNA, small molecules, ions, and other biomolecules. This means drug development may undergo a fundamental shift: in the past, scientists searched for molecules; in the future, AI will first understand disease mechanisms, then design molecules.

VI. AI Drug Discovery: Transforming Healthcare's Most Expensive Segment

Traditional drug development faces a formidable problem: a new drug typically requires 10–15+ years, development costs can reach $1–2 billion or more, and approximately 90% of drug candidates that enter clinical trials ultimately fail. The primary cause of failure is not the absence of drugs, but the failure to find truly effective ones.

AI seeks to transform this process. The traditional pathway is: disease → target identification → mass molecule screening → experiments → optimisation → animal studies → clinical trials. The AI pathway is: massive biological data → AI identifies disease mechanisms → predicts potential targets → generates candidate molecules → AI predicts properties → automated experiments → clinical trials. The greatest value is not making scientists 20% faster, but potentially eliminating large numbers of candidates that do not warrant entering the laboratory at all.

The First Real Case: Insilico Medicine

Insilico Medicine (03696.HK) is a particularly noteworthy case. Its AI platform contributed to the development of the TNIK inhibitor ISM001-055 for idiopathic pulmonary fibrosis, with the discovery-to-candidate process taking approximately 18 months, compared to a traditional pathway of approximately 4–5 years. The project has completed Phase IIa and reported positive efficacy results.

This is significant. The market's previous question was 'Can AI actually create real drugs?' Now that question is gradually evolving into: 'Can AI consistently create successful drugs?' — these are two fundamentally different questions.

Core AI Drug Discovery Companies

Recursion Pharmaceuticals (RXRX): Core approach is AI + automated laboratories + large-scale cellular data, attempting to build a machine that discovers drugs automatically. Strength is vast experimental data; risk is that clinical success rates remain unguaranteed.

Insilico Medicine (03696.HK): Core is generative AI + drug design + automated experiments, currently one of the very few AI-native pharmaceutical companies listed publicly. The key validation point is whether AI-designed drugs can ultimately become commercialised products.

Schrödinger (SDGR): Takes a physics + molecular simulation + AI approach, closer to the physical mechanisms of drug molecules. Its technology has already advanced multiple candidates into late-stage clinical development, and computational chemistry is gradually transitioning from a tool to drug development infrastructure.

VII. mRNA: One of the Most Important Bridges Connecting AI to Therapy

If AI can understand disease, the next step is to design treatments. mRNA is critically important here. Traditional pharmaceuticals manufacture a fixed molecule, whereas mRNA is closer to sending a manufacturing instruction to the body's cells. This means that when a virus presents a new variant, the theoretical flow — AI analyses virus → designs new mRNA sequence → rapidly manufactures vaccine → enters clinical use — can be dramatically compressed. This is why COVID-19 made the world recognise for the first time that mRNA is a platform technology, not merely a vaccine.

Moderna and BioNTech: From Vaccines to Cancer

The truly compelling long-term story for Moderna (MRNA) and BioNTech (BNTX) is no longer COVID — it is cancer. The future pathway may be: tumour sequencing → AI identifies tumour-specific antigens → designs personalised mRNA → trains the immune system → attacks cancer cells. This means AI + gene sequencing + mRNA + immunotherapy may form an entirely new cancer treatment system.

VIII. Gene Editing: From Treating Symptoms to Modifying Disease at Its Source

If mRNA sends instructions to the body, gene editing directly modifies DNA. CRISPR Therapeutics (CRSP) has demonstrated that CRISPR gene editing can reach the commercialisation stage. Its Casgevy became one of the world's first approved CRISPR gene editing therapies, for sickle cell disease and transfusion-dependent beta-thalassemia. This means gene editing has moved from scientific experiment to real medicine.

Beam Therapeutics: Next-Generation Base Editing

Beam Therapeutics (BEAM) represents another direction: Base Editing. Traditional CRISPR acts like scissors, while Base Editing is more like changing a single letter in DNA, with higher precision.

The 2026 BEAM-302 clinical data is particularly noteworthy. Company reports show that at the 60mg dose, mean steady-state total AAT reached 16.1 μM, with all patients consistently above the 11 μM protective threshold, while mutant Z-AAT decreased by 84%, and normal M-AAT accounted for 94% of total AAT. The company plans to initiate a global pivotal study in the second half of 2026. If future evidence proves that a single in vivo Base Editing can permanently correct genetic diseases, the gene editing market may enter an entirely new phase.

IX. Cancer: AI Healthcare's Most Important Battlefield

Why is cancer particularly important? Because cancer is not one disease — it is a collection of hundreds of different diseases. Different patients have different genes, different mutations, different tumour microenvironments, and different immune responses. This means precision medicine + AI is a natural fit for cancer.

The future cancer value chain may form: blood/tissue → gene sequencing → AI identifies mutations → predicts cancer risk → selects treatment → personalised drugs → treatment → ctDNA testing → determines whether recurrence occurs. Along this chain, a group of critically important companies will emerge.

Natera: The 'Molecular GPS' After Cancer Treatment

Natera's (NTRA) core product Signatera focuses on MRD — minimal residual disease. After cancer surgery, the absence of visible cancer cells does not mean cancer cells have truly disappeared. Through ctDNA testing, it is possible to detect extremely low levels of tumour DNA in the blood. This may change the cancer treatment paradigm: in the past, periodic CT scans; in the future, continuous molecular-level monitoring.

Guardant Health: From Tissue Biopsy to Liquid Biopsy

Guardant Health's (GH) core is liquid biopsy. The traditional approach requires cutting tissue; the future requires only a blood draw, followed by analysis of DNA, RNA, epigenetic information, and more. In 2026, the FDA approved the new version of Guardant360 Liquid CDx, significantly expanding detection scope and further strengthening the role of liquid biopsy in precision oncology. Cancer diagnosis is moving from 'detecting tumours' to 'understanding tumours'.

Tempus AI: The Core Platform for Healthcare AI Data

Tempus AI (NASDAQ: TEM) has a fundamentally different logic from companies that simply perform testing. It attempts to build a comprehensive medical data platform integrating clinical data + genomic data + pathology data + AI. Future value may come from the closed loop: data → AI → clinical decisions → drug development → new data. This is one of the most important business models in healthcare AI: data is not a one-time asset, but infrastructure that can continuously train AI and generate new value.

X. Robotic Surgery: AI Ultimately Enters the Operating Theatre

If data is AI's brain and drugs are AI's therapeutic tools, then robots are AI's hands. Intuitive Surgical (NASDAQ: ISRG), with its Da Vinci surgical robot, has demonstrated the most mature commercial case in robotic surgery. Its true strength lies not in selling robots, but in the complete commercial closed loop formed by robots + physician training + surgical data + consumables + hospital ecosystem.

In the future, AI may further enter surgical path planning, tissue identification, risk prediction, surgical action analysis, automated assistance, and physician training. Robotic surgery may gradually evolve from 'physician operates robot' to 'physician supervises AI-assisted robot'.

XI. The Genuine Long-Term Opportunity: A Five-Layer Value Chain Framework

Many investors, when looking at healthcare AI, first search for the 'AI physician', but this may not be the largest opportunity. The true long-term value is more likely to come from systematic positioning across the entire value chain: Layer 1 Data (TEM/NTRA/GH) → Layer 2 AI Models (medical large models, biological models) → Layer 3 Drug Discovery (RXRX/03696/SDGR) → Layer 4 Therapeutic Technologies (BNTX/MRNA/CRSP/BEAM) → Layer 5 Robotics (ISRG) → Layer 6 AI-Driven Complete Healthcare Closed Loop.

Why Healthcare AI May Become the Next Giant Industry of the AI Era

Healthcare simultaneously possesses four structural conditions that drive AI penetration:

• ① Massive economic scale: global healthcare spending is itself an enormous industry
• ② Extremely high information density: DNA, RNA, proteins, imaging, medical records, drugs, clinical trials — all are data
• ③ Extremely high labour costs: physicians, researchers, nurses, clinical trial personnel all require substantial human investment
• ④ Extremely high cost of failure: a single drug failure can waste hundreds of millions of dollars and years of time

Therefore, even if AI improves efficiency by just 10%, the economic value could be enormous.

XII. Core Risks: Healthcare AI Is Not a Highway Without Barriers

The greatest risk in healthcare AI is not that AI is insufficiently intelligent, but that healthcare is not a pure information industry. Data quality, clinical validation, clinical trial failure rates, and regulatory uncertainty constitute the four most important barriers in this sector.

Risk One: Data Quality. Healthcare data is highly fragmented — different hospitals use different formats, different countries have different privacy regulations, and different patients present diseases differently. If AI model training data contains bias, it may produce erroneous results.

Risk Two: Insufficient Clinical Validation. Accuracy in the laboratory does not equal effectiveness in real hospitals. Although the number of FDA-approved AI medical devices is increasing rapidly, research indicates that real-world validation remains insufficient. A JAMA Network Open study found that among 950 FDA-authorised AI/ML devices as of 2024, 723 belonged to radiology, and among radiology devices with complete submission data, only 29% included clinical testing, and only 5% conducted prospective testing. 'FDA approval' does not equal 'AI has been proven effective in all real-world medical scenarios'.

Risk Three: AI Cannot Eliminate Clinical Trial Failure. AI can identify targets, design molecules, screen patients, and optimise trials, but the human body remains the ultimate test. AI cannot guarantee that a drug will be effective once it enters the human body. A 2026 Tufts Center for the Study of Drug Development study shows that AI has begun helping cancer clinical trials improve efficiency, with analyses estimating AI can shorten late-stage trials by approximately 10 weeks and reduce direct operating costs by up to approximately $5.6 million, but the study also notes that AI cannot solve all clinical trial problems — patient recruitment, informed consent, drug supply, and other issues persist.

Risk Four: Regulatory Uncertainty. The regulatory framework for healthcare AI is itself evolving rapidly. The FDA has issued guidance on AI in drug development and continues to refine the regulatory framework for AI in drug and biological product development in 2026. Meanwhile, the FDA is exploring how to evaluate generative AI medical devices, as such systems may produce variable, non-fixed outputs, fundamentally different from traditional medical devices.

Therefore, the companies that will truly possess moats in the future are likely not those with the most flashy AI technology, but those that combine AI + clinical validation + FDA experience + healthcare data + commercialisation capability.

XIII. Competitive Evolution: From Model Competition to Data + Closed Loop Competition

Competition in healthcare AI will evolve through four stages: Stage One — whose model is smarter; Stage Two — whose data is more abundant; Stage Three — who possesses real clinical data; Stage Four — who can translate AI results into drugs, treatments, and revenue.

Ultimately, the moat in healthcare AI will not be just an algorithm, but the complete closed loop of data, models, experiments, clinical validation, regulation, and commercialisation.

XIV. Five Main Investment Themes for the Next Decade

Complete Company Map

XV. Conclusion: The Self-Reinforcing Cycle of the AI Healthcare Value Chain

The question of whether AI will enter healthcare has essentially been answered. FDA-authorised AI medical devices have reached 1,357; generative AI adoption in healthcare institutions is transitioning from pilot to scale. McKinsey's 2026 survey shows that approximately half of US healthcare organisations have implemented generative AI, with over 80% having deployed at least one application to end users.

The real question has become: to what degree will AI restructure the healthcare industry's value chain?

2026–2028, AI-Assisted Healthcare Phase: Physician efficiency gains, imaging diagnosis, medical records, clinical trials, drug screening.

2028–2032, AI Participates in Medical Decisions Phase: AI begins genuinely participating in drug design, patient selection, personalised therapy, cancer monitoring, surgical assistance.

2030s, AI-Driven Healthcare Phase: AI is no longer just a tool, but becomes the infrastructure connecting data, biology, drugs, therapy, and surgery.

Ultimately, the healthcare industry may shift from the past 'discover disease → treat disease' to 'predict disease → prevent disease → precision treat → continuously monitor → proactively intervene'. The most important change is not simply reducing healthcare costs — more importantly, AI may for the first time enable healthcare to move from large-scale standardised treatment to truly large-scale personalised medicine.

AI healthcare may ultimately form a self-reinforcing technology cycle: more patients → more data → stronger AI → more accurate diagnosis and drug discovery → more high-quality data generated, forming a closed loop of data → AI → biological discovery → treatment → new data. This is why, over the next decade, AI × Healthcare is likely to become one of the most worthwhile technology sectors to track long-term, after the internet, cloud computing, and AI infrastructure.

AI healthcare may ultimately form a self-reinforcing technology cycle: more patients → more data → stronger AI → more accurate diagnosis and drug discovery → more high-quality data.