By Efstathios Iliopoulos and Olivia Zumbach
In memory of Fransoui-Franzeska Giannakopoulou
The “Bundesrepublik” recently announced that it will invest €1.6 billion in AI as part of its AI Action Plan, with the aim of strengthening Germany’s position as a global AI leader¹. Germany also recently joined the European Union’s InvestAI initiative, which aims to mobilize €200 billion in public and private investment for artificial intelligence across Europe².
Undoubtedly, AI development has received unprecedented investment over the last decade from private and public sectors. However, there is no single verified global total for all AI investment from 2015–2025 because countries and companies do not use a common accounting system for AI spending. The most reliable figures come from sources that measure private AI investment by country and national AI-related investment estimates.
The strongest available evidence shows that AI investment has reached hundreds of billions of dollars annually. According to Stanford University’s AI Index Report 2025, global private AI investment reached $252.3 billion in 2024. The United States was the largest investor, with $109.1 billion, followed by China with $9.3 billion and the United Kingdom with $4.5 billion³.
The OECD provides a broader estimate for the European Union, measuring AI investment across both private and public sectors. In 2023, total AI investment in the EU27 was estimated at €257 billion, including €188 billion from the private sector and €69 billion from the public sector. These figures include AI-related investments in areas such as software, hardware, data, and equipment, rather than only scientific research⁴.
The concentration of AI investment is highly uneven. The United States currently dominates private AI investment because of its large technology sector and venture capital ecosystem. China is the second-largest national investor, while countries such as the United Kingdom, Canada, France, Germany, Israel, India, Japan, South Korea, and the United Arab Emirates represent smaller but significant AI investment centres⁵.
In summary, what occurs from available data is that AI development has attracted hundreds of billions of dollars of investment annually, with private-sector investment accounting for the largest share, tough a precise global decade total remains unverifiable for the reasons outlined above.
In particular, investment in AI for medicine (overall) has grown rapidly over the past decade as governments, universities, healthcare organizations, and private companies increasingly recognize AI’s potential to improve diagnosis, treatment, drug discovery, and healthcare delivery.
The U.S. National Institutes of Health (NIH) is the largest public funder of biomedical research worldwide. In 2017, the NIH invested approximately US$264 million in machine learning-related clinical research (about 2% of its clinical research budget). By 2023, NIH funding for AI and machine learning research had increased to approximately US$2.3 billion⁶, representing nearly 4.7% of the agency’s total budget, which signifies an impressive inflation-adjusted increase since 2019⁷.
In Europe, AI research in medicine is funded through large collaborative research programmes, particularly the European Union’s Framework Programmes and Horizon Europe. The scope is, rather than maintaining a dedicated “medical AI” budget, the EU supports AI projects across many areas⁸. Overall, these investments amount to hundreds of millions to over €1 billion annually distributed across healthcare-related AI initiatives⁹.
The private sector also contributes significantly. Major pharmaceutical companies, such as Pfizer, Roche and Novartis including those focused on drug discovery and precision medicine, as well as general technology companies such as Google, Microsoft, NVIDIA, and others are supposedly¹⁰ investing substantial amounts in AI-enabled biomedical research every year. Because companies rarely report AI-specific R&D expenditures separately, exact totals are difficult to determine.
Considering both public and private investment, it could be estimated that global funding for AI research in medicine likely exceeds US$10–20 billion annually.
Given the above figures, a rough scenario estimate would be that healthcare, and medical applications account for approximately 5–10% of global AI investment. Applying this range to Stanford’s estimate of $252.3 billion in private AI investment in 2024 suggests that roughly $13–25 billion may have been directed toward medical AI in that year. This estimate includes AI applications in drug discovery, medical imaging, diagnostics, clinical decision support, and biomedical research, but excludes many healthcare IT investments that are not explicitly classified as AI.
The above conclusion raises three difficult questions: 1) Did this investment pay out? 2) Is the allocated funding enough? 3) Are the funds allocated to the most effective causes?
The short answer to the first question is: partly, but not yet at the scale expected from the investment itself and the general public’s feeling.
AI investment in medicine has produced several measurable successes: In Drug discovery, AI has accelerated protein structure prediction and biological modelling, which in turn supports downstream drug candidate identification¹¹.
In the field of medical imaging, AI systems have achieved expert-level performance in some narrow diagnostic tasks, including radiology, ophthalmology, and pathology. However, most of these performance results come from retrospective studies on curated, often single-institution datasets, rather than prospective validation in routine clinical workflows. Performance frequently drops when models are tested on new populations or imaging equipment (‘domain shift’), which is a major reason why relatively few AI diagnostic tools have been widely adopted in clinical practice.
Furthermore, in the sector of Clinical workflows, AI tools are already reducing administrative workload and enhance the possibilities of better data inputs.
However, financial returns are harder to demonstrate:
- Many AI medical companies have raised large amounts of capital but have not yet produced profitable products.
- Regulatory approval, clinical validation and integration into hospitals remain major barriers.
- A large proportion of AI research prototypes never become routine clinical tools.
A 2023 review in Nature Medicine noted that although AI performance has improved substantially, translation into routine healthcare has been slower because of issues such as data quality, workflow integration, validation, and regulation¹².
The case of cancer.¹³
Although significant investments have been made in healthcare, AI has so far made progress in cancer research primarily in specific areas, including improved image analysis and the identification of biological patterns. However, these achievements have not yet translated into a major change in cancer treatment outcomes or a broadly available AI-derived cancer therapy. A measurable, attributable reduction in cancer mortality has also not yet been demonstrated, though this is partly expected: mortality is a multifactorial outcome that typically requires years of follow-up to detect, so its absence, so far, does not necessarily reflect a failure specific to AI¹⁴. Instead of that, our societies keep counting considerable losses (among them many young people) to this complex of diseases.
The gap between expectations and clinical reality is particularly visible in oncology. Cancer remains one of the most complex diseases because of tumour heterogeneity, evolutionary adaptation, and individual differences between patients. Although AI can analyse large datasets and identify promising targets, translating these discoveries into safe and effective treatments still requires years of laboratory validation, clinical trials, regulatory approval, and real-world implementation. The U.S. National Cancer Institute (NCI) highlights that AI applications in cancer are promising but emphasizes the continuing need for randomized clinical trials, validation, reproducibility standards, and integration into clinical workflows before widespread adoption¹⁵.
On the other side, someone could argue that the funding in medical AI is not enough. That claim depends on the goal.
If the goal is just scientific progress, funding is already very large and has enabled major breakthroughs.
If the goal is transforming healthcare, the funding may still be insufficient because successful medical AI requires much more than just researching algorithms:
- high-quality clinical datasets and secure computing infrastructure
- translation of algorithms into effective, clinically validated treatments
- clinical trials and regulatory approval
- real-world implementation into hospitals.
For comparison:
- Global healthcare spending exceeds $9 trillion annually (World Bank/OECD estimates).
- AI investment represents only a small fraction of total healthcare expenditure.
Therefore, even if medical AI receives tens of billions annually, it remains a small investment relative to the size of the healthcare system it aims to improve. Given the potential social value of AI-driven improvements in prevention, diagnostics, and treatment, a stronger allocation – potentially double the current 5-10% share – could be justified.
For the final question, if funding is allocated to the most effective causes:
This question does not have a single answer: the same investment pattern can look effective or misallocated depending on which criterion is applied.
By application type, most healthcare AI funding is going toward lower-risk, faster-return applications rather than frontier clinical research. AI adoption follows the path of least resistance: fields like radiology are among the earliest adopters precisely because they already generate large volumes of standardized, labeled digital data (images) and involve well-defined pattern-recognition tasks – conditions that suit current AI architectures particularly well. This technical “fit” reinforces the funding pattern: according to a 2024 SVB analysis of US venture funding, administrative AI accounts for 60% of total AI investment in US healthcare, and healthcare AI companies focused on patient diagnostics account for 52% of total US AI investment in clinical solutions.¹⁶ A 2026 npj Digital Medicine analysis of 3,807 AI health startups similarly found that, cumulatively since 2010, imaging and diagnostics attracted $11.87 billion and drug discovery $18.5 billion in funding, while domains like mental health and rehabilitation received comparatively little despite substantial unmet need – a pattern the study attributes to data and scalability limitations rather than a lack of clinical need¹⁷. These figures are consistent with the broader pattern reflected in the $13–25 billion estimated for global medical AI investment in 2024 alone (see above): capital concentrates in domains that are not necessarily the most clinically urgent, but the easiest to build AI for.
By disease burden versus commercial potential, funding tends to track market size rather than global need. Neglected tropical diseases affect more than one billion people worldwide, yet traditional and AI-driven drug discovery alike have largely overlooked them due to limited commercial profitability¹⁸. Oncology, by contrast, attracts outsized capital and political attention – illustrated by the UK’s £19 million PharosAI cancer initiative announced in 2025, and by the $500 billion Stargate Project, a broader US AI-infrastructure initiative that names cancer research as one strategic priority¹⁹ – despite oncology already being one of the best-funded disease areas globally.
By risk profile, investors are systematically favoring incremental tools over higher-risk research. Providers and payers are targeting administrative AI first since it carries less risk and leads to clear efficiency gains. This “flight to quality” was explicit in 2024 SVB data, where investors disclosed that reduced risk is taking priority in their late-stage spending. In the European context, this risk calculus is sharpened by regulation. Clinical AI applications that process patient data typically fall under the GDPR and, increasingly, under the EU AI Act’s “high-risk” category, triggering extensive conformity assessment, documentation, and post-market monitoring obligations.
In Germany, physician confidentiality (ärztliche Schweigepflicht, §203 StGB) further restricts how patient data can be shared with or processed by third-party AI systems, including cloud-based tools. Administrative AI applications, which typically handle less sensitive data, face substantially lower regulatory barriers – one more reason capital and deployment concentrate there rather than in direct clinical decision-making.
Geographically, medical AI investment mirrors the broader AI investment pattern described earlier in this essay: highly concentrated in a few wealthy countries. The United States remains the epicentre of digital health funding with around 70% of global venture funding in 2024, consistent with its dominant $109.1 billion share of overall global private AI investment noted above²⁰, while AI health startups more broadly remain concentrated in high-income countries, leaving regions with the greatest unmet healthcare needs comparatively underfunded.
Taken together, these four lenses point to the same underlying conclusion: medical AI funding is allocated efficiently by market logic – toward applications, diseases, and regions offering the fastest, lowest-risk returns – but not necessarily toward the causes with the greatest potential global health impact. Whether this constitutes genuine misallocation depends on whether AI investment is viewed as a market-driven innovation process or as a resource that should be deliberately steered toward maximizing global health outcomes.
Ultimately, this raises a question larger than investment allocation alone: healthcare systems worldwide are under mounting structural pressure (aging populations, workforce shortages, and rising costs) and this pressure is unlikely to be resolved without AI in the loop. If that premise holds, then AI cannot remain concentrated in the most commercially lucrative corners of medicine. It needs to be integrated across the full breadth of patient care, not only where the data is cleaner and the returns faster, but also where the clinical and societal need is greatest. Realizing that shift will require the medical field to stop treating AI as a competing force and start treating it as a partner in sustaining and improving healthcare delivery.
References
¹https://www.bundeswirtschaftsministerium.de/Redaktion/DE/Pressemitteilungen/2026/04/20260428-abschlusssitzung-der-expertenkommission-wettbewerb-und-kuenstliche-intelligenz.
²https://digital-strategy.ec.europa.eu/en/news/eu-launches-investai-initiative-mobilise-eu200-billion-investment-artificial-intelligence.
³ Stanford University Human-Centered Artificial Intelligence — AI Index Report 2025, Economy Chapter.
⁴ https://oecd.ai/en/ai-investment
⁵ https://hai.stanford.edu/ai-index/2025-ai-index-report?sf225800102
⁶ Nananukul, N., & Kejriwal, M. (2026). An Analysis of Artificial Intelligence Adoption in NIH-Funded Research. arXiv. This study reports that NIH AI/ML funding reached approximately US$2.3 billion in 2023 and analyzes trends in AI-related biomedical research.
⁷ https://report.nih.gov/award/index.cfm
⁸ Such as medical imaging, personalized medicine, clinical decision support, and digital health
⁹ Fajardo-Ortiz, D., Thijs, B., Glänzel, W., & Sipido, K. (2023). Evolution of funding for collaborative health research towards higher-level patient-oriented research. arXiv. This paper compares European Union and NIH health research funding and discusses the evolution of collaborative health research investment
¹⁰ There is not a definite official number.
¹¹ For example, DeepMind developed AlphaFold, which predicted the structure of virtually all known proteins and significantly changed structural biology research. It is worth noting that AlphaFold itself does not identify or design drugs – it predicts protein structures, which researchers then use as a starting point for structure-based drug design. The step from structure prediction to an actual approved drug candidate still requires extensive additional research.
¹² Nature Medicine – Artificial intelligence in healthcare: past, present and future.
¹³ Cancer is a group of related diseases characterized by the uncontrolled growth and spread of abnormal cells, caused by genetic changes affecting genes that regulate cell growth, division, and DNA repair. These abnormal cells can invade surrounding tissue and spread to other parts of the body (metastasis). Source: National Cancer Institute (NCI), “What Is Cancer?” — cancer.gov/about-cancer/understanding/what-is-cancer
¹⁴ National Cancer Institute (NCI), “Artificial Intelligence (AI) and Cancer” — discusses AI opportunities in cancer research while explicitly noting the need for clinical validation, randomized trials, and implementation research before AI can improve cancer care at scale
¹⁵https://www.cancer.gov/research/infrastructure/artificial-intelligence?cid=soc_fb_en_enterprise_nca50
¹⁶ Silicon Valley Bank (SVB), “Healthcare Investments and Exits” report, 2024\u20132025. Figures refer to US venture capital deal data (proprietary SVB data and PitchBook), not global investment; not a peer-reviewed source.
¹⁷ “Mapping AI startup investment and innovation in healthcare using a five-tier AI systems complexity framework,” npj Digital Medicine (2026). Figures are cumulative totals for 2010\u20132024, not annual 2024 figures.
¹⁸ Nishan, M.D.N.H. (2025). “AI-powered drug discovery for neglected diseases: accelerating public health solutions in the developing world.” Journal of Global Health, 15, 03002.
¹⁹ UK PharosAI initiative (\u00a319 million, 2025) and the US Stargate Project ($500 billion AI-infrastructure initiative naming cancer research as one of several priorities); reported in The Lancet Oncology (2025), “Cancer drug discovery at warp speed: can AI deliver?”
²⁰ SVB / industry reporting on 2024 global digital health venture funding share by region.
