Davos 2026 was wall-to-wall AI talk. Whether you were listening to heads of state or CEOs, the conversation kept circling back to how artificial intelligence is reshaping every industry. Healthcare came up a lot, which makes sense given the sector’s complexity and the stakes involved. Demis Hassabis discussed AlphaFold. Novartis pitched precision medicine. But amid all the optimism about AI-designed drugs and personalized treatments, there was a blind spot: most of the world won’t benefit from any of this.
The Uncomfortable Reality
Google’s James Manyika highlighted one bright spot at Davos. In India, the company performed retinal scans for diabetic retinopathy on over 600,000 patients using AI diagnostics. It’s working with partners to scale this to millions. That’s real impact in a region where healthcare access is limited.
But this example is the exception, not the rule. The broader picture reveals a growing divide between AI healthcare in wealthy countries and the Global South. While Davos attendees discussed trillion-dollar infrastructure investments and AlphaFold-designed drugs, over 5 billion people remain largely invisible to the AI health systems being built.
Training Data From Another World
The core problem is simple: AI healthcare systems are trained almost entirely on data from high-income countries. Over 80% of genetics studies include only people of European descent, representing less than 20% of the global population. When you build diagnostic algorithms on datasets that exclude most of humanity, those algorithms fail when applied to excluded populations.
This isn’t theoretical.
A cancer detection algorithm trained primarily on lighter skin tones will miss tumors on darker skin. An AI diagnostic tool built on Western disease patterns won’t recognize how conditions present differently in other populations. Cardiac risk assessment models calibrated for one genetic background produce inaccurate predictions for others.
The consequence is straightforward: the populations who could benefit most from low-cost, scalable AI diagnostics are the least represented in the systems being developed. Rwanda’s Babyl Health program shows what’s possible when AI tools are actually designed for local contexts. Using machine learning to analyze patient symptoms via mobile apps, the program has facilitated over 2 million consultations, reducing diagnostic turnaround times by 40% in rural areas. But Babyl remains an outlier.
Infrastructure Reality
The 85 trillion dollar AI infrastructure investment Huang described at Davos won’t reach most of the world. His five-layer framework requires massive capital: energy generation, chip manufacturing, data centers, AI models, and applications. Countries struggling with basic healthcare delivery can’t fund this buildout.
Even where infrastructure exists, it’s often inadequate. Many low-resource healthcare systems still rely on paper-based records. Data is fragmented, decentralized, and incomplete. Digital health initiatives face barriers from language diversity to low health literacy affecting data quality and completeness.
India’s Ayushman Bharat Digital Mission, connecting nearly 750 million people to digital health records, was discussed at Davos as a success story. It is. But it also highlights how much ground needs to be covered. Most of Africa, large parts of Asia and Latin America lack comparable systems.
Google’s 15 billion dollar investment in India, spanning infrastructure, data centers, and research, demonstrates what scaled commitment looks like. But similar investments aren’t happening across most of the Global South. Post-pandemic, healthcare development financing has declined even as defense spending rises globally.
The Precision Medicine Paradox
Novartis’s session at Davos focused on precision medicine enabled by AI and advanced therapeutics. The pitch was compelling: fewer than 10% of known diseases have meaningful treatments, but AI can change that by delivering unprecedented insights into human biology and enabling interventions matched to individual patients.
There’s a problem. Precision medicine requires genetic data, diagnostic infrastructure, and treatment access. AlphaFold can identify novel drug targets for previously untreatable diseases. Isomorphic Labs can design molecules computationally. But if those drugs cost tens of thousands of dollars and require genetic testing unavailable in most of the world, precision medicine becomes precision inequality.
The pharmacogenomics gap is particularly stark. Pharmacogenomics matches drugs to patients based on genetic profiles, predicting which medications will work and which will cause adverse reactions. It’s central to the precision medicine vision articulated at Davos. But implementing pharmacogenomics requires genetic sequencing capacity, bioinformatics expertise, electronic health record integration, and clinical decision support systems.
These capabilities exist in high-income countries. They’re largely absent elsewhere. When Novartis and others talk about tailoring treatments to individual biology, they’re talking about a fraction of the global population with access to that level of care.
What Actually Needs to Happen
Addressing this divide requires more than acknowledging it exists. Several concrete steps could narrow the gap:
Representative Training Data: AI health systems need to be built on globally diverse datasets. This means investing in data collection infrastructure in underrepresented regions, not just analyzing existing data from wealthy countries. Community engagement is critical to ensure data reflects local contexts and respects privacy concerns.
Studies in Guatemala developing AI-enabled ultrasound tools for Indigenous midwives or in Bangladesh creating depression screening for non-specialized providers show what locally-grounded development looks like. These projects involve affected communities, use representative datasets, and operate in local languages and cultural contexts.
Local Infrastructure and Expertise: The Davos infrastructure conversation focused on chips and data centers. Equally important is building diagnostic capacity, training local AI researchers and healthcare workers, and creating regulatory frameworks appropriate to local contexts rather than importing Western models.
Microsoft Research’s Peter Lee discussed the brain economy initiative at Davos, examining how human and artificial intelligence can jointly deliver outcomes. That framework needs application in resource-limited settings, where AI could augment scarce medical expertise rather than replacing it.
Equitable Partnerships: Most AI healthcare development happens in high-income countries and gets deployed elsewhere. True progress requires reversing that flow. Research and development should happen in the Global South for the Global South’s benefit, not just technology transfer of tools built elsewhere.
Isomorphic Labs has 3 billion dollars in partnerships with major pharmaceutical companies. Similar investment in AI drug discovery focused on diseases primarily affecting low-income populations would demonstrate commitment to equity rather than just market opportunity.
Pharmacogenomics for Global Populations: If precision medicine is the future, pharmacogenomics needs global implementation. This requires several things: building genetic databases representing diverse populations, developing low-cost sequencing and testing infrastructure, training healthcare workers to interpret and apply genetic information, and integrating pharmacogenomic data into clinical workflows in resource-limited settings.
Current pharmacogenomic research focuses overwhelmingly on populations of European descent. Drug-gene interaction data for other populations is limited. Clinical trials rarely include genetic stratification for diverse ethnic groups. Without addressing this, precision medicine remains precise for only a small subset of humanity.
Guatemala’s NatallA project provides a model. By developing AI tools that help midwives perform diagnostic ultrasounds without specialized personnel, it addresses a real healthcare gap in a culturally appropriate way. Similar approaches could bring pharmacogenomics to settings where genetic testing is currently unavailable.
Regulation and Governance: Many countries lack AI-specific health regulations, relying on broader medical device frameworks that don’t address AI challenges around data privacy, algorithmic bias, and accountability. Without appropriate governance, AI deployment risks exacerbating inequalities.
The Davos Alzheimer’s Collaborative discussed brain health and prevention-oriented approaches. Their framework emphasized upstream investments and measurable economic returns. But implementing this globally requires regulatory systems that ensure AI tools are safe, effective, and equitable across diverse populations.
Addressing Algorithmic Bias: Even well-intentioned AI systems can perpetuate bias if training data reflects historical inequities. Research shows algorithmic bias produces 17% lower diagnostic accuracy for minority patients in some studies. The digital divide excludes 29% of rural adults from AI-enhanced tools.
Mitigating this requires diverse development teams who can identify potential disparities, robust testing across populations, transparency in how algorithms make decisions, and mechanisms for identifying and correcting bias in deployed systems.
The Business Case No One Made at Davos
Here’s what wasn’t discussed at Davos: there’s a business case for equitable AI healthcare development, not just a moral one. The Global South represents the majority of humanity and a growing share of global GDP. Companies that develop AI health solutions working effectively across diverse populations have access to larger markets.
More importantly, diseases don’t respect borders. Tuberculosis, which goes undiagnosed in nearly 40% of cases globally according to Manyika, affects everyone if left unaddressed. Pandemic preparedness requires healthcare infrastructure everywhere, not just in wealthy countries. AI tools that work only for well-resourced populations leave critical gaps in global health security.
The 85 trillion dollar infrastructure investment Huang described could include meaningful allocation to Global South healthcare AI if framed as necessary infrastructure for global economic growth and stability rather than charitable development assistance.
What This Means for Pharmacogenomics
Pharmacogenomics sits at the intersection of AI-driven drug discovery and personalized medicine deployment. Companies developing AI-designed drugs need to think beyond just molecular design to implementation realities.
If Isomorphic Labs creates drugs targeting previously undruggable proteins using AlphaFold, that’s a breakthrough. But if those drugs only work effectively in patients with certain genetic profiles, and genetic testing is unavailable to most potential patients, the breakthrough has limited impact.
Integrating pharmacogenomics from the beginning means several things: including diverse genetic data in AI drug discovery models, designing companion diagnostics that work in low-resource settings, conducting clinical trials with genetic stratification across populations, and developing low-cost pharmacogenomic testing methods.
The precision medicine vision articulated at Davos won’t achieve its potential if precision is available only to the wealthy. Pharmacogenomics can bridge the gap between computational drug discovery and equitable treatment access, but only if developed with global implementation in mind.
Beyond Davos Optimism
Davos 2026 presented AI as transformative for healthcare. It can be. But transformation that reaches only a fraction of humanity while leaving billions behind isn’t transformation, it’s technological apartheid.
The sessions on brain health, cardiovascular disease prevention, and breakthrough innovation all emphasized data-driven, AI-enabled approaches. None adequately addressed how these approaches scale to populations currently invisible to the AI systems being built.
India’s digital health success and Google’s AI diagnostics work show possibilities. Rwanda’s Babyl Health, Guatemala’s NatallA project, and Bangladesh’s depression screening tools demonstrate what locally-grounded development achieves. But these remain exceptions.
The 5 billion people at risk of exclusion from AI healthcare advances aren’t a side concern to be addressed after the technology matures. They’re the majority of humanity. If AI healthcare development continues on its current trajectory, optimized for wealthy populations with data from wealthy populations, we’ll build sophisticated systems that work brilliantly for 15% of the world while failing everyone else.
Pharmacogenomics offers a test case. If precision medicine means matching drugs to individual genetic profiles, but genetic profiling is available only in high-income countries, precision medicine becomes a luxury good. The alternative is building pharmacogenomic capacity globally from the start, using AI to analyze diverse genetic data, developing low-cost testing methods, and integrating genetic information into clinical care across all resource settings.
That requires different priorities than those evident at Davos 2026. Infrastructure investment needs to include diagnostic capacity and data systems in low-resource settings. Drug discovery needs to focus on diseases affecting global populations, not just profitable markets. Research and development needs to happen where the health challenges are most acute, not just where the funding is most available.
The choice is whether AI becomes a tool for global health equity or global health inequality. Davos 2026 leaned heavily toward the latter without acknowledging it. Changing that trajectory requires recognizing that healthcare AI development excluding most of humanity isn’t just ethically problematic, it’s strategically shortsighted and practically incomplete.
