A data-first counterpoint to AI’s medical moonshots
Fidji Simo’s assertion that AI could “cure all diseases” lands in a moment when the technology sector is being pressed—by regulators, investors, and even AI leaders themselves—to separate aspiration from deliverable science. Against the backdrop of Anthropic CEO Dario Amodei’s warnings about overpromising, Simo’s message is notable less for its ambition than for its condition: medical AI will only reach its potential if the industry builds the biological data and infrastructure capable of supporting it.
This framing subtly shifts the debate away from model theatrics and toward the unglamorous substrate of progress: high-quality, longitudinal, and clinically grounded datasets. In other words, the limiting factor is not whether transformers, multimodal models, or graph neural networks can be made larger or faster; it is whether they can be fed the kind of structured, diverse, and validated information that medicine demands.
Simo’s move from OpenAI to co-found ChronicleBio underscores a broader industry pattern: some of the most credible next steps for AI in healthcare may come not from consumer-facing applications, but from platform-building—the painstaking work of curating disease-specific datasets, harmonizing clinical records, and making biology computable at scale.
—
Why chronic and rare diseases expose AI’s biggest bottleneck: missing biology
Oncology has become the poster child for data-rich medicine. Decades of investment have produced deep reservoirs of genomic sequencing, tumor pathology, clinical trial endpoints, and standardized protocols. That density of information makes cancer a comparatively fertile domain for machine learning: patterns can be learned, validated, and iterated upon with a degree of statistical confidence.
Chronic and rare diseases, by contrast, often live in the shadows of the data economy. They are frequently characterized by:
- Fragmented care journeys spread across multiple providers and years
- Inconsistent diagnostic criteria and shifting clinical definitions
- Sparse labeled outcomes, especially for quality-of-life and functional measures
- Underrepresentation across demographics, geographies, and socioeconomic groups
- Limited longitudinal tracking, despite conditions unfolding over long time horizons
This is where Simo’s thesis becomes strategically sharp. If AI is to meaningfully reduce the “diagnostic odyssey,” improve treatment selection, or enable earlier intervention for chronic conditions, it needs datasets that capture the full arc of disease—from symptoms and biomarkers to lifestyle, environment, and response to therapy.
The implication is that the next wave of healthcare AI will be judged less by benchmark scores and more by whether it can operate on real-world clinical complexity: messy records, missing values, confounding variables, and heterogeneous patient populations. Without that foundation, even the most advanced models risk producing outputs that are technically impressive but clinically brittle.
—
ChronicleBio and the rise of biomedical data platforms as the new AI infrastructure layer
ChronicleBio’s stated ambition—to construct and curate datasets designed to unlock AI-driven insights—places it within an emerging infrastructure category: biomedical data platforms. These companies resemble MLOps or cloud data platforms in spirit, but they face a different order of constraints. Healthcare data is not just large; it is sensitive, regulated, and semantically inconsistent.
To create durable value, platforms in this tier must solve for multiple layers simultaneously:
- Interoperability and standardization: aligning ontologies, coding systems, and clinical vocabularies across institutions
- Data provenance and auditability: ensuring every transformation is traceable for scientific and regulatory scrutiny
- Privacy and compliance: operationalizing HIPAA, GDPR, and evolving digital health rules without crippling usability
- Multimodal integration: combining EHR data with genomics, imaging, wearables, lab results, and digital biomarkers
- Annotation and clinical validation: pairing machine learning pipelines with domain expertise to reduce label noise and bias
This is also where the concept of AI-enabled “digital twins” becomes more than a buzzword. High-fidelity patient simulations—models that can forecast disease trajectories or treatment responses—require datasets that connect biology across scales: genome to phenotype, physiology to behavior, and environment to outcomes. The promise is compelling: faster hypothesis generation, more targeted trials, and more personalized care. The prerequisite is equally clear: data depth, not just model depth.
Simo’s stance effectively argues that the most important “model” in healthcare may be the data model—the schemas, standards, and governance that make medical reality legible to computation.
—
Capital, partnerships, and governance: where the business stakes are headed
From a market perspective, Simo’s approach aligns with a visible investment rotation toward “picks-and-shovels” opportunities—companies that enable AI rather than merely market it. Data-centric platforms can become strategic chokepoints in the value chain, especially if they secure trusted partnerships with hospitals, payers, and life sciences firms.
Several forces are converging here:
- Healthcare cost pressure: chronic diseases drive a disproportionate share of spending, making better diagnosis and treatment selection economically attractive to payers and health systems.
- Pharma’s productivity imperative: access to de-identified, high-resolution patient data can reduce late-stage trial failures and compress discovery timelines.
- Consortium economics: rare and chronic disease datasets often require multi-institution pooling to reach meaningful scale and diversity.
Yet the same dynamics heighten risk. Data quality failures—bias, imbalance, poor annotation—can produce models that mislead clinicians and erode trust. Regulatory uncertainty remains a moving target, and public tolerance for opaque data practices is shrinking. The winners in this category are likely to be those who treat governance as a product feature, not a legal afterthought—deploying privacy-preserving techniques (such as federated learning or secure computation where appropriate), maintaining transparent consent and de-identification practices, and building audit-ready pipelines from day one.
Simo’s core proposition ultimately reads as both a vision and a constraint: AI can only be as curative as the biomedical reality we can responsibly measure, standardize, and share. In the race to transform healthcare, the decisive advantage may belong to the companies that make biology computable—carefully, credibly, and at scale.
By
By
By
By
