Key Takeaways

  • NCI is supporting artificial intelligence in cancer research and care through funding opportunities, events, data infrastructure, and interoperability programs.
  • The Cancer Research Data Commons now encompasses about 17 petabytes across seven public data commons, giving AI developers access to substantial research assets.
  • Data quality, controlled access, clinical validation, and consistent terminology remain central challenges for healthcare organizations and technology suppliers.

Artificial intelligence is becoming a larger part of how the National Cancer Institute approaches cancer research and care. NCI is backing that shift through funding opportunities, industry and research events, standardized data resources, and cloud infrastructure designed to make complex datasets more usable.

For technology suppliers, AI requires a well-structured data environment. Cancer models may draw from genomic sequences, medical images, proteomic measurements, clinical records, and treatment histories. Those sources use different formats, terminology, and access controls, which can turn model development into a lengthy integration project before meaningful analysis begins.

NCI is addressing part of that problem through the Cancer Research Data Commons, a secure, cloud-based infrastructure for submitting, sharing, and analyzing cancer datasets. As of 2026, the CRDC contained about 17 petabytes and seven public data commons.

Named resources include the Genomic Data Commons, Imaging Data Commons, Proteomic Data Commons, and Clinical and Translational Data Commons. Together, they give researchers routes into multiple forms of cancer data without requiring every institution to build its own large-scale storage and distribution environment.

A computer-vision model trained on pathology or radiology images has different requirements from a system analyzing genomic variants. Multimodal research makes the engineering challenge even harder because records must be connected without erasing important clinical context. A sophisticated model provides little value if the underlying datasets describe the same diagnosis in incompatible ways.

Storage volume alone does not create AI-ready data. NCI's Enterprise Vocabulary Services maintains more than 208,600 standardized cancer-research concepts. Its caDSR repository provides common data elements, forms, and models intended to support interoperable data collection and analysis. Broader healthcare workflows also commonly use OMOP and PCORnet common data models, along with SNOMED CT and LOINC terminologies.

This standards layer has direct commercial implications. Vendors supporting research institutions may need to map their schemas to NCI Common Data Elements, preserve provenance, and document transformations used in model pipelines. Platforms such as REDCap and Medidata Rave can participate in NCI-aligned workflows, but interoperability depends on implementation details rather than the presence of a familiar product name.

The data estate continues to grow. The NCI Genomic Data Commons released more than 8,000 new whole-genome-sequencing variant calls in January 2025. That expansion illustrates both the opportunity and the operational burden. More genomic records can support broader analyses, but they also increase requirements for compute capacity, metadata management, version control, and reproducible processing.

Access management introduces additional technical requirements. According to the Journal of the National Cancer Institute, NCI's Data Access Committee approved more than 16,000 requests for controlled-access data during 2025. This demand demonstrates why data governance requires continuous technical enforcement. Identity controls, authorization, consent restrictions, audit trails, and approved-use conditions must follow data through the entire AI lifecycle.

For cloud providers, clinical software companies, data-engineering specialists, and AI developers, NCI's direction creates an addressable market around secure research environments and interoperable workflows. Opportunities include data harmonization, privacy-preserving computation, model monitoring, lineage tracking, and tools that help researchers reproduce results across institutions.

Translating research performance into clinical value introduces distinct validation requirements. Cancer-care models can encounter bias, changing patient populations, inconsistent documentation, and differences between controlled research datasets and routine clinical records. Validation therefore requires multidisciplinary review alongside accuracy scoring.

NCI's combination of funding, shared infrastructure, and standards indicates that cancer AI is developing as an ecosystem reliant on structural data governance rather than isolated algorithms. Organizations navigating this space must connect models to governed, well-described data while giving researchers clear evidence of how results were produced to achieve durable clinical integration.