- NVIDIA AI GPUssemiconductor · materiality high
High-performance GPUs (H100, A100) required for training and inference of large language models at scale. NVIDIA chip design in US, manufacturing primarily TSMC Taiwan. Known constraint: Industry-wide GPU supply constraints for AI workloads, allocation-based distribution.
- Cloud compute infrastructuredata center · materiality high
Massive distributed compute capacity for model training runs and serving billions of inference requests. Primarily US-based hyperscale data centers. Known constraint: Power availability and cooling capacity limits in key regions.
- Electricity for computeenergy · materiality high · inference
Sustained multi-megawatt power draw for training runs lasting weeks to months and continuous inference serving. Known constraint: Data center power capacity constraints in key AI hubs.
- Venture capital fundingfinancing · materiality high
Multi-billion dollar funding rounds necessary to sustain capital-intensive model development and infrastructure costs before profitability. US venture capital markets.
- AI research talentlabor · materiality high · inference
Elite machine learning researchers and engineers with expertise in transformer architectures, reinforcement learning, and AI safety. Concentrated in San Francisco Bay Area, some remote. Known constraint: Limited global pool of top-tier AI researchers, intense competition from other AI labs.
- Internet infrastructurecommunications · materiality medium · inference
Global network connectivity to deliver API responses and web-based Claude interface with sub-second latency requirements.
- AI governance frameworksregulatory · materiality medium
Evolving AI safety standards, export controls on advanced models, and data privacy regulations affecting product deployment and capabilities. US federal (Executive Orders, NIST), EU AI Act, UK frameworks. Known constraint: Uncertain regulatory trajectory for frontier AI systems.