$CEREBRAS - Cerebras has developed wafer-scale chips that are 58x larger than competitors, using fast memory to achieve 15x faster inference than GPUs. Recently IPO'd at $64B valuation with major contracts from OpenAI ($20B+) and AWS, positioning them as a key player in the AI inference market.
$OPENSOURCE - Open source AI models are significantly cheaper per unit of intelligence compared to closed source models, with only 3-5% difference in quality. Major companies are quietly shifting from closed to open source models, creating a competitive threat to closed source providers.
$DATACENTER - Data centers are the binding constraint across the entire AI industry, limiting all vendors regardless of chip technology. This constraint is expected to persist for 15-18 months, creating sustained demand for data center capacity and infrastructure.
$TSMC - TSMC is described as the greatest manufacturing company on earth and a strategic national champion. They are a critical partner enabling advanced chip production and have been providing sufficient capacity to Cerebras while others face supply constraints.
$PHARMA - Pharmaceutical companies with unique proprietary datasets (genomics, patient records, drug trials) are positioned to extract significant value from AI while being protected from disruption, unlike white-collar professions. Their decades of data collection creates a defensible moat.
Bearish:
$NVDA - NVIDIA's CUDA moat has dramatically eroded, losing 70% market share in frontier models. Two of three leading AI models now use no CUDA at all, and CUDA has no role in inference. GPUs are also extremely expensive and power-hungry at fast token generation.
$WHITECOLLAR - White-collar professionals requiring expertise over a body of knowledge (accountants, lawyers) are most directly threatened by AI, as this is exactly what AI excels at. These professions that stood between ordinary people and specialized knowledge are under significant disruption risk.
$HBM - HBM memory faces extreme supply constraints with very long lead times and high costs. Cerebras has designed around this bottleneck entirely, suggesting structural challenges for those dependent on HBM supply from Samsung, Hynix, and Micron.