- Author is long and bullish TEM: diagnostics, longitudinal clinical data, models and clinical distribution could monetize rising demand for usable healthcare data.
- BMY expanded its relationship through 2031 for $137.5M in fixed payments; a separate Pathos neurology arrangement adds up to $55M, while Data and Applications revenue grew 28% to $93.2M in Q2 2026.
- Starting from a roughly $1.6B 2026 revenue guide, author’s 22–25% growth scenario reaches a $50B valuation around 2032–33; 30–35% growth could bring $100B into 2031–33, depending on sales multiples.
- The case requires profitable execution: Personalis remains pending, frontier-lab contracts are prospective, and reimbursement, validation, competition, integration and dilution could limit shareholder value.
Investment pitches
filter:
76 78 - CBRS’s multiwafer SRAM approach expands model capacity while preserving local access: four CS-3s serve Llama 3.1 70B, while GPT-5.6 Sol reached 750 output tokens/s on an unconfirmed multiwafer installation.
- Capacity is already a constraint: management says Cerebras is sold out, OpenAI has substantial reservations, and remaining contracted obligations total $25.4B; dedicated-capacity contracts are generally take-or-pay.
- Core cloud-and-other-services revenue rose 287% year over year to $127.7M in Q2 2026, while gross margin reached 41.8%. The author remains bullish, but CS-5/6 targets and realized total-cost savings require validation.
64 - At roughly $35B, the author sees CBRS as cheap for 2029–30, comparing its wafer-scale IP and customer relationships with NVDA’s ~$20B Groq deal.
- Commercial traction is strong: Q2 core revenue grew 103%, cloud revenue nearly quadrupled, and remaining performance obligations reached $25.4B; management plans to more than triple 2027 revenue.
- Benchmarks show Cerebras delivering roughly 3.5–7.3× Groq’s tokens per second. The author attributes this to 44GB of SRAM per wafer versus Groq’s 500MB, reducing device communication.
- CS-5 targets up to 10,000 tokens/sec on medium models and 5,000 on frontier models in 2027. The author remains bullish but acknowledges further semiconductor-sector downside and unknowable timing.