Second-order optimizer enabling Shampoo-quality LLM fine-tuning on 16GB GPUs

Second-order optimizer enabling Shampoo-quality LLM fine-tuning on 16GB GPUs targets a $12.0B = 1.5M ML teams x $8K ACV (ML infra, tooling & optimizers across enterprises and SMBs) total addressable market with medium saturation and a year-over-year growth rate of 30% (developer tools and ML infra demand growth driven by LLM adoption).

Key trends driving demand: Local and hybrid model training -- more teams want to fine-tune models off-cloud for cost, latency, and privacy reasons, creating demand for consumer-grade tooling.; Efficiency-first ML engineering -- growing focus on compute- and memory-efficient algorithms to reduce costs and environmental impact.; Algorithmic progress in low-rank methods -- recent research makes practical approximations of second-order curvature feasible on smaller hardware.; PyTorch ecosystem maturity -- widespread adoption and extension points lower integration friction for new optimizers..