KM4BI KM4BI

Thinking Machines’ first AI model, California loses $3.2B to Texas, TSMC adds $100B

TBPN breaks down Inkling, Thinking Machines’ open-weights bet, and the industrial signals from Saronic, California Forever, and TSMC.

Inkling, the first open-weights model from Thinking Machines Lab, is framed less as a raw benchmark victory and more as infrastructure for customization. The episode connects that strategy to Tinker, the distillation debate, geopolitical pressure around Chinese open models, and broader industrial signals from Saronic, California Forever, and TSMC.

What Inkling signals

Thinking Machines Lab released a model with 975 billion total parameters and roughly 41 billion active at a time. The hosts emphasize the positioning: Inkling is not pitched as the strongest overall model, but as a broad, adaptable foundation model that fits a fine-tuning and customer-integration business.

Open weights, distillation, and sovereignty

The conversation treats Inkling as a potentially important Western open-weights option at a moment when access to Chinese models may become more politically sensitive. The distillation discussion is deliberately nuanced: Thinking Machines says it used synthetic data from open-weight models during supervised fine-tuning, so the question is not binary.

Industrial policy in practice

California Forever lost Saronic’s automated Port Alpha shipyard project to Brownsville, Texas. The practical takeaway is that large industrial projects move toward jurisdictions that can provide fast approvals, clarity, and incentives.

Chips and the long AI buildout

TSMC raised capex and plans another $100 billion of investment in Arizona fabs. Even though investors worried about overspending, the episode treats the move as a strong signal that a central semiconductor supplier still sees durable AI-driven demand.

Source