The infrastructure arms race is accelerating while the actual competitive advantage narrows. Microsoft is shipping Maia 300, its custom silicon play to reduce dependence on NVIDIA, exactly as the chip market demanded. NVIDIA responds by flooding the zone: opening standards (Vera Rubin architecture), shipping new interconnect hardware with Cisco (USFP modules), and conducting a European roadshow to lock in regional compute commitments before competitors gain ground. The message is clear: NVIDIA owns the factory floor, and it's betting that controlling the entire stack, from optics to open standards, matters more than any single component. Meanwhile, GitHub is quietly pushing Copilot deeper into developer workflows through custom agents and code-aware instruction sets, turning the IDE into an agentic interface. Anthropic, by contrast, is publishing red team findings about model self-replication and distillation defenses, which reads less like product momentum and more like defensive posturing on a regulatory track. The gap between infrastructure builders moving product and safety researchers moving papers is widening. One side is scaling factories across continents and embedding AI into daily work; the other is documenting risks in controlled settings. The market is voting with deployment speed, not caution.
Sloane Duvall
A curated reference of models from major AI labs, with open/closed weight status, input modalities, and context window size. American labs tend towards closed weights models and Chinese labs tend toward open weights models.
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None