Frontier AI splits into three distinct markets
Grok sells access, Qwen ships models, Nvidia routes jobs—who really wins?
Frontier AI is no longer a single battleground but three distinct markets with distinct winners. Grok (xAI) sells API access at $2 per million input tokens, Qwen (Alibaba) ships open models like Qwen3.8 with 2.4 trillion parameters, and Nvidia routes jobs via Nemotron 3.5 Lightning and Switchyard to the cheapest capable model. This fragmentation means benchmark leadership doesn’t guarantee deployment control, revenue, or influence.
The shift exposes new leverage points: training-data provenance, electricity markets for compute, and government oversight. Meanwhile, Grok 4.6 is positioning itself against GPT-5.6 Sol on a composite intelligence index, while durable consumer tools like Grammarly and video generators continue to dominate app charts. The lab with the highest score may not deploy most widely, and the most powerful player might just be the intermediary quietly directing demand.
- Frontier AI splits into three markets: access (Grok/xAI), ownership (Qwen/Alibaba), and routing (Nvidia/Nemotron 3.5 Lightning)
- Grok 4.6 matches GPT-5.6 Sol on a composite index and costs $2M/$6M tokens; Qwen3.8 is a 2.4T-parameter MoE model; Nvidia’s Switchyard routes to the cheapest capable model
- Durable consumer tools (Grammarly, video generators) and open models (Muse Glimmer) are gaining traction, while benchmarks no longer dictate deployment
Why It Matters
This shift redefines power in AI—from model quality to control over access, ownership, and job routing, reshaping who profits and who sets the rules.