Why there isn't any top LLM providers investing on diffusion LLM?

A year ago, I would’ve said Diffusion LLMs were an interesting idea but still far from practical. They’re still pretty rough, but Mercury 2 now makes it seem like they might finally be getting close to usable. That said, aside from Meta, Ant, and Inception/Mercury, it doesn’t seem like many labs are seriously investing in them - especially the major ones like OpenAI, Anthropic, Google, xAI, or even architecture-focused teams like DeepSeek and Kimi.