Nvidia has been strategically investing in AI application layers in recent years to bind long-term consumers of its GPU computing power, which is its consistent investment strategy—previous strategic investments in companies like CoreWeave, Databricks, xAI, Mistral, and Perplexity essentially aim to make the invested companies stable purchasers of its computing power and cloud services, rather than just pursuing financial returns.

As a leading early-stage venture capital firm in Silicon Valley, Lightspeed has long focused on seed and Series A investments in AI infrastructure and applications. Leading the $50 million Series A for Nuance Labs aligns with its ongoing investment rhythm in conversational AI and digital human sectors; Nvidia's participation appears more like strategic positioning for potential large computing power clients rather than purely financial investment.

The low-latency AI digital human niche is in the same competitive quadrant as conversational/digital AI companies like Character.AI, Synthesia, and HeyGen, which have previously received significant funding, but focuses more on the higher-level interaction form of "face-to-face real-time conversation." Its technological iteration path is similar to the evolution of real-time audio and video technologies like Zoom and Tencent Meeting from "able to call" to "low-latency high-definition calls," except this time the interaction object is an AI digital human instead of a real person.

This is essentially a concentration of capital: leading chip manufacturer Nvidia is strategically directing capital further towards application-layer startups that require substantial real-time inference computing power and will continuously consume its GPU products. The mechanism is that Nvidia's business model heavily relies on the explosive growth of downstream AI applications; by investing in such companies, it can lock in potential large customer relationships in advance and indirectly share the growth dividends of the application layer through financial returns, forming a capital closed loop of "selling shovels while holding shares in gold miners."

ABAB News · Cognitive Law

Those who sell shovels will eventually buy shares in gold mines.
For every millisecond of latency reduced, the valuation increases by one position.
Chip manufacturers are not investing in companies, but in future computing power orders.