Multimodal AI Competition Shifts Toward Cost as Google's EmbeddingGemma 2 Raises Questions

At a glance

Google DeepMind has announced EmbeddingGemma 2, a technology that converts text, images, and audio into numbers so they can be compared in one system. The article argues that this research is shifting the focus of AI competition from large models to operating costs.

Key points

1. 1

   Companies look at GPU usage, response speed, and monthly operating costs before asking how smart an AI is.

2. 2

   Multimodal AI handles photos, voice, and text together, which increases data volume and quickly raises computing demand.

3. 3

   The article says Korean companies need the ability to organize and connect their data well more than they need to own their own models.

4. 4

   However, the announcement does not include performance figures or the size of cost savings, so it is hard to draw firm conclusions about its effects.

Why it matters

If the cost structure of AI services changes, how companies adopt the technology and how vendors compete will also change.