I just noticed an interesting development in AI infrastructure over the past period. At the end of last year, major companies acquired Groq, a chip company focused on low-latency processing, for US$20,000 million, and from what I saw at GTC last week, they have just launched the Groq 3 LPU chip, manufactured using Samsung’s 4nm technology.

What caught my attention is the rationale behind this move. It seems that the AI services market is shifting. It’s no longer just about boosting traditional processing power; it’s about giving users more choices. Some people want speed, some want fast responsiveness, and some want both—while being willing to pay different prices.

In fact, it’s similar to expanding the scope of the market—not just the usual expansion. Now there are new market segments with low latency and a high unit cost. Beyond the traditional approach that emphasizes volume and high speed, the same model can be priced differently depending on response speed. Even if the processing volume is lower, the price per unit can make up for it.

What’s interesting is that Groq’s LPU architecture is well known for its guaranteed low latency. When combined with the high-speed GPU approach from major players in the market, it’s like filling the missing gap in their product lineup. This isn’t just straightforward competition—it’s a complementary move that enhances their overall capabilities, which is a very smart strategic move.