Dylan PatelFounder and Chief Analyst, SemiAnalysis · The semiconductor-AI stack

Why NVIDIA Built the LPU

22 Apr 2026 · 1 min

Watch on YouTube

0:32

Dylan Patel explains NVIDIA's LPU, a chip designed to outperform GPUs for certain inference tasks. The move toward disaggregated inference lets NVIDIA optimize for what customers actually need instead of building one chip for everything.

What do you think about the LPU stuff, right? Because that's kind of like a hot topic right now. NVIDIA does know the most about its customers. Like, of course, the customers who directly ask NVIDIA for stuff that will improve their inference performance. So I guess in a way, disaggregated inference is the future. And to really optimize each specific part of disaggregated inference is that this LPU does in fact surpass GPU in some aspects. And that's why NVIDIA developed it.

More from Dylan

2:04

Even the frontier labs underestimated their own demand.

3d ago

1:52

The AI buildout needs more capital than backstops can cover.

4d ago

1:52

The fastest finish is not always the most efficient one.

5d ago

1:23:18

Ep. 037 - Who's Funding the $11 Trillion AI Buildout? (Capital Markets)

6d ago

0:48

"It's like a zombie apocalypse, but you don't know how many waves of zombies are coming."

6d ago

2:01

Open-source models are already taking the easier tasks. The bull case for frontier labs rests on one

1w ago

0:44

A $200 AI subscription can be worth $12,000 in tokens. Which plan delivers it is not close.

1w ago

50:14

Ep. 036 - Anthropic Subscriptions Beat OpenAI 5x, Depending on Your Workload (Tokenomics)

1w ago

0:39

What four weeks of SemiAnalysis consulting can do to a company

1w ago

34:41

Ep. 035 - Tech DD’s, Performance Projections, Supply Chain, Investment Thesis (Consulting)

1w ago

0:50

Are GPUs money printers?

1w ago

See all content