The mobile 235H quotes 94 TOPS where the desktop 235 says 27 — which figure should AI planning use?

  Do not take either figure at face value. TOPS sums several engines, so dividing 94 by 27 across a laptop and a desktop yields a multiple with no practical meaning. What this part reliably delivers is NPU-resident light work plus short inference bursts from 14 cores at a 5GHz boost with 18MB of L3. A rating that triples does not mean the models you can host triple too.

  It is a solid pick when you want the full mobile light-AI set and a 14-core part for occasional data work. Reading 94 as a ticket to big local models points you at the wrong aisle. The larger the mobile figure, the more you should ask how the chassis is tuned, because a weak body will not deliver that 94 anyway.