Razor Lake-AX 700 TOPS vs Panther Lake: analysis

Seven hundred TOPS is the kind of number that kicks the lab door open and demands its own breaker panel. Set it beside Panther Lake’s 180 TOPS and Intel’s Razor Lake-AX target looks like a clean four lane highway into local AI compute. But this is where the spec sheet starts doing card tricks. For AI PC builders, the useful question is not whether 700 is bigger than 180. It is what kind of work is being counted, where that work runs, and whether the number describes your workload or just the chandelier in the showroom.

The key context is that this is not currently framed as a normal laptop reveal. VideoCardz reports that Intel outlined Razor Lake-AX for 2028 as a system on chip for robotics and humanoid systems, after Intel executive Cho Min-sung presented a robotics processor roadmap at the Semiconductors Powering Humanoids conference in South Korea on October 16. That makes the roadmap valuable for AI PC watchers, but not because it proves a specific laptop is coming. It is valuable because robotics is the nasty stress test for on device AI, where perception, decision making, and control all fight for compute like three interns sharing one soldering iron.

VideoCardz puts the roadmap on the bench

According to VideoCardz, Intel’s current Panther Lake based robotics development platforms, also known as Core Ultra Series 3, reach up to 180 TOPS of combined AI performance. The same report places Nova Lake in early 2027 and Razor Lake-AX in 2028 with a 700 TOPS target. That gives us a neat roadmap ruler, even if the ruler has some numbers printed in invisible ink. It tells us Intel wants a much larger on die AI budget for future physical AI systems, but it does not yet tell us the full electrical bill.

@title Roadmap compute anchor
@source VideoCardz.com
@source Notebookcheck News

  Product        Time     AI compute         Role
  Panther Lake  current  180 TOPS combined  platform
  Razor Lake-AX 2028     700 TOPS target    SoC

@caption Panther Lake and Razor Lake-AX show the clean comparison and the useful caveat.

This is the tidy part of the teardown. Panther Lake is the known comparison point in the reporting, and Razor Lake-AX is the future target. If all you need is a directional read, Intel is signaling a large jump in local AI compute for robotics class silicon. If you are planning a machine, though, directional reads are like checking a power rail with your thumb. Entertaining, maybe, but not how you sign off a board.

Notebookcheck shows the missing screw

Notebookcheck reports the same broad shape: Intel has publicly acknowledged Razor Lake-AX, with a 2028 plan and roughly 700 TOPS of AI processing performance, while Panther Lake based robotics hardware reaches up to 180 TOPS. Then comes the line that matters more than the big number: Notebookcheck says Intel has not specified the numerical precision behind the figure, or how much comes from the CPU, GPU, and NPU. That is the tiny screw under the bench that changes the whole repair.

Here is why builders should care without turning this into alphabet soup. A combined AI number can describe a whole chip’s peak capability, but it does not tell you whether your model lands on the NPU, spills onto the GPU, or leans on the CPU like a chair with one short leg. The CPU, GPU, and NPU split matters because software support, scheduling, power behavior, and latency can all depend on which block is doing the lifting. The missing precision matters because comparing two TOPS claims is only clean when the counting method is also visible.

VideoCardz and Notebookcheck make this an AI PC caution light

VideoCardz says Razor Lake-AX would represent almost four times Panther Lake’s theoretical AI compute, and the math is a useful headline anchor. Notebookcheck’s caveat keeps it from becoming a shopping recommendation. Until Intel discloses the precision and the block level contribution, 700 TOPS is best treated as a roadmap voltage, not a benchmark waveform. It tells you potential energy exists. It does not show the signal quality under load.

For future AI PCs, that distinction is the whole game. A robotics SoC aimed at humanoid systems may share ideas with client platforms, but available reporting does not establish that Razor Lake-AX will appear in conventional laptops, mini PCs, or desktops. The right move for builders is to track the platform plumbing: supported software stacks, which accelerators are exposed to developers, how sustained performance behaves, and whether real applications can actually keep the AI blocks fed. Thermal throttling, as always, is where optimistic slides go to learn humility.

The forward looking read is constructive: Intel’s Razor Lake-AX roadmap gives us a strong signal that on device AI compute is becoming a central design axis, not a garnish. The 700 TOPS target is worth watching precisely because it is incomplete today. When Intel fills in precision, CPU, GPU, and NPU details, the number will stop being a movie poster and start becoming an engineering document. That is when AI PC buyers, robotics teams, and workstation builders can decide whether this is the clean getaway car or just a beautiful chassis waiting for a drivetrain.