Intel Razor Lake-AX Targets Robotics AI With 700 TOPS
Intel's Razor Lake-AX platform will deliver 700 TOPS of AI compute for robotics and humanoid machines, launching in 2028 with a 3.88x performance leap over current Panther Lake chips.

Intel has confirmed its Razor Lake-AX processor will launch in 2028 with 700 trillion operations per second of AI compute power. The platform is specifically targeted at premium robotics and humanoid AI applications, representing a nearly fourfold increase in theoretical performance over current Panther Lake mobile chips.
Intel senior vice president Cho Min-sung announced the roadmap at the Semiconductors Powering Humanoids summit in South Korea on October 17, 2026. The company forecasts that customers will begin mass-producing autonomous mobile robots using its technology in 2027, with humanoid robots expected to follow in 2028. This on-device processing capability could enable robots to handle camera feeds, identify objects, and make decisions with reduced latency and less dependence on cloud connectivity.
Performance and Architecture Details
Razor Lake-AX delivers 3.88 times the AI performance of Intel's current Panther Lake-based robotics platform, which offers up to 180 TOPS. This also makes it 2.5 times more powerful than NVIDIA's Jetson AGX Orin, which is rated for 275 TOPS. The chip is projected to reach approximately 700 TOPS at INT8/FP8 precision.
Subsequent leaks suggest the platform will feature either 16 or 32 Xe3P graphics cores, though Intel has not confirmed these details. The GPU die is reported to measure approximately 162.84 square millimeters. The company also plans to reintroduce on-package memory, similar to its Lunar Lake architecture, likely using LPDDR5X or LPDDR6. The chip will utilize a BGA-4326 package with 4,326 pins and employ EMIB and Foveros 3D interconnect technology.
Platform Technology and Interconnects
The platform will include a large integrated GPU based on Panther Lake's Xe3 graphics IP. Intel had previously considered a different system-on-chip, internally rumored as Nova Lake-AX, which was designed to compete with AMD's Strix Halo lineup for consumer PCs. That earlier concept featured a 48-core Xe3 GPU and a 28-core CPU cluster but has been put on hold. Razor Lake-AX is a distinct design focused on physical AI.
Leaks from sources such as Jaykihn claim Razor Lake-AX will reuse CPU cores from the Nova Lake architecture, specifically Coyote Cove performance cores and Arctic Wolf efficiency cores. The chip is also reported to lack the big Last Level Cache rumored for other Razor Lake consumer processors.
Target Markets and Availability
Intel is creating its own SoC to accelerate agentic and physical AI segments, a market where its current Panther Lake platform, offered through the Robinson Bay development kit, is already a potent solution. Razor Lake-AX is not a traditional computing platform for consumer PCs. It is targeted squarely at premium devices in the robotics and humanoid machine sector.
The company has not confirmed whether Razor Lake-AX will ever appear in conventional laptops or mini PCs. Key details such as power consumption, exact memory capacity, and real-world graphics performance remain unspecified. Intel describes the 700 TOPS figure as a roadmap target rather than a finalized performance specification, and has not broken down how much performance will come from the CPU, GPU, or dedicated neural hardware.
Intel plans to introduce a separate Nova Lake platform for robotics applications in early 2027. The company supports its hardware with software tools including OpenVINO, Physical AI Studio, and the Robotics AI Suite. Mass production of autonomous mobile robots by Intel's customers is slated to begin in 2027.





