Summary: Intel launched Core Ultra Series 3, codenamed Panther Lake, on January 5, 2026 as the first AI PC platform with a compute tile made on Intel 18A. It is set for worldwide availability on January 27 with more than 200 partner laptop designs. Intel claims about 50% higher performance than Lunar Lake at comparable power, with performance figures still awaiting independent verification.
On the evening of January 5, 2026, Intel took a Las Vegas stage to launch Core Ultra Series 3, codenamed Panther Lake — the first AI PC platform whose compute tile is fabricated on Intel’s 18A process. Jim Johnson, senior vice president and general manager of the Client Computing Group, ran the keynote, while CEO Lip-Bu Tan made an unscheduled appearance to underline a promise kept: the first 18A products shipped before the end of 2025, as committed.
The stakes stretch well past a single processor generation. 18A is the wager Intel’s foundry business rides on, and the launch arrived with hard numbers bolted on — worldwide availability January 27, more than 200 laptop designs from partners, and Johnson’s boast that this becomes “the most broadly adopted AI PC platform Intel has ever shipped.”
18A in Volume: The Process Is the Story
Panther Lake stands as Intel’s first client silicon with its compute tile built on 18A, and the node is already running in high volume. Johnson labeled 18A “the most advanced process in the world,” crediting RibbonFET gate-all-around transistors and PowerVia backside power delivery; Intel’s own press kit bills it as the most advanced process node in the United States. For buyers and product teams, the subtext outweighs the slogan: Intel needs this product to succeed commercially so that 18A reads as a credible foundry option for outside customers, not merely an internal trophy.
Architecture: Cougar Cove, Darkmont, and Arc B390
The design marries Lunar Lake’s efficiency with Arrow Lake’s scalability: Cougar Cove performance cores and Darkmont efficiency cores, with Intel claiming roughly 50% more performance than Lunar Lake at comparable power. The integrated Arc B390 GPU shouldered half the show — Intel positioned it at discrete-class gaming performance, 2x a competitor in select titles and 73% faster on average, and it becomes the first integrated graphics to support multi-frame generation. EA took the stage to run Battlefield 6 at up to 145 fps using XeSS with 4x frame generation. The GPU now lives on its own chiplet, and 4K streaming sips a third of the power the previous generation burned.
The AI Pitch Shifts to GPUs and the Edge
Where rivals aim the spotlight at the NPU, Tom’s Hardware observed that Intel loaded its weight onto the GPU for this launch. The concrete AI claims: a 70-billion-parameter model running on-device with a 32K context window, fresh hybrid local-plus-cloud branding under “AI Super Builder,” and the edge and embedded product lines folded into Series 3 beneath the “agentic actioning” banner. A robotics footnote rounded out the show: Oversonic Robotics’ Italian-built RoBee humanoid is being tested with Series 3 edge processors to recognize people and gestures on-device, cloud untouched. It is a different doorway into the same physical-AI era Jensen Huang proclaimed from the same CES stage a day earlier — Intel is selling the client-side compute that physical AI stands on.
January 27: Ecosystem by the Numbers
Preorders opened the morning after the keynote, with worldwide availability on January 27, 2026. More than 200 partner laptop designs are in the pipeline, covering consumer, commercial, and edge use cases. Tan’s framing skipped the embroidery: Series 3 is “the next evolution of the PC.” Given Intel’s past few years, the numbers themselves carry the message — first 18A product in volume, 200-plus designs committed, a firm on-sale date — the platform cadence, on this evidence, has not slipped.
What It Means for Developers and Product Teams
Three practical effects. First, the envelope for local models on x86 laptops keeps widening: should the 70B-at-32K claim survive independent verification, hybrid architectures that infer locally and fall back to the cloud become plannable on mainstream hardware. Second, pulling edge into the same platform lets consumer and industrial workloads share one toolchain, trimming the cost of cross-device deployment. Third, the AI PC is settling in as baseline rather than differentiator: with 200-plus designs shipping NPU-plus-Arc-B390-class silicon, product decisions may simply assume on-device inference exists. The standing caveat holds: figures like “2x” and “73% faster” come from the vendor — wait for independent reviews before staking anything on them.