Fullhan Micro Net Profit Soars 10x, AI-ISP Chip Targets Robotics
On July 31, Fullhan Micro (300613), a leading visual chip maker, published its 2026 half-year earnings forecast. Revenue is projected at 1.4-1.5 billion yuan, up 103.48%-118.01% year-on-year. Net profit attributable to shareholders is 270-350 million yuan, a surge of 1072.72%-1420.19%. Q2 saw record-high quarterly revenue and profit, with shares closing at 56.23 yuan, market cap around 13.1 billion yuan.
The results could be easily read as a cyclical rebound driven by memory price hikes. In the first half of 2026, global electronics faced sharp DRAM/NAND price increases and PCB/component shortages. Fullhan raised product prices, while shipments grew across smart video, IoT, and mobility segments, creating a classic volume-price mix.
However, from a robotics supply chain perspective, the key is not cyclical elasticity but a subtle yet significant comment: "New AI-ISP chips have received positive market feedback, and customer expansion into new fields like robotics is progressing steadily."
Robots need clear vision. In dim factories, humanoid robots navigating staircases against backlight, or vacuum cleaners avoiding obstacles under beds, raw camera frames often contain noise, motion blur, and overexposure. Traditional architectures separate ISP and AI inference, but AI-ISP embeds lightweight neural networks into the imaging pipeline, reconstructing images during denoising and HDR. Fullhan leverages its expertise in surveillance IPC and automotive ISP to package these capabilities into AI-ISP SoCs with on-device NPUs, making the transition to robot camera modules natural.
Fullhan is not a pure-play robotics stock but occupies a critical layer in the robot supply chain: the perception entry point. A robot with cameras may use Rockchip RK3588 or Horizon Juexi as the main controller, but each lens needs a vision-front-end chip. Fullhan's position is low unit price but high volume, with reuse of security and automotive certifications, and high customer switching costs. As industrial robots, AGVs, inspection robot dogs, and home companion robots scale, AI-ISP orders represent a new growth curve beyond the surveillance base, not an internal replacement.
Comparing peers clarifies the logic. Rockchip's 2026 interim forecast shows revenue up 40%-42% and net profit up 60%-71%, driven by edge AI controllers and AIoT. Allwinner's net profit grows 195%-220%, benefiting from affordable AIoT and robot vacuums. SigmaStar's revenue up 87%-94% and net profit up 584%-650%, with edge AI SoCs ramping in surveillance, automotive, and service robots. Fullhan's profit growth of up to 21x (excl. non-recurring) is amplified by a low base and memory pass-through, but the AI-ISP-into-robotics part is the least cyclical.
Of course, price hike benefits cannot be touted as pure skill. Fullhan is fabless; DRAM/NAND dominate its BOM costs. Memory price increases lead to chip price hikes and margin repair—this is industry beta that may fade. In Q1-Q3 2025, net profit fell over 50%, so high H1 2026 growth partly reflects low base. The real alpha lies in two things: AI-ISP design wins with leading robot customers, and whether management's 3-5 year revenue target of 4 billion yuan materializes as the company pivots from a 1.6 billion yuan surveillance ceiling to industrial vision, automotive, and embodied AI.
For the robotics sector, Fullhan's interim report is a probe. It proves that edge visual chips have crossed the surveillance bridge and reached the robot perception deck. Short-term earnings rely on cycles, mid-term valuation on edge AI, and long-term optionality on robot multi-camera penetration. While robot makers are still burning cash on gait tuning, chip companies making 'eyes' have already made their financials positive—a footnote to the 2026 robotics bull run where components hit inflections earlier than complete machines. (For more in-depth humanoid robot articles, follow WeChat account '人形大讲堂'.)