Fu Sheng, chairman and CEO of Cheetah Mobile and founder of its robotics subsidiary OrionStar, has spent 2026 making a specific, technical argument about why most AI agents on the market today aren’t actually ready for the jobs they’re being sold to do — and backing it with a robotics business whose revenue growth suggests his narrower, accuracy-first approach is finding real commercial traction.
The “1% Wall” and the “Devil’s Zone”
Speaking at WAIC Up! 2026 in an interview on January 16, 2026, Fu laid out what he called the core limitation holding back general-purpose AI agents: a system that’s 98% accurate sounds impressive, but that remaining 1-2% gap is where enterprise deployment actually breaks down. “In some scenarios, being just 10% better than others allows you to capture the entire market,” he said. “In others, a 10% error rate is fatal.” His proposed fix isn’t chasing a universal, do-everything model to higher accuracy — it’s deliberately narrowing an AI agent’s scope to a specific domain, where hitting reliable, near-total accuracy becomes achievable, a pattern he compared to the early smartphone market, where the first mobile phones dominated despite inferior battery life because they solved one problem well rather than many problems adequately.
For robots specifically, Fu argued the bar is even higher: 99.9% accuracy, not 98% or 99%. That final 0.1% gap, which he called the “devil’s zone,” is what separates a robot that’s merely usable from one that’s genuinely trustworthy in a real physical environment, and closing it requires three concrete safeguards working together — sensor redundancy, verification through smaller, purpose-built models double-checking a larger model’s output, and continued human oversight rather than fully unsupervised autonomy. His broader framing described the goal of AI development as a shift “from machines evolving around humans to true machine-human adaptation” — natural-language interaction paired with genuinely autonomous task completion, rather than humans continuing to adapt their own behavior to fit a machine’s limitations.
The Business Case: Robotics Now a Standalone, Fast-Growing Segment
That accuracy-first thesis is showing up directly in Cheetah Mobile’s financials. The company’s Q1 2026 results, reported in June 2026, showed total revenue of roughly 259 million yuan (about $37.5 million), essentially flat year-over-year — but within that flat top line, the robotics and other segment (OrionStar’s core business) grew 176% year-over-year to about 51 million yuan, now representing roughly 20% of total company revenue and, for the first time, broken out as its own independent reporting segment starting that quarter. The segment’s adjusted operating loss also narrowed by 57% year-over-year, evidence that the business is scaling toward profitability rather than simply growing revenue while losses widen. Cheetah Mobile’s separate cloud and AI infrastructure business grew 68% year-over-year, contributing another 18% of total revenue — meaning more than a third of the company’s business now comes from the two newer, AI-and-robotics-driven segments Fu has pushed the company toward, even as its legacy consumer software business remains largely flat.
A Company Explicitly Calling 2026 a Transition Year
Cheetah Mobile has described 2026 in its own investor materials as a transitional year, moving the company “from a traditional internet company toward AI-enabled applications, AI agents and robotics, with early-stage commercial validation underway” — language that frames the current growth numbers as evidence of an emerging business proving itself, not yet a mature, fully scaled one. That framing matters given the company’s history: Cheetah Mobile built its original scale internationally through consumer mobile utility apps, including security and device-cleaning tools that reached hundreds of millions of downloads worldwide, a very different product category from the enterprise service robots — cleaning, delivery, and reception robots — that OrionStar now sells commercially.
A Founder Reframing His Own Company Around a Narrower Bet
Fu’s public commentary and Cheetah Mobile’s own segment reporting both point to the same underlying strategic choice: rather than chasing the broad, general-purpose AI agent race that dominates headlines from larger labs, Fu has directed OrionStar toward physical robotics applications where his own “1% wall” argument suggests a narrower, more rigorously validated product can actually clear the reliability bar real customers require — a bet that Cheetah Mobile’s own Q1 2026 numbers suggest is beginning to pay off financially, even as the segment remains a minority share of the company’s overall revenue.
What This Means for Philippine Founders
Fu’s “1% wall” framing is directly useful for Philippine AI founders under pressure to build a broad, general-purpose product: his argument that narrowing scope is what actually gets a product from “impressive demo” to “enterprise-deployable” is a concrete strategic lesson for local startups with far less capital than the general-purpose labs they’re implicitly competing against. His robotics-specific standard — 99.9% accuracy, not 98%, achieved through sensor redundancy, model-based double-checking, and continued human oversight rather than full autonomy — is also a useful benchmark for any Philippine company evaluating a robotics or physical-automation vendor pitch: a vendor unable to articulate a comparably specific reliability architecture is likely still in demo-stage territory, not ready for real deployment in a Filipino warehouse, hotel, or retail environment.
Share this article