Honor’s Robot Phone and the New AI Chip Arms Race

Honor launched its “Robot Phone” in China on August 12, 2026, a Snapdragon 8 Elite Gen 5 device with a motorized four-axis camera gimbal built into the body, starting at RMB 9,999 (about $1,500) according to Honor’s own announcement and 9to5Google’s coverage of the launch. The device is the most visible example of a broader pattern this year: Qualcomm, MediaTek, Apple and Samsung have all shipped new mobile chips built around substantially upgraded neural processing units (NPUs), each pitched as the engine for “agentic AI” that runs on the phone itself rather than in the cloud. The specs are real and independently documented. Whether they translate into daily usefulness that matches the marketing is a separate, less settled question.

What actually launched

Honor’s Robot Phone pairs a Qualcomm Snapdragon 8 Elite Gen 5 chipset with a 4-degree-of-freedom titanium gimbal — described by Honor as using “100+ precision components” and a 2.6-gram titanium motor capable of 360 degrees of rotation per second — mounted behind a dual 200-megapixel rear camera system co-developed with ARRI, the cinema camera maker, per Honor’s official product page. The gimbal lets the camera track a subject or rotate the frame between portrait and landscape without the user turning the phone. Honor’s MagicOS software layer adds a “YOYO Robot Mode” AI agent that can, per the company, execute multi-step tasks and follow voice or gesture commands. Forbes contributor Prakhar Khanna, who tested the device, confirmed the gimbal activates via hand gesture and that the YOYO agent successfully handled a spoken drink order during his hands-on, while also reporting that subject tracking “worked in bright light but struggled to track” a subject in low light with multiple people present. The phone is currently a China-only release with no announced international rollout, according to both Honor’s materials and 9to5Google’s reporting.

The chips underneath: what an NPU actually does

An NPU is a processor block designed specifically for the matrix-multiplication math that neural networks run — the same category of calculation used to recognize a face, transcribe speech, or generate a sentence of text. Unlike a phone’s general-purpose CPU or graphics-focused GPU, an NPU is narrower in what it can do but far more power-efficient at that one job, which is why manufacturers now build them as a third distinct block on the same chip. Running AI tasks locally, on the NPU, rather than sending them to a cloud server has two practical implications companies emphasize: the request doesn’t need a network connection, and the raw data (a photo, a voice recording, a calendar entry) doesn’t necessarily have to leave the device to be processed.

The 2026 chip generation, claim by claim

Qualcomm’s Snapdragon 8 Elite Gen 5, announced in September 2025 and now shipping in devices like the Robot Phone, includes a Hexagon NPU that Qualcomm’s own newsroom describes as “37% faster” than the prior generation, alongside a 20% CPU performance gain and 23% GPU improvement, according to Qualcomm’s official release. Qualcomm frames the chip’s purpose explicitly around what it calls “agentic AI assistants” that can “see what you see, hear what you hear and think with you in real time,” in the words of Chris Patrick, SVP and General Manager of Mobile Handsets at Qualcomm Technologies, quoted in the same release. Notably, Qualcomm’s public materials do not disclose a specific TOPS (trillion operations per second) figure for the Hexagon NPU — the improvement is stated only as a relative percentage.

MediaTek’s Dimensity 9500, covered by GSMArena and detailed in MediaTek’s own press release, uses a ninth-generation NPU 990 with what the company calls “Generative AI Engine 2.0,” claiming double the compute throughput of its predecessor, a 100% speedup running 3-billion-parameter language models locally, and up to 56% lower power draw at peak performance. MediaTek also touts support for “BitNet 1.58-bit” compressed models, a technique that shrinks model size to run more efficiently on limited hardware. As with Qualcomm, MediaTek’s release does not publish an absolute TOPS figure — its NPU claims are all comparative to the prior Dimensity generation.

Apple’s approach, per MacRumors’ review of the September 2025 launch materials, keeps a 16-core Neural Engine across the A19 and A19 Pro chips but adds per-core “Neural Accelerators” paired with each GPU core, which Apple says deliver “4x the peak compute” of the A18 Pro for on-device model tasks such as Apple Intelligence and on-device Siri. Apple also does not publish a TOPS figure for the Neural Engine in its public materials.

Samsung’s Galaxy S26, per Samsung’s own U.S. product page, claims a “38% boost to the advanced NPU” and “36% faster AI performance” compared to the Galaxy S25+, powering features like Photo Assist (natural-language photo editing) and Now Brief (personalized daily summaries), alongside a “Personal Data Engine” Samsung says processes personal data on-device with Knox Vault encryption.

Marketing claims versus what’s independently verified

Every percentage figure above — Qualcomm’s 37%, MediaTek’s up-to-100%, Apple’s 4x, Samsung’s 38% — is a manufacturer’s own comparison against its own prior-generation chip, not a figure produced or replicated by an independent lab in the sources reviewed for this article. None of the four companies’ public materials disclosed a standardized, cross-comparable TOPS number, which makes head-to-head comparison between Qualcomm, MediaTek, Apple and Samsung chips difficult using company materials alone. What is independently confirmed, from Forbes’ hands-on testing of the Robot Phone specifically, is that the gimbal hardware and gesture activation function as advertised, that the AI voice agent can execute a simple real-world task, and that AI subject tracking is inconsistent in low light — a concrete gap between the marketing framing (“embodied AI”) and tested performance.

Competitive landscape

The pattern across Qualcomm, MediaTek, Apple and Samsung in 2025–2026 is that each company is now marketing its NPU generation-over-generation improvement as the headline chip feature, ahead of traditional CPU/GPU benchmarks — a shift DigiTimes’ reporting on Honor’s launch describes as Honor repositioning itself as “an AI device maker rather than just a smartphone brand.” Honor’s Robot Phone is notable less for its NPU (a Snapdragon chip shared with many other 2026 flagships) and more for pairing that silicon with new mechanical hardware — the gimbal — as a second, physical bet on what “AI hardware” means beyond the chip itself. Honor also unveiled a companion humanoid robot at MWC 2026, according to Honor’s own newsroom, suggesting the company sees the phone as one node in a wider embodied-AI hardware strategy rather than an isolated product.

Privacy and practical implications

The consistent claim across all four chipmakers is that moving AI inference on-device reduces how much personal data has to travel to a cloud server — Qualcomm’s release specifically states user data “remains on-device,” and Samsung’s page describes encrypted on-device storage via Knox Vault for its Personal Data Engine. These are architecture claims worth taking seriously, since on-device processing is a genuinely different data flow than cloud inference. But none of the sources reviewed here include an independent security audit of what data these NPU-driven features actually retain, transmit, or share with cloud components for features that exceed on-device model capacity — a gap consumers should weigh against the marketing language.

What happens next

Honor’s Robot Phone remains China-only for now, with no confirmed global launch date in the sources reviewed. The broader NPU generation it runs on — Snapdragon 8 Elite Gen 5 — will spread into many more 2026 flagship phones from other manufacturers, making Qualcomm’s, MediaTek’s, Apple’s and Samsung’s competing NPU claims a recurring comparison point through the rest of the year as more devices ship and independent reviewers get hands-on time to test them.

What Smashology verified

Evidence limits and unresolved questions

• None of the major chipmakers published a standardized TOPS figure for their newest NPUs in the official materials reviewed, making direct cross-brand performance comparison impossible from company sources alone.
• All performance percentage claims are manufacturer comparisons to their own prior-generation chip, not measurements against a common independent benchmark suite.
• Independent hands-on testing of the Honor Robot Phone specifically was limited, in the sources reviewed, to a single Forbes contributor’s impressions; broader reviewer consensus was not yet available in the window searched.
• No independent security or privacy audit of any of these NPUs’ on-device data handling was located; “stays on-device” claims currently rest on manufacturer statements alone.
• The Honor Robot Phone’s international availability, pricing outside China, and long-term durability of the mechanical gimbal system remain unknown, since the device has only been on sale in China for a matter of days as of this writing.

Sources

• Honor Global: HONOR Launches Revolutionary Robot Phone, Ushering a New Era of Cinematic Mobile Filmmaking and Embodied AI
• Honor Global: HONOR Advances Its AI Vision at MWC 2026 with Robot Phone, Humanoid Robot and Magic V6
• 9to5Google: Honor finally launches Robot Phone starting at around $1,500
• Forbes (Prakhar Khanna): The Honor Robot Phone Is A Ridiculous Idea That Actually Works
• DigiTimes: Honor’s Robot Phone launch signals a bigger bet: turning smartphones into AI hardware
• Qualcomm: Snapdragon 8 Elite Gen 5 Establishes New Consumer Experiences and Sets New Industry Benchmarks
• GSMArena: Qualcomm Snapdragon 8 Elite Gen 5 SoC features, specs
• MediaTek: MediaTek Dimensity 9500 Unleashes Best-in-Class Performance, AI Experiences, and Power Efficiency
• GSMArena: MediaTek announces Dimensity 9500 flagship chipset
• MacRumors: A19 vs. A19 Pro: iPhone 17 Chip Differences
• Samsung: Galaxy S26 & S26+ | Galaxy AI

Method note: Smashology verifies technology claims by tracing each figure to the manufacturer’s own newsroom or product page and, separately, to independent hands-on testing or trade press coverage where available. Where a specific number appears only in company marketing with no independent replication found, this article states that explicitly rather than presenting it as confirmed fact.