Smartphone makers are ditching generic processors in favor of custom-designed neural chips built specifically for on-device AI tasks. Vivo's upcoming X Fold6 marks a rare industry milestone: a flagship processor redesigned from the ground up to handle multiple AI assistants running simultaneously without draining the battery or overheating the device.

What's Driving the Move Toward Custom Neural Processors?

For years, running artificial intelligence on smartphones meant forcing rapidly evolving AI models onto fixed, generic processor architectures. When new AI techniques emerged, the underlying hardware couldn't adapt, resulting in wasted power and inefficient performance. The problem became especially acute when devices needed to run multiple AI tasks at once, which triggered thermal throttling, stuttering interfaces, and catastrophic battery drain.

Vivo spent two years co-developing a bespoke system-on-chip with MediaTek called the Dimensity 9500 Super Edition, specifically engineered for foldable devices running heavy multitasking and edge AI workloads. The results speak to why this approach matters: the custom neural processing unit (NPU) delivers 111 percent more peak performance while consuming 56 percent less power than the standard version.

How Does Custom Silicon Improve Real-World Performance?

The performance gains translate directly into practical benefits for users. The X Fold6's custom AI voice engine handles meeting transcription 7 times faster than conventional approaches, with a 57 percent improvement in word summarization speeds. Text summarization across multiple documents runs 20 percent quicker, and the device prevents UI lag when resizing or dragging files across four to five active windows simultaneously.

This efficiency matters because modern foldable phones with large internal displays demand sustained, peak-load performance from their neural processors. Running four or five applications alongside heavy AI workloads presents a massive engineering challenge. By tailoring the silicon to the specific workload, Vivo solved the thermal and power constraints that would otherwise make such multitasking impractical.

Steps to Understanding Custom AI Chip Design

• Domain-Specific Instructions: Custom neural processors add specialized instructions directly into the chip's core pipeline to accelerate specific AI tasks like matrix multiplication and vector processing, eliminating the need to shuttle data to external accelerators.
• Unified Memory Architecture: Instead of moving data between a standard CPU, GPU, and separate neural processor, custom chips integrate these functions into a single tightly coupled system, drastically lowering latency and power consumption.
• Application-Specific Optimization: Chip designers can strip out unnecessary architecture and tune the silicon precisely for the AI models and workloads the device will actually run, shifting edge AI from a cloud-dependent luxury to truly autonomous local processing.

Why Is This Trend Accelerating Across the Industry?

The semiconductor industry is recognizing that generic processors no longer serve the needs of AI-heavy applications. McKinsey predicts that autonomous driving chips will shift value dramatically toward neural processing units, memory bandwidth, and low-latency interconnects, with autonomous driving semiconductor revenue expected to reach $46 billion by 2035, up from $5.6 billion in 2025.

Beyond smartphones, custom neural processors are becoming standard across edge AI applications. Automotive manufacturers are building specialized chips for advanced driver assistance systems, industrial equipment relies on custom silicon for real-time sensor processing, and consumer devices from security cameras to smart appliances increasingly feature tailored neural engines.

The broader shift reflects a fundamental change in how the industry approaches hardware design. Rather than forcing fluid, rapidly evolving AI software onto rigid, proprietary silicon architectures, companies are now co-designing hardware and software from day one. The silicon is built for the workload, not the other way around.

Vivo's investment in custom silicon for the X Fold6 demonstrates that even consumer smartphone makers are willing to commit significant engineering resources to specialized neural processors. This signals that custom AI chips are no longer a luxury reserved for data centers; they are becoming essential for delivering responsive, power-efficient on-device intelligence across mainstream consumer products.