Qualcomm Unveils Snapdragon 8 Elite Extreme Gen 6, First 2nm Mobile AI Chip
AI compute capacity continues to expand with new on-device processors, while infrastructure build-outs face increasing challenges from power demands and supply chain constraints.
The story
Qualcomm has introduced its Snapdragon 8 Elite Extreme Gen 6 processor, marking a significant step for on-device AI. Built on TSMC's 2nm process, a first for Snapdragon chips, the Elite Extreme variant boasts a 44% improvement in GPU performance and 40% better GPU efficiency over its predecessor.
Its re-engineered Hexagon NPU delivers 35% better performance and 33% improved AI performance-per-watt, capable of running 30-billion-plus parameter mixture-of-experts (MoE) models on-device. This advancement underscores a growing trend towards powerful edge AI, reducing reliance on cloud infrastructure for certain workloads.
Meanwhile, the broader AI infrastructure build-out continues to grapple with power availability, with a recent Brookings Institution report estimating AI projects will cost $10.3 trillion from 2025 to 2032. Companies like Azio AI Holdings are now prioritizing securing power first, recognizing electricity as the new bottleneck for deploying compute capacity.
Silicon
Snapdragon 8 Elite Extreme Gen 6
Maker: Qualcomm
What: First mobile chip on TSMC's 2nm process, with 44% better GPU performance and 35% better NPU performance for on-device AI.
For Whom: Flagship Android phones and other mobile devices.
Dimensity CX C10 Max
Maker: MediaTek
What: Built on TSMC's 3nm process, features an 8th-generation NPU delivering up to 55 TOPS for AI-first laptops.
For Whom: Upcoming Googlebook devices and AI-first laptops.
The build-out
| Project | Who | Scale | Where |
|---|---|---|---|
| Encinal Gas-Fired Power Plant | South Korea (as part of its US investment commitment) | $22.3 billion | Encinal, Texas, US |
| Nvidia R&D Lab Acquisition | Blackstone (from Nvidia) | $99.6 million for 10MW capacity | Sunnyvale, California, US |
| Compute Power Contract | Anthropic (with Akamai Technologies Inc) | $11.6 billion, seven-year deal | Akamai's distributed infrastructure |
Supply & policy signals
Nvidia CEO Jensen Huang expects to double chip sales in 2027 compared to 2026.
Implication: High-bandwidth memory (HBM) supply is identified as the primary constraint for this growth.
Brookings Institution report projects $10.3 trillion in AI infrastructure spending from 2025-2032.
Implication: This represents an average of 3.63% of US GDP annually, highlighting the unprecedented scale of investment.
Goldman Sachs Research forecasts US data center power demand to reach 66 gigawatts by 2027.
Implication: Electricity supply, rather than chip availability, is becoming the critical bottleneck for AI infrastructure deployment.
What we'll be watching
- Microsoft's planned unveiling of its Maia 300 AI accelerator, expected this fall.
- Further details on South Korea's $350 billion investment commitment in the US, particularly for AI-related energy projects.
- Updates on the deployment of AMD's next-gen MI450 GPUs as part of its 6-gigawatt deal with OpenAI, starting in 2H 2026.
Reporting + analyst voices: grounded via Google Search at publish time.