VENDORiQ: Microsoft’s New AI-Enabled PCs – Strong ARM Tactics or Shape of Things to Come?

Delve into the analysis of Microsoft's latest AI-enabled PCs, uncovering the performance enhancements, compatibility hurdles, and crucial data privacy implications.

The Latest

24 May 2024 – Microsoft announced that their new laptops are 58% faster than Apple’s M3 models. The new ‘Copilot+ PC’, a Surface laptop, runs Windows on ARM architecture using the Qualcomm Snapdragon Elite chip. Notable features include a 40-TOPS neural processing unit (NPU) for AI tasks and an on-device Recall feature that captures user activities. While key applications like Chrome and Microsoft Office are compatible, others, particularly from Adobe, may not work on ARM architecture. Data privacy concerns have been raised due to the recall feature that constantly takes snapshots of activities. These may be overblown for many organisations, but must still be carefully considered.

Why It Matters

Microsoft’s Copilot+ PC announcement is significant because of its potential impact on enterprise laptop refresh cycles. 

The ARM-based architecture offers substantial performance gains and improved battery life, which could transform productivity and mobile working conditions. 

While the ARM chip offers many advantages for business users, the ARM set presents challenges for IT professionals deciding on laptop directions. ARM is not new, having been used in the Surface Pro X line-up. However, compatibility issues with existing x86-based software could disrupt workflows and necessitate re-evaluation of software portfolios. In addition, Intel is developing the Lunar Lake x86 chipset which may see it deliver the same benefits now available with ARM.

What is new (for ARM laptops) is the inclusion of NPUs specifically aimed at locally accelerating AI processes. The NPUs in the new laptops run at 40 Tera Operations Per Second (TOPS)1.  

The addition of NPUs in devices is the shape of things to come. IBRS expects that AI accelerator hardware will be standard in all devices – not just laptops but tablets, smartphones, and potentially new audio and visual form factors, within the next three years.

While innovative for search and productivity, the Recall feature introduces new data privacy considerations that must be carefully managed to comply with regulations and protect sensitive information. 

The Recall function constantly takes screen snapshots of the device’s activities and holds the data for over a three-month period. Recall can then search for user activity and files based on past activities. This is a significant and deep productivity improvement. Since the Recall software can classify images on the device screen and provide deep searching and analysis, it raises data privacy concerns. Such concerns may increase significantly as advantages in new AI models – in particular Graph AI – evolve.  

Data captured by Recall remains on-device and can be deleted by users in much the same way as clearing a browser cache. 

At this time, Microsoft has not indicated that it will access Recall data, and its Enterprise privacy statement suggests that it will not do so in the near future. However, enterprises will need to grapple with local admins’ rights over the Recall cache of workers’ devices.

Who is Impacted

  • Desktop managers: Need to manage integration and compatibility with existing systems.
  • Cyber security officers: Must address potential data privacy and security concerns introduced by the Recall feature.
  • Software developers: Likely required to adapt or optimise existing applications for ARM architecture and prepare for leveraging new AI acceleration capabilities of devices.

Next Steps

  • Assess the compatibility of your organisation’s critical software with ARM architecture.
  • Develop and implement policies to ensure the secure use of the Recall feature, prioritising data privacy and security.
  • Monitor developments in x86 and ARM architecture to inform future hardware procurement decisions.

Additional Reading

  1. TOPS is a metric used to measure the theoretical peak performance of NPUs or AI accelerators for executing neural network operations. TOPS quantifies how many trillion operations (multiplication, addition, etc.) an NPU can perform per second when running at maximum utilisation. ↩︎

Trouble viewing this article?