Commentary: The rise of Arm on the desktop spells trouble for Intel

This week, Dell, HP, Lenovo, and others have announced new laptop and tablet hardware based on the Qualcomm Snapdragon X processor. With integrated NPU (and the only one supporting Microsoft Copilot+), can Intel continue to compete on the desktop?

Background

In an eagerly anticipated announcement, this week, Microsoft announced new Surface Pro and Surface Laptop models based on the Qualcomm Snapdragon X processor. The Snapdragon X Elite and Plus are Systems-on-Chip designs based on Arm cores and with an NPU (neural processing unit) capable of delivering 45 TOPS.

At a launch event on 20^th May, Microsoft set the minimum requirements for Copilot+ systems at 16GB of system memory, 256GB of SSD capacity and 40+ TOPS of NPU performance. The new Snapdragon X systems are currently the only hardware solutions capable of meeting these requirements, notably including the recently announced AI PC processors from Intel.

AI PC

The AI PC has been a big marketing push for Intel, starting with the release of Meteor Lake processors in December 2023. We previewed the technology in a blog post in September 2023. Meteor Lake is a SoC design incorporating a Neural Processing Unit (NPU), which manages the heavy lifting maths calculations needed for AI inferencing. It also incorporates both performance and efficiency cores as a technique to lower power draw and improve the battery life of portable devices.

From Microsoft’s perspective, new hardware capabilities enable software enhancements, specifically Copilot. Microsoft is betting big on the productivity improvements that Copilot can bring, which also attracts an additional monthly service charge.

TOPS

Unfortunately for Intel, the first wave of AI PC hardware announced in December doesn’t meet the minimum specification for running Copilot+. So far, only the Snapdragon X Arm-based processors meet the minimum requirement of 40+ TOPS. It’s worth noting that the Meteor Lake processors deliver 34 TOPS, while Lunar Lake, due later this year, should achieve 100 TOPS (Intel published a press release after the Qualcomm-related announcements to reiterate this).

Windows on Arm

Microsoft has developed a version of Windows specifically for Arm-based processors. Arm-native applications are starting to appear (including Google Chrome, Teams, Office applications, Zoom and WhatsApp). In the interim, Microsoft has developed a cross-ISA emulator called Prism that will translate x86-native applications and enable them to run on Windows for Arm. The details of Prism haven’t been made clear. However, we are searching for additional information (we believe Prism is based on Arm64EC).

The Architect’s View®

We can see several scenarios developing, as Windows 11 and the latest Snapdragon-based devices start to arrive in the market.

The Arm Scenario. Prism works as described and isn’t encumbered by Microsoft and partners attempting to gain additional revenue by licensing the use of Prism for existing applications. Battery lifetime and application performance exceed that experienced with Intel devices, driving adoption.
The Intel Scenario. Microsoft and partners look to monetise the migration to Arm, licensing the use of Prism for existing applications and hindering the adoption of Arm-based devices. Intel releases Lunar Lake processors in time for the Copilot+ release of Windows 11, matching or exceeding the performance of the current wave of Arm devices. Arm adoption is poor and hinders future hardware development.
The Balanced Scenario. Neither Intel nor Arm gains substantially with the first wave of AI devices. Both Intel and Arm products exist in the market, with Intel losing some market share.

It isn’t clear exactly how the market will move forward. We suspect that the factors affecting market direction will include:

Price – Arm-based systems need to be comparable or cheaper than their Intel equivalents.
Performance – Arm-based devices need to be at least equivalent in performance but with substantially better battery life. Alternatively, customers may trade some battery life for better performance than offered by Intel.
Productivity – Copilot delivers tangible productivity improvements, making the move to Arm a decision about more than just battery lifetime.
Reliability – Prism needs to make the Arm transition seamless. Any issues with application support will put uncertainty into customers’ minds and introduce inertia to move platforms, especially for business users.

What is the benefit for Microsoft in adopting Arm? Obviously, in corporate environments, more efficient designs mean lower operating costs for the customer. In reality, though, Microsoft wants to drive revenue growth from additional services, in this case, the monthly Copilot charges. As a result, the company will go with whichever vendor can deliver the hardware to meet the minimum Copilot(+) specifications – which is currently Qualcomm with Snapdragon.

How does all this affect the enterprise data centre? We already have Arm competing against Intel in the public cloud. AWS, Azure, and GCP all have Arm-based virtual instances in production or preview (Microsoft announced Cobalt-based virtual instances in preview at Build this week).

The most apparent result of the widespread adoption of Arm on consumer devices is that the process normalises the acceptance of Arm as a genuine competitor to x86 from a cost, performance, and efficiency perspective. This is a scenario we first discussed nearly four years ago. The only area where Arm hasn’t yet gained a foothold is in the private data centre on server systems. It is arguably only a matter of time before a processor vendor such as Ampere partners with a server vendor to produce a viable set of products.

Intel needs an Arm killer and needs it quickly. Otherwise, x86 will lose the dominance it currently has in the consumer and enterprise landscape. However, for the end user, the benefits are only positive, as the increased competition in the processor market should drive down pricing and improve overall operational costs.