Microsoft Architects a New Breed of PCs

May 22, 2024
Microsoft Architects a New Breed of PCs

Every year, personal computer (PC) manufacturers come out with the new versions of their laptops.

It’s never much of an event. The improvements each year have always been incremental.

More RAM (short-term memory), longer battery life, a faster processor, and a nicer screen always tend to be the selling points… for what has largely become a commodity product. This is especially true for Microsoft Windows-based machines.

But not this year.

Microsoft’s Big Move

Microsoft has rearchitected the PC, in what I consider to be the biggest development in the world of Windows since Windows 95.

It has been a loooooong wait for something really interesting.

Microsoft announced its new line of Surface Pro tablets and Surface Laptops branded with Copilot + PC:

Source: Microsoft
Source: Microsoft

And yes, the announcement was all about artificial intelligence (AI) and how it has been incorporated into Microsoft’s PC hardware design.

It was too obvious that the timing of the announcement was synchronized with the latest developments at OpenAI. 

And as we learned last week in Outer Limits — Is Google Doomed?, OpenAI’s latest developments did not include a new release of its large language model (LLM), expected to be called GPT-5.

Instead, OpenAI released GPT-4o, a multi-modal upgrade to GPT-4 capable of accepting inputs of voice, video (including real-time video), images, data, software code, audio, with the ability to generate text, audio, or image outputs.

Given Microsoft’s effective control over OpenAI due to its majority ownership stake having invested around $13 billion, the latest developments really put the product decisions into context.

The reason is that a multi-modal version of GPT-4 (like GPT-4o) is far more valuable to Microsoft from a product perspective than a larger LLM would have been.

Said another way, GPT-4o is incredibly valuable as a human-like interface to a computing system. 

In the issue of Outer Limits — Is Google Doomed?, I shared some examples of how GPT-4o can be used:

  • Real-time translation provided between two people that speak different languages. The users give the AI the instruction to translate one person’s language from English to Spanish, and the other from Spanish to English.
  • Recognizing the context of an event due to props on a table — in this case a birthday.
  • Acting as a tour guide while in London — the camera provides the “sight” for the AI to understand location and context.
  • Customer service applications
  • Two AIs assigned with the tasks of how to interact with one another.
  • Assistance in preparing for an interview (even judging the interviewees outfit).
  • Demonstrating a very human trait, like sarcasm.
  • Math instruction for educational purposes with a child.

It’s worth seeing some of this in action, and you can do so here on OpenAI’s blog post.

This is the kind of functional technology that can be productized by Microsoft. And that’s exactly what was announced early this week.

Microsoft’s AI product, Copilot, is built on GPT-4o.

It is now clear that Microsoft has had a heavy influence over the product development at OpenAI, as it is critical to its own product announcements.

And yet, Microsoft’s announcement was actually much bigger.

The New PC Architecture

Microsoft’s announcement demonstrated a hard shift in PC architecture, specifically for AI.

The media focused on the Microsoft Surface products powered by AI — Copilot — as the big news. But that wasn’t it at all.

Microsoft detailed specifications required for any Windows-based PC to be considered a Copilot Plus PC. Minimum requirements are:

  • 256 GB of SSD (solid state drive storage)
  • Integrated neural processor (Microsoft calls this the NPU — neural processing unit)
  • 16 GB of RAM

These specifications are designed to one-up Apple and its MacBook Air line of laptops.

It’s probably no surprise how similar the Copilot PCs look to Apple’s design, seen below.

Apple Macbooks
Source: Apple

But the real reason for the hardware horsepower, specifically the neural processing unit (NPU), is to support running AI software locally on the laptop (as opposed to in the could).

This is the PC architecture shift that I referred to earlier. Apple has already been designing AI hardware into its laptops and tablets, and now Microsoft has taken the leap.

And the key point for us to understand is that this announcement isn’t about Microsoft trying to increase its own branded PC sales… 

After all, it has less than 5% of the global PC market.

Computer hardware really isn’t Microsoft’s business…

The New Reference Design

Microsoft, with its Windows operating system (OS), has adopted a strategy very similar to Google and its Android OS for smartphones.

Google’s branded smartphones have less than 3% market share in the U.S. and even less worldwide. The goal has never been to become a major smartphone manufacturer.

Both Google, and now Microsoft, are building their hardware in a way that acts as a reference design to show off what the software — in this case Microsoft’s new Copilot (GPT-4o) — can do.  It’s like setting a standard for Windows-based machines capable of running Copilot.

It has already been confirmed that all major Windows laptop partners will be manufacturing Copilot Plus PCs. That means Dell, Lenovo, Hewlett Packard, Asus, Acer, and Samsung. More will follow.

As we discussed in Outer Limits — Is Google Doomed?, this is the same strategy that Google uses in its own smartphone reference designs, used to highlight its Android operating system.

Microsoft’s current forecast is for 50 million Copilot Plus branded PCs to be sold in the next 12 months.

But to me, this is an accelerant.

In general, a Windows-based average lifespan is between 3-5 years. And that’s probably generous. Usually by three years out, the machines get so slow because the Windows software requires more and more horsepower with each new version. Consumers end up upgrading just to get their PC performance back to some kind of normal.

And in this case, to run something like Copilot (GPT-4o) locally, new hardware is required. That’s the neural processing unit (NPU) — an application-specific semiconductor designed for running AI. Microsoft used an ARM-based semiconductor from Qualcomm for its own Surface products, but it is possible to use other ARM-based NPUs from other companies.

ARM based semiconductors are the most widely adopted semiconductor architectures for mobile applications where power consumption is important. They are powerful, efficient semiconductor designs that are licensed from ARM Holdings by almost all semiconductor companies.

The newfound functionality enabled by AI is an actual incentive for consumers to upgrade their laptops so that they can access the new features enabled by Copilot (GPT-4o). It’s kind of like wanting to upgrade to a 5G phone for a big jump in performance compared to 4G. But in this case, even bigger.

And the most notable AI-powered feature of Copilot is Recall.

“Watching” Everything

Recall may make some subscribers uncomfortable.

The software records everything we do on a Windows Copilot Plus PC. Everything.

It logs what websites we visit, what applications we use, all of our photos and videos, communications in video meetings, basically any activity that we engage in with the laptop.

Do we trust Microsoft with that information? I’ll leave that for you to decide.

The purpose of this technology, from the user’s perspective, is to be able to recall just about anything that we might be searching for.

Microsoft Copilot Recall
Source: Microsoft

All of our interactions with the PC become searchable.

If we can’t remember where we saw that picture of a red barn, Copilot can find it from our history. If we’re looking for some obscure reference to an inventor like Benjamin Franklin that we referenced on some call years ago, Copilot can give us the time, date, and meeting when it came up.

Copilot basically gives us a human interface for our computing system, that will save us wasted time searching for things that we need to dig up. 

The utility is obvious. It will make life easier. It will reduce or eliminate friction in working with a computing system.

And for that reason, it will be successful.

Microsoft has about a 72% global market share in desktop and laptop computer operating systems. And combined with about 15% for Apple, that’s almost 90% of the global market share for operating systems.

With both players jumping headfirst into AI-enabled PC hardware architectures, it won’t be long before the world will be interfacing with their computers in a completely different way.

We always welcome your feedback. We read every email and address the most common comments and questions in the Friday AMA. Please write to us here.

Previous Post Next Post