The AI PC needs to deliver more than performance - it needs to deliver security

Scanning the future

Sponsored Feature The AI revolution is changing how we do business, how we innovate and how we relax.

And like all good revolutions it is causing other revolutions up and down the tech stack, not least as hardware manufacturers race to provide AI capable platforms.

Almost a third of personal computers shipped in 2025 will be AI PCs, according to figures from Gartner, double the proportion just a year ago. A few years ago, skeptics saw the AI PC label as a prime example of hype. Now, says Gartner, it will be the norm by 2029.

But this breakneck pace of development poses dangers, if manufacturers and enterprises rush to support newer AI workloads, while neglecting the disciplines of the past. That's particularly true when it comes to security.

As truly AI capable devices find their way onto the market, things that would have happened in the cloud, now happen in your lap. That means benefits such as increased productivity and autonomy and reduced latency. But it also means much more AI-related activity is happening outside the embrace of hyperscalers' security wrappers.

The same threats and vulnerabilities associated with traditional on-device computing apply, with AI potentially amplifying some of these, while adding others.

For example, the data being analyzed or used to train a model presents a tempting target for attackers, whether to exfiltrate or encrypt or both. The model itself, perhaps one tuned to the needs of your organization, is also a target. Attackers could seek to manipulate models or extract underlying data from them. And, of course, attackers themselves have been quick to adopt and adapt AI, from improving phishing lures, to automating their operations, refining malware, and carrying out more targeted reconnaissance.

Then the demands of AI have to be considered. Security scans have always been a bugbear for enterprise PC users, slowing down operations and diverting system resources from other applications.

This mismatch becomes more pronounced when AI workloads come into the equation, given they are likely to be more resource intensive and data heavy than traditional enterprise workloads.

Add to this the question of energy, and it's clear that the requirements for a true AI PC are more than just a blisteringly fast GPU, NPU or even CPU performance. Because if the type of workload is not matched with the right xPU, it is sapping the platform's ability to carry out AI work, and the amount of time it can spend doing that.

These three issues (running AI on-device, security, and power/performance), are inseparable. And all three have informed Intel's second-gen Core Ultra processor architecture and the vendor's vision for what constitutes a true AI PC.

Protect and thrive

As Intel partner technical sales specialist Jimmy Wai explains, "What we are trying to do on the AI PC is focus on two things. One is protecting the hardware itself and also using the hardware to protect the data or the workload running on the PC."

This requires looking at both the overall performance the platform provides but also considering how that performance is managed and allocated. This means more than simply meeting the requirements of Microsoft's Copilot+ AI PC definition, which calls for a neural processing unit capable of 40TOPS or more in addition to GPU and CPU.

It also requires support for Pluton, Microsoft's on-board zero trust security processor, which adds security support beyond the Trusted Platform Module 2.0 specification, and which is part of the Copilot+ standard.

Intel's second-gen Core Ultra processor architecture brings all these parts together on a single die. That includes up to ten x86 cores, a GPU with up to eight Intel Arc Graphics Xe graphics cores, and the Intel Boost NPU.

The entire system delivers up to 120 TOPS, 48 of which come from the NPU, while the GPU delivers 67 TOPS, with the balance coming from the CPU.

In addition, it carries a Platform Controller tile, which handles I/O functions, but also carries built-in security functions including Intel Platform Trust Technology, and that critical Pluton capability.

But Pluton is table stakes when it comes to security in Copilot+ compatible PCs. Enterprise use stretches the architecture much further when it comes to AI and security.

The Intel platform's built-in Intel Threat Detection Technology (Intel TDT) delivers hardware-level monitoring, detecting how different workloads are using the CPU resources. "Based on those behaviors, we can detect whether there's a ransomware or crypto mining malware running on that PC," explains Wai. EDR solutions can use the technology to enhance malware detection and to provide better security protection to the PC..

Because threat protection focuses on behavior, Wai continued, "It's really difficult for the hackers to hide those attacks. Even if they are sophisticated enough to hide that workload or malware in a virtual machine, Intel TDT can still detect that, because eventually it has to be executed on the CPU.'"

Intel's vPro Enterprise delivers additional protection in firmware, such as safe boot and firmware protection. But it also offers hardware management that's independent of the operating system, with Intel Active Management Technology.

This allows enterprises to manage users' devices, even when the operating system is inaccessible. For example, says Wai, "In a case where you have a security breach, and you need to reach out to the PC and fix that, it becomes a very valuable tool."

Support when you're down

Remote manageability is also invaluable if security or infrastructure teams have to deal with an outage due to a flawed update and have to reach out to their fleet of PCs to apply a patch or delete files.

"Intel AMT gives you the way to do that without physically touching the PCs. So, if you have a security incident, you can use that too to quickly recover your PCs," he says.

But the GPU also has a role to play when it comes to security scanning.

Some algorithms just work better on GPUs, says Wai, while the graphics architecture can offer increased efficiency.

One key example is accelerated memory scanning, he continues. Many modern malware threats just sit in RAM, meaning they won't be spotted in an SSD scan.

But, Wai continues, "If you are an ISV and you want to do memory scanning it is very costly if you use a CPU to do that, it takes out a lot of performance right from the end user."

That focus on energy and performance also applies to Intel's decision to solder DRAM in place in second-gen Core Ultra processors.

But having memory on the die also has a dramatic impact on overall performance and power efficiency.

Again, this is part of the architecture's broader focus on energy management, from more granular clock speed management to a smaller transistors. Among other benefits this means advanced threat management and other security techniques can be applied without compromising overall performance or battery life.

So, users can expect all day performance, even when running intensive AI tasks, without compromising on overall security.

That might sound like a small thing compared to the some of the more eye catching possibilities that AI promises. But for today's enterprise users crunching numbers, collaborating with far flung colleagues, generating marketing collateral, or developing their own small language models, it will be truly revolutionary.

For more on Intel AI PC’s for business, click here.

Sponsored by Intel.