In today's rapidly evolving tech landscape, the **interest in AI PCs is surging**, breathlessly reinvigorating what has long been a plethora of stagnant endpoint devices. Yet, amid this excitement, there remains a challenge: the **absence of clearly defined use cases** that resonate with everyday users.
Initially, hardware and software vendors championed use cases focused on **unified communications and collaboration**. While these advancements demonstrate the remarkable capabilities of AI technology in enhancing audio and video quality, they failed to address significant needs for the majority of users. These features are impressive, yet **not enough to make anyone say "I need this right now!"**
As industry discussions continue, we find ourselves in a state of confusion—will a universal **AI PC** use case ever emerge? Drawing upon my insights from early 2023, I speculated whether AI would quietly integrate itself into our daily operations or whether a compelling reason for widespread adoption would ultimately surface. Currently, I find myself straddling both perspectives: the undeniable utility of AI is clear, but I'm yearning for a **truly compelling use case** to justify broader acceptance.
While searching for **broad, must-have use cases for local AI resources**, many of us are left wanting. Although a few promising applications are creeping into view—such as security enhancements and agentic AI—most prominent AI implementations still depend heavily on cloud-based services. These services are undeniably beneficial, prevalent, and have been delivering substantial advantages across various sectors. Still, they raise the question: if cloud services suffice, why do we genuinely need local AI?
In recent months, **some use cases for local AI** have emerged, albeit with mixed feedback from users. For instance, Copilot+ launched with Recall, which garnered responses ranging from "We can do this, but should we?" to "Please, no!" Additionally, the ability to build and disseminate customized models using open-source large language models (LLMs) is gaining traction, particularly among developers, but applicable use cases may extend beyond them.
However, there are significant challenges for those looking to train their models:
It's **costly**.
Keeping up with rapid advancements in cloud-scale LLMs is **challenging**.
Models can quickly become **outdated**.
Frequent **retraining** is often necessary, creating a cycle of continuous investment.
So, where does that leave us? The sensation of a “**solution in search of a problem**” looms large. This sentiment may sound harsh, but after **two months of using an AI PC in my daily work**, it became apparent that my neural processing unit (NPU) was predominantly engaged only during Teams meetings.
Yet, all is not bleak. In fact, an exciting opportunity is emerging: **security**. This might just be the **game-changing use case** that we've all been searching for. **Security** applications could solidify the practical value of AI PCs as other use cases gradually evolve alongside their adoption.
Security and Agentic AI: The Future of AI PCs
Before diving deeper, it's essential to clarify what constitutes an **AI PC**. I recently came across an insightful definition from an Intel representative: an AI PC is equipped with dedicated hardware optimized for specific tasks. The **CPU handles quick, lightweight operations**, the **GPU is geared for data-intensive AI processes**, and the **NPU serves as an "AI accelerator,"** enabling efficient workloads to run consistently while saving energy.
An AI PC is not just a powerful GPU. Rather, it balances the strengths of CPUs, GPUs, and NPUs to deliver the best performance for various workloads.
Key Use Cases: Security
Starting from the foundational level of AI utilization—think audio and video enhancements—**endpoint security** stands tall as the next significant application of local AI capabilities. I eagerly anticipate discovering more about **AI-driven endpoint security** during the upcoming RSA Conference.
My experience last year was underwhelming, as AI’s involvement in endpoint security could be summarized with the mundane label of “chatbots.” However, this year presents newly emerging applications, such as **ESET's announcement** to leverage Intel NPUs for increased efficiency and reduced strain on system resources—a trend likely to expand.
Exploring Agentic AI
Diving further into the AI capabilities, **agentic AI** emerges as the next exciting frontier. While its potential impact seems boundless, the path toward practical implementation involves navigating numerous complexities. Concerns surrounding **security, identity, compliance, and trust** must first be addressed, particularly if agents are to function autonomously on behalf of organizations.
Nevertheless, **the endpoint could serve as the ideal battleground for agentic AI**, enabling agents that operate specifically for end users—handling tasks like filing expenses, compiling reports, or strategizing based on meeting notes. Offloading these mundane responsibilities to local AI could alleviate burdens on cloud resources, allowing them to focus on more demanding assignments.
Conclusion
As we await **AI's defining "Excel moment,"** akin to how Microsoft Excel revolutionized corporate forecasting, it's refreshing to witness **practical use cases** gradually coming to light. I recently learned about **ESET’s innovative use of AI PCs** to enhance their endpoint security offerings, which highlights the exciting potential of AI's capabilities. After the RSA Conference, I hope to share even more remarkable applications that showcase the value of AI PCs.
Gabe Knuth is the senior end-user computing analyst at Enterprise Strategy Group, now part of Omdia.
Enterprise Strategy Group is part of Omdia, and its analysts maintain business relationships with various tech vendors.
