The PC industry is hurtling toward a future where every new laptop and desktop sold will carry a neural processing unit (NPU) and run AI workloads locally. That's the audacious forecast from Luca Rossi, Lenovo's President of Intelligent Devices Group, who declared at IFA 2025 in Berlin that “every PC will be an AI PC in four, maximum five years.” Rossi's timeline, backed by current sales momentum and silicon roadmaps, paints a picture of a market in rapid transition—but it also glosses over stubborn gaps in software maturity, user awareness, and the very definition of what makes a PC “AI.”

Right now, the global penetration of AI PCs sits at about 5%, according to Lenovo's internal estimates. The company claims roughly a 30% share of that nascent segment and expects to leap to 50% within 12 to 18 months. That's aggressive growth, but credible given the OEM's scale, its tight partnerships with Intel, AMD, and Qualcomm, and a massive tailwind: Microsoft will end support for Windows 10 on October 14, 2025, forcing a wave of enterprise and consumer upgrades over the next two years.

The hardware is ready, but where are the ‘killer apps’?

The AI PC revolution has an engine problem that has nothing to do with silicon. “I think there is a relatively modest percentage of users that fully understand and fully embrace their AI,” Rossi admitted during a roundtable discussion. “I would say it's not a large majority of users.” Instead, the current boom is fueled by hardware appeal: Copilot+ laptops like Lenovo's ThinkPad X9 or the ASUS Zenbook A14 are celebrated for their thin profiles, 12-hour battery life, and premium build quality. The AI label is a secondary feature for many buyers.

Rossi's analogy is apt: the app ecosystem for NPUs today is like the early days of smartphone app stores. There is plenty of potential, but no single application has emerged that makes a local neural engine indispensable for the masses. Microsoft's own Copilot+ features—like live captions, Windows Studio effects, and Cocreator in Paint—are neat but have not set the mainstream on fire. My colleague Richard Devine recently noted that after a year of Copilot+, he still doesn't care about it.

The NPU inside: what TOPS really means

Underpinning the AI PC pitch is a new class of silicon. Microsoft's Copilot+ specification mandates an NPU capable of at least 40 trillion operations per second (TOPS). Qualcomm's Snapdragon X series started it, and Intel (Core Ultra) and AMD (Ryzen AI) have quickly followed with their own TOPS-compliant processors. But TOPS is a marketing-friendly metric, not a user-experience benchmark. Real performance depends on memory bandwidth, model optimization, software frameworks, and thermal constraints. Industry analysts caution against treating TOPS as the sole indicator of AI prowess.

In practice, the PC market is bifurcating between two NPU tiers: entry-level accelerators around 10 TOPS that handle background noise suppression, transcription, and camera effects, and the 40+ TOPS Copilot+ class designed for heavier on-device generative workloads. Yet even high-TOPS devices today struggle to run large language models locally beyond demo scenarios. Lenovo itself touts “AI Now,” a local personal knowledge base that can query documents, but these features remain niche.

The software gap: 80–100 ISVs are porting, but to what end?

Rossi revealed that Lenovo is in dialogue with 80 to 100 independent software vendors (ISVs) who are “now porting their applications to have the ability to use the NPU.” That's a tangible number, but “porting” is a vague term. Will these ISVs merely offload a video effect to the NPU, or will they rebuild workflows to take advantage of local inference for real productivity gains? Adobe, for instance, has shown AI-powered features like Generative Fill in Photoshop, but much of that heavy lifting still happens in the cloud.

The analogy to smartphone apps is instructive, but also cautionary. It took years after the iPhone's launch for apps like Uber and Instagram to redefine computing. AI PCs may follow a similar trajectory, but there's no guarantee the pace will satisfy investors or consumers. Rossi himself pegs 2024 as the year hardware laid the groundwork, 2025 as the year it matured, and the coming years as when software and “agent ecosystems” finally deliver. That's a long runway.

The super agent vision: intent-based computing

Lenovo's long-term bet extends beyond NPUs. Rossi envisions a “super agent” that blends local and cloud intelligence to anticipate needs, trigger tasks, and operate across devices—desktops, laptops, phones, wearables, even AR glasses. It's an intent-based computing model where users interact less with apps and more with an intelligent layer that orchestrates everything. “The machine or the agent will anticipate your needs and will trigger the application or the task you need autonomously over time,” Rossi said.

Parts of this future are already visible in China, where ecosystems like WeChat integrate services across devices. Lenovo, with its dual-ecosystem OEM role and chip partnerships, believes it can pioneer such cross-platform agents globally. But the reality is that operating systems—Windows and Android—remain siloed, and no universal agent standard exists. A “super agent” that works seamlessly across Microsoft, Google, and Apple ecosystems is a distant dream at best.

The Windows 10 EOL accelerator

For enterprises, the AI PC push has a more prosaic driver: fear. Microsoft's end-of-support date for Windows 10 will force security-conscious organizations to refresh hardware. Many will opt for Copilot+ PCs simply because they run Windows 11 and include the latest silicon. “Windows 10 end-of-life is a real procurement signal,” noted the forum analysis. Procurement templates are already shifting to include NPU requirements as a baseline, and certain verticals (healthcare, finance, legal) see real value in on-device AI for privacy and low-latency transcription. But blanket adoption across all industries will depend on proven ROI, not just compliance deadlines.

Strengths of Lenovo’s vision

  • Product depth: Lenovo's IFA lineup—from Copilot+ ThinkPads to folding phones, AR glasses, and gaming handhelds—shows a company executing on AI hardware that maps to multiple device categories.
  • Channel scale: With a top-three global PC share and a claimed 30% of AI PC shipments, Lenovo can push inventory and create momentum that others will follow.
  • Ecosystem leverage: Partnerships with Microsoft, Intel, AMD, and Qualcomm, combined with homegrown software experiments, give Lenovo a seat at every table where AI PC standards will be set.

Risks that could derail the timeline

  • Software never catches up: If ISVs fail to deliver apps that demonstrably improve daily life, AI PCs will remain premium hardware purchases driven only by conventional specs.
  • TOPS obsession: Buyers mistaking TOPS for real-world capability could lead to disappointment and backlash, slowing trust in the category.
  • Privacy and security: Agent-based computing requires deep access to personal data and app contexts. Without ironclad, transparent consent frameworks, enterprise adoption will stall.
  • Fragmentation: If every OEM and cloud provider builds proprietary agents, the “super agent” will dissolve into a Tower of Babel, undermining universal adoption.
  • Price: Most AI PCs today command a premium. For the prediction to hold, NPU-equipped devices must reach mainstream price bands below $700, which hasn't happened yet.

What to watch: signs the timeline is on track

  1. ISV porting velocity: Announcements from major suites (Microsoft 365, Adobe Creative Cloud, etc.) of full-feature local AI inference that significantly outperforms cloud alternatives.
  2. Benchmarks beyond TOPS: Independent, real-world metrics for AI tasks—transcription latency, image generation speed, battery impact—that give buyers clarity.
  3. Enterprise RFP changes: Procurement documents that explicitly mandate Copilot+ capabilities or minimum NPU TOPS, moving from pilot to standard.
  4. Agent standards: Public APIs or alliances that enable cross-platform agent behavior, such as an assistant that works across Outlook, Slack, and a phone.
  5. Price compression: Availability of 40+ TOPS Copilot+ laptops under $700, making AI PC the default choice for budget buyers.

What this means for you

  • Consumers: If your current laptop is fine, you can hold off. But if you're due for an upgrade and want to be ready for Copilot+ features as they mature, look for devices with explicit NPU specs and Windows 11 Copilot+ branding. Don't pay extra just for TOPS; focus on build quality, battery life, and display—the things you actually use.
  • IT decision makers: Use the Windows 10 EOL as a risk-management trigger, but pilot on-device AI features against real workflows before committing. Not every user will need an NPU on day one, but future-proofing your fleet now may save costs later as software matures.
  • Developers and ISVs: Target the most portable inference stacks (ONNX, DirectML, OpenVINO) and optimize for memory usage. Integration with OS-level assistant APIs will be key to discoverability when agent layers materialize.

The bottom line

Lenovo's five-year prophecy is a plausible, well-funded bet. NPUs are becoming standard, Windows 10's demise forces a refresh, and the sheer weight of OEM marketing will move units. But the leap from hardware ubiquity to indispensable AI experiences depends entirely on software that doesn't yet exist. Rossi's analogy to the smartphone app revolution is fair, but revolutions don't run on industry roadmaps—they run on unpredictable killer apps that change behavior. The next 18 months will reveal whether AI PCs are the next smartphone platform or just the next 3D TV.