The PC upgrade cycle that vendors promised would rocket with the arrival of “AI PCs” has instead become a slow, uneven crawl — a market shaped less by a single, dramatic buyer pivot than by technical gates, enterprise risk calculations, rising component costs, and a murky, still‑forming value proposition for on‑device artificial intelligence. Recent vendor calls and analyst commentary show the picture clearly: AI‑capable devices are growing as a share of shipments, but meaningful, broad‑based upgrades are stalling as buyers ask whether the premium for NPUs, Copilot integrations, and Windows 11 readiness actually delivers day‑to‑day return on investment.
Background
The stage for the current market dynamic is familiar: Microsoft set a higher hardware baseline for Windows 11 (TPM 2.0, UEFI Secure Boot, and CPU generation cutoffs), then announced the end of mainstream support for Windows 10 — a calendar that intended to accelerate refreshes. But that hardware compatibility cliff left a substantial pool of functional devices unable to upgrade without replacement, and Microsoft’s Extended Security Updates (ESU) program provided a pragmatic bridge that many buyers used instead of replacing machines immediately. At the same time, OEMs shifted messaging toward an “AI PC” era — thin‑and‑light notebooks and workstations equipped with NPUs/MPUs and optimized stacks that promise low‑latency, private on‑device inference. Microsoft and OEMs framed those capabilities as the next compelling reason to refresh.
The result is a market that is uneven: enterprise refresh programs continue on traditional cycles, commercial purchases skew earlier and larger, and consumers — faced with modest perceived benefits, inflationary pressures, and tradeoffs like battery life — are more likely to defer. That “hurry up and wait” reality is reflected in vendor commentary and independent telemetry.
What vendors say today
HP: productivity claims, cost actions, and AI PC share
HP’s latest earnings commentary leaned hard into AI as both a product and an operational lever. CEO Enrique Lores told investors HP is using its own deployments as “customer zero” to demonstrate gains, claiming double‑digit productivity improvements from curated AI workflows and reporting that AI‑capable devices represented north of 30% of shipments in the quarter. HP also announced a workforce reduction tied to a plan to realize roughly $1 billion of gross run‑rate savings over several years while accelerating AI adoption internally. Those are bold claims: HP framed AI PCs as a rising revenue driver, but paired that narrative with clear caution about cost and supply headwinds. Key vendor points from HP:
- AI PCs exceeded 30% of shipments in the quarter, and management expects the mix to grow.
- HP says internal AI deployments have shown up to 16–17% productivity improvements in selected workflows; management uses these pilots to support its productivity narrative.
- HP is cutting 4,000–6,000 roles as part of a program to reallocate resources toward AI and efficiency, targeting $1 billion in gross savings.
Caveat: vendor‑reported productivity gains are selectable and context‑dependent — pilots and curated application sets rarely generalize without careful rollout, verification, and third‑party measurement.
Dell: installed‑base math, server growth, and cautious optimism
Dell’s investor messaging focused on two realities: the company sees a large installed base that hasn’t migrated, and its fastest growth right now is in servers and AI infrastructure rather than mainstream client devices. Dell’s COO Jeffrey Clarke publicly estimated roughly 1.5 billion Windows PCs in the field, with about 500 million capable of upgrading to Windows 11 but not yet upgraded, and another ~500 million that are too old to support Windows 11 without replacement. Dell pairs that installed‑base math with a view that NPUs, small language models, and Copilot‑style experiences will create a multi‑year runway for premium, higher‑ASP replacements — but Dell simultaneously forecasts a roughly flat PC market in the near term, relying on server growth to power corporate results. What Dell’s statements imply:
- There is an addressable upgrade pool — but converting it requires more than vendor claims; it needs clear ROI, financing, and timing aligned with procurement cycles.
- Dell expects server and infrastructure strength (AI datacenter demand) to sustain the company while client refreshes creep forward.
Caveat: Dell’s installed‑base estimates are directional planning math for investors; they are not a device‑level census and should be reconciled against telemetry and fleet inventories.
Lenovo: market share in AI PCs, and an aggressive product push
Lenovo has leaned into a “personal AI” strategy and reports strong adoption of AI‑capable SKUs. Management pointed to AI PC penetration in the low‑to‑mid 30% range of shipments and asserted leadership in the Windows AI PC category, claiming more than 30% market share in that segment. Lenovo frames AI PCs as part of a broader “One AI, Multiple Devices” ecosystem, betting that cross‑device agent experiences will cement its advantage. Vendor takeaway:
- Lenovo’s product and supply discipline has translated into measurable share gains in the premium, AI‑focused segment.
The technical and economic frictions slowing adoption
1) The compatibility cliff
Windows 11’s hardware baseline deliberately raised the bar for security and future capabilities. But the immediate side effect was a compatibility cliff: many still‑serviceable PCs lack TPM 2.0, compatible firmware, or a supported CPU generation and therefore cannot be upgraded without motherboard or full system replacement. For users and IT teams that means an investment decision rather than a simple in‑place OS update. This hardware gating is one of the most concrete reasons adoption has been slower.
2) Enterprise risk calculus and migration economics
Large organizations do not flip operating systems overnight. They must validate line‑of‑business applications, test drivers and images, and coordinate staged rollouts — often on fiscal timelines and procurement windows that span quarters or years. Even where hardware is capable, migration timing is a function of budget, compliance, and change management capacity. ESU offerings let many enterprises and consumers buy time rather than accelerate capital spending.
3) Component inflation and margin pressure
DRAM and NAND pricing cycles, compounded by high AI datacenter demand, lifted BOM costs across the industry in 2024–2025. Higher component costs compress OEM pricing flexibility and make aggressive consumer discounts harder to sustain. That raises the effective cost of replacement and tilts buying behavior toward deferral or refurbished devices.
4) Perceived value of on‑device AI
Most consumer‑facing AI features today are accessible through cloud APIs and browser tools; they don’t strictly require specialized client hardware. On‑device inference brings advantages — privacy, low latency, offline capability — but the software ecosystem must deliver compelling, measurable gains for users to justify paying a premium. While vendors promise productivity lift and privacy benefits, widespread, repeatable “killer apps” for on‑device AI are still emerging.
5) Sustainability and e‑waste
Pushing millions of still‑functional PCs into replacement cycles raises environmental and equity concerns. Advocacy groups and procurement officers are increasingly factoring refurbishment, trade‑in, and lifecycle clauses into buying decisions, which can prolong device lifetimes and delay net new sales.
What “AI PC” actually means (and why it matters)
The term “AI PC” is often used as a marketing umbrella, but it encompasses a set of concrete hardware and software attributes:
- Dedicated NPUs/MPUs for efficient on‑device inference.
- Firmware and OS integration (Windows hooks for Copilot and platform services).
- Larger memory footprints and faster NVMe storage to host models locally.
- Software delivery mechanisms for on‑device models and privacy controls.
AI PCs can deliver real, measurable advantages for specific workflows: local transcription with near‑zero latency, privacy‑sensitive personal assistants, offline generative features, and lower recurrent cloud costs for high‑volume inference. But these benefits require software vendors and IT teams to adopt those local APIs; silicon alone is not enough. Buyers should demand measured benchmarks (latency, throughput, battery impact) and validated use cases before paying a premium.
Enterprise vs. consumer behavior: two very different upgrade engines
- Enterprise:
- Migration decisions are driven by security, compliance, and total cost of ownership.
- ESU is used tactically to buy time, not as a long‑term strategy.
- Early AI PC adopters are often sectors with high privacy or latency needs (finance, healthcare, legal) or knowledge‑worker groups with quantifiable productivity outcomes.
- Consumer:
- Decisions are price, battery life, and feature driven; visual tweaks to the OS are rarely a lone trigger.
- Many consumers will use cloud AI (ChatGPT, Gemini) or continue on older devices until hardware fails or a clear, widely useful capability emerges.
- Sustainability and affordability make refurbished and trade‑in channels important levers.
Practical guidance — what IT leaders and consumers should do now
- Inventory and classify devices immediately: tag endpoints as Windows 11‑ready, firmware‑upgradeable, or replacement‑required.
- Prioritize risk exposure: upgrade internet‑facing and regulated endpoints first; low‑risk users can be staged later.
- Use ESU strategically: treat Extended Security Updates as a time‑limited bridge, not a permanent solution. Microsoft’s consumer ESU program runs through October 13, 2026 and requires enrollment mechanics that vary by account type.
- Pilot AI PC value: run targeted pilots with clear KPIs (transcription time saved, task automation hours, reduced cloud spend) before wide procurement.
- Negotiate procurement levers: bulk trade‑ins, certified refurbished channels, and financing reduce per‑seat costs.
- Standardize validated SKUs: choose a small set of tested configurations per workload to simplify driver management.
This staged, evidence‑based approach minimizes disruption and aligns IT spend with measurable outcomes rather than marketing narratives.
Strategic risks and what could go wrong
- Vendor claims vs. real world: Productivity gains reported inside vendor pilots (HP’s reported 16–17% uplift, for example) are promising but contextual; they require independent validation and scalable software integration to translate into fleet‑level ROI.
- Price sensitivity: If component inflation persists, OEMs may be forced to maintain higher ASPs on premium AI SKUs, widening the affordability gap for mainstream buyers.
- Environmental backlash: Aggressive replacement campaigns without robust refurbishment programs risk reputational and regulatory consequences.
- Fragmented security posture: A prolonged mixed fleet (Windows 10 + Windows 11) increases operational complexity and potential exposure unless mitigations (segmentation, endpoint protection, ESU) are enforced.
Vendor playbooks: how OEMs will try to convert opportunity into revenue
OEMs have a clear playbook:
- Target enterprise refresh programs where procurement windows and security needs align with replacements.
- Upsell AI capabilities and Copilot+ bundles to push ASPs higher on premium SKUs.
- Offer trade‑in, financing, and certified refurbished options to lower consumer friction.
- Use internal deployments (customer zero) to generate case studies and measurable KPIs for sales motions.
This strategy is rational — higher ASPs offset flat unit growth — but converting the mid‑market requires proving day‑to‑day value and lowering cost and migration friction.
The near‑term outlook (what to expect in 12–24 months)
- A continued, multi‑year refresh runway rather than a single “upgrade wave.” Market movement will be segmented by enterprise, region, and vertical.
- AI PCs will gain share in shipments (vendors report ~30–33% penetration in recent quarters), but mainstream consumer replacement will be gradual unless clear, ubiquitous use cases emerge.
- OEMs will extract margin through premium AI SKUs while server and infrastructure sales continue to be the fastest growth lever for companies like Dell.
Two plausible scenarios to watch:
- Accelerated uplift: measurable, widely useful on‑device features (local agents integrated into browsers, productivity suites, and line‑of‑business apps) plus vendor trade‑in incentives ignite stronger consumer take‑up within 12–24 months.
- Stretched migration: enterprise prudence, affordability concerns, and sustainability priorities slow adoption into a 2–5 year taper where OEMs monetize services and premium segments while overall unit volumes stay flat.
Final assessment
The AI PC narrative is
real — vendors are shipping devices with NPUs, and OEMs report growing shares of AI‑capable SKUs. But the upgrade boom many predicted has not materialized into a singular, large‑scale consumer rush. Instead, the market is rewriting itself into a longer, more complex refresh cycle driven by hardware eligibility, enterprise timing, component prices, and the still‑maturing software ecosystem for on‑device AI. HP and Lenovo report meaningful AI PC penetration and internal productivity claims from pilots, while Dell’s installed‑base math highlights a large deferred upgrade opportunity that will be realized only on buyers’ terms. That combination — product readiness, pilot success, and buyer caution — yields a simple conclusion: upgrade timing will be deliberate, not frenetic.
For buyers and IT leaders the right posture is pragmatic: inventory, pilot, measure, and align procurement to validated value. For OEMs the pressure is to convert promising AI slogans into measurable, repeatable outcomes and affordable trade‑in paths that close the gap between marketing and day‑to‑day work. The revolution hasn’t been canceled; it’s simply on a runway longer than many expected.
Conclusion
The PC market is not failing; it is evolving into segmented, higher‑value lanes where AI‑capable hardware will matter most in contexts that require privacy, low latency, or sustained local inference. Vendors can and should keep pushing the software ecosystems that make NPUs matter — but the ultimate arbiter remains clear: buyers will pay for what measurably improves outcomes, protects data, or reduces total cost of ownership. Until those gains are broadly demonstrated and affordable, the AI PC upgrade cycle will progress — but it will do so on buyers’ terms, not vendors’ calendars.
Source: Constellation Research
The AI PC upgrade cycle is crawling amid murky value