⚡ Quick Summary
- Asia's mid-tier and smaller chipmakers are implementing price hikes, following larger peers, as AI infrastructure demand creates supply constraints across the entire semiconductor ecosystem.
- Global AI-related capital expenditure is forecast to surpass $136 billion in 2026, a 25% year-on-year increase driven by hyperscaler commitments from Microsoft, Google, Amazon, and Meta.
- Advanced packaging substrates and power management components are seeing the steepest price increases of 10–20%, as bottlenecks shift beyond raw compute silicon.
- Enterprise cloud AI costs are expected to rise significantly over 2025–2026 as hardware inflation propagates through hyperscaler pricing, affecting Microsoft Azure and competing platforms.
- Businesses should urgently revisit AI infrastructure budget models, explore committed-use cloud pricing agreements, and optimise software licensing costs to offset rising hardware expenditure.
What Happened
A wave of price increases is rippling through Asia's semiconductor industry, and this time it isn't just the headline names driving the movement. According to reporting by Nikkei Asia, smaller and mid-tier chipmakers across the region are following their larger counterparts in raising prices, with the catalyst being an unprecedented surge in capital expenditure tied directly to artificial intelligence infrastructure build-out. Industry projections now place global AI-related capex at over $136 billion in 2026, representing a year-on-year increase of approximately 25%.
The companies raising prices span a broad spectrum of the semiconductor supply chain — from packaging specialists and substrate manufacturers in Taiwan, South Korea, and Japan, to analogue chip producers and power management IC suppliers operating in markets that rarely make international headlines. What unites them is a shared reality: their components, once treated as commodity inputs, are now bottlenecked by demand from hyperscalers, AI accelerator manufacturers, and server OEMs racing to build out data centre capacity at a pace the industry has never before sustained.
The price hikes vary by segment. Advanced packaging substrates — the critical interposers and high-bandwidth memory packaging components essential to chips like NVIDIA's H100 and Blackwell series — are seeing some of the steepest increases, with certain substrate categories reportedly rising 10–20% depending on lead time commitments. Power semiconductors, thermal management components, and specialised connectors are also climbing. Even companies supplying relatively mature-node silicon are benefiting from the overflow demand as leading-edge fabs remain constrained.
The timing is notable. This announcement arrives as TSMC continues to ramp its N3 and N2 process nodes, as Samsung Foundry battles yield challenges on its 3nm GAA process, and as Intel Foundry Services attempts to rebuild credibility with its 18A node roadmap. The entire ecosystem is under pressure to deliver, and cost increases at the component level will inevitably propagate upstream.
Background and Context
To understand why this moment feels structurally different from previous semiconductor cycles, it helps to trace the arc from the 2020–2022 chip shortage to today's AI-driven demand environment. The COVID-era shortage was characterised by broad, relatively indiscriminate demand across consumer electronics, automotive, and industrial sectors — a demand spike that ultimately corrected sharply in 2022 and 2023 as inventory overhang accumulated across the supply chain. Many smaller Asian chipmakers that had scrambled to expand capacity found themselves sitting on excess stock and idle equipment by mid-2023.
The AI inflection changed everything. The commercial launch of OpenAI's ChatGPT in November 2022, followed by the explosion of large language model deployments throughout 2023, triggered a qualitatively different kind of semiconductor demand — one concentrated in high-value, high-complexity components rather than distributed across commodity categories. NVIDIA's data centre revenue grew from approximately $15 billion in fiscal year 2023 to over $47 billion in fiscal year 2024, a trajectory that forced every link in the semiconductor supply chain to reassess its capacity assumptions.
Crucially, this demand is not cyclical in the traditional sense. The hyperscalers — Microsoft Azure, Google Cloud, Amazon Web Services, and Meta — have all publicly committed to sustained AI infrastructure investment. Microsoft alone announced plans to invest $80 billion in AI-capable data centres in fiscal year 2025, with the majority of that spend directed at US facilities but with significant portions flowing through Asian supply chains. Google's parent Alphabet disclosed capex guidance of approximately $75 billion for 2025. These are multi-year commitments, not quarter-by-quarter procurement decisions, which gives chipmakers and their suppliers unusual visibility into forward demand.
Smaller Asian chip companies — firms like Unimicron and Ibiden in substrate manufacturing, or companies like Chroma ATE in test equipment, or the dozens of Taiwanese analogue and mixed-signal IC houses — had largely been spectators to the initial AI boom. Their products were necessary but not glamorous. That dynamic has now shifted as the bottleneck has moved from raw compute silicon to the surrounding ecosystem of packaging, power delivery, and signal integrity components.
Why This Matters
For enterprise technology decision-makers, the implications of this price escalation extend well beyond the semiconductor industry itself. The cost increases being absorbed at the component level will inevitably translate into higher prices for AI servers, GPU clusters, and the cloud infrastructure that underpins modern enterprise software — including the AI-augmented productivity tools that millions of businesses depend on daily.
Consider the Microsoft ecosystem specifically. Microsoft's Copilot AI features, now deeply integrated into Microsoft 365, Windows 11, and the broader Azure platform, are computationally intensive services that run on data centre hardware. When the cost of building and operating that hardware rises — driven by more expensive chips, substrates, power components, and cooling systems — those costs don't disappear. They get absorbed into cloud service pricing, passed to enterprise customers through licensing adjustments, or reflected in the pace of feature rollouts. Microsoft has already raised Microsoft 365 commercial pricing multiple times since 2022, and the structural cost pressures now building in the semiconductor supply chain provide additional justification for future increases.
For IT professionals managing enterprise technology budgets, this creates a planning challenge. Cloud compute costs for AI workloads were already difficult to forecast; they become harder still when the underlying hardware supply chain is in a sustained inflationary phase. Organisations running AI inference workloads on Azure, AWS, or Google Cloud should be modelling scenarios where per-token or per-hour compute costs increase 15–30% over the next 24 months, not as a worst case, but as a plausible baseline.
There is also a strategic implication for businesses evaluating whether to run AI workloads on-premises versus in the cloud. On-premises AI server pricing — for systems from Dell, HPE, Lenovo, and others built around NVIDIA's HGX platform — will also rise as component costs increase. The calculus between cloud and on-prem AI infrastructure is about to get more complex, and procurement teams that lock in multi-year cloud commitments now may find themselves either advantaged or disadvantaged depending on how aggressively hyperscalers pass through cost increases.
Businesses looking to manage costs in this environment should explore every avenue for efficiency, including sourcing affordable Microsoft Office licences through legitimate resellers rather than paying full retail prices, as software cost optimisation becomes increasingly important when hardware and cloud costs are rising.
Industry Impact and Competitive Landscape
The competitive dynamics triggered by this price escalation are multidimensional and will play out differently across the major technology ecosystems.
NVIDIA remains the primary beneficiary of the AI capex surge in the short term. Its H100 and H200 GPUs continue to command extraordinary premiums, and the forthcoming Blackwell B200 and GB200 NVLink rack-scale systems are already sold out well into 2025. But NVIDIA's dominance is precisely what is motivating competitors to accelerate. AMD's Instinct MI300X has gained meaningful traction with hyperscalers seeking supply diversification, and AMD's MI350 series is expected to close the performance gap further. Intel's Gaudi 3 AI accelerator has found limited but real adoption at Microsoft and other customers. Custom silicon — Google's TPU v5, Amazon's Trainium 2, Microsoft's Maia 100, and Meta's MTIA — represents a structural effort by hyperscalers to reduce NVIDIA dependency, though these chips still rely on the same constrained packaging and substrate supply chain.
For companies like TSMC, the price environment is broadly positive. TSMC has already implemented pricing increases on advanced nodes and is in a position of extraordinary leverage as the sole credible manufacturer of sub-5nm logic. Samsung Foundry and Intel Foundry Services are competing aggressively on price to win customers, but their yield and volume limitations mean they cannot fully absorb demand overflow from TSMC.
The secondary chipmakers raising prices — the substrate manufacturers, power IC suppliers, and packaging specialists — occupy a strategically interesting position. Companies like ASE Technology, Amkor Technology, and SPIL are seeing advanced packaging demand (CoWoS, SoIC, and related technologies) far outstrip their current capacity. ASE has guided for significant capex increases through 2025 and 2026. These are not household names, but their capacity constraints are a genuine bottleneck for the entire AI hardware ecosystem.
For Microsoft specifically, the competitive landscape with Google and Amazon in cloud AI services means that cost increases cannot always be passed through immediately without risking customer attrition. This creates margin pressure that will influence product and pricing strategy for Azure AI services throughout 2025 and 2026. Organisations building on enterprise productivity software platforms should monitor these dynamics closely.
Expert Perspective
From a strategic standpoint, what we are witnessing is the semiconductor industry's version of a commodity supercycle — but with important structural differences from historical precedents. Unlike the oil supercycles of the 1970s or the memory chip cycles of the 1990s, this demand wave is being driven by a technology transition that shows no signs of reversal. Enterprises are not deploying AI because it is fashionable; they are deploying it because early adopters are demonstrating measurable productivity gains, and competitive pressure is forcing laggards to follow.
Industry analysts at firms like IDC and Gartner have projected that AI-augmented enterprise software will become the default expectation rather than a premium feature by 2027. That transition requires sustained infrastructure investment, which means the capex numbers being discussed today — $136 billion in 2026 — may themselves prove conservative if AI adoption accelerates faster than current models anticipate.
The risk scenario worth monitoring is a demand air pocket. If enterprise AI adoption disappoints — due to ROI challenges, regulatory constraints, or a broader economic slowdown — hyperscaler capex could moderate sharply, leaving smaller chipmakers with expanded capacity and falling prices. This is not the consensus view, but it is a real tail risk that supply chain planners should not dismiss entirely.
The opportunity, meanwhile, is for companies that can credibly offer supply chain diversification. TSMC's new facilities in Arizona, Japan, and Germany represent a geopolitical as much as a commercial strategy, and the companies that establish themselves as reliable suppliers in these new geographies will command premium relationships with hyperscalers anxious about concentration risk in Taiwan.
What This Means for Businesses
For business leaders and IT decision-makers, the practical takeaway from this semiconductor pricing shift is straightforward: AI infrastructure costs are entering a sustained inflationary phase, and budget models built on 2023 or early 2024 cloud pricing assumptions need to be revisited.
Organisations with significant cloud AI workloads should engage their hyperscaler account teams now to understand committed-use discount structures and whether locking in longer-term pricing agreements makes sense given the forward cost trajectory. For those running Windows-based infrastructure and considering AI-augmented workflows through Microsoft Copilot, understanding the full licensing cost — including both the Copilot add-on and the underlying Microsoft 365 subscription — is essential for accurate TCO modelling. Sourcing a genuine Windows 11 key through authorised resellers can help manage the software side of the cost equation while hardware and cloud costs rise.
IT departments should also begin evaluating workload placement more rigorously — not every AI inference task needs to run on the most expensive GPU infrastructure. Edge AI, smaller fine-tuned models, and hybrid architectures can deliver significant cost savings compared to defaulting to large foundation model APIs for every use case. The companies that navigate this cost environment most successfully will be those that treat AI infrastructure as a strategic resource to be optimised, not a utility to be consumed without scrutiny.
Key Takeaways
- Asia's smaller semiconductor companies are raising prices alongside industry giants, signalling that AI-driven demand has permeated the entire chip supply chain, not just leading-edge logic manufacturers.
- Global AI-related capital expenditure is projected to exceed $136 billion in 2026, a 25% year-on-year increase, driven by hyperscaler commitments from Microsoft, Google, Amazon, and Meta.
- Advanced packaging substrates, power management ICs, and specialised connectors are among the components seeing the sharpest price increases as bottlenecks shift from raw compute silicon to surrounding ecosystem components.
- Enterprise cloud costs for AI workloads are likely to rise 15–30% over the next 24 months as hardware cost inflation propagates through hyperscaler pricing models.
- Microsoft's Copilot and Azure AI services face margin pressure as infrastructure costs rise, with potential implications for enterprise licensing and feature rollout timelines.
- Businesses should reassess cloud AI workload placement, explore committed-use pricing agreements, and optimise software licensing costs to offset rising infrastructure expenditure.
- The demand wave appears structurally sustained rather than cyclical, but a demand air pocket remains a tail risk if enterprise AI ROI fails to meet expectations at scale.
Looking Ahead
Several near-term developments will determine how this story evolves. TSMC's quarterly earnings calls through 2025 will provide the most authoritative read on whether leading-edge capacity constraints are easing or tightening. NVIDIA's next earnings disclosure — expected to show continued extraordinary data centre revenue growth — will either validate or temper the capex projections underpinning current pricing assumptions.
Watch also for announcements from the major substrate manufacturers — Ibiden, Shinko Electric, and Unimicron — regarding capacity expansion timelines. Advanced packaging capacity is currently the most acute bottleneck in the AI hardware supply chain, and any credible expansion announcements will be closely scrutinised by investors and procurement teams alike.
On the demand side, the enterprise AI adoption data from Microsoft's fiscal Q3 2025 earnings (expected April 2025) will be particularly telling. Copilot seat counts, Azure AI revenue growth rates, and management commentary on enterprise purchasing behaviour will provide a ground-level view of whether AI demand is sustaining the trajectory that is driving these extraordinary capex commitments — and whether the semiconductor price increases now rippling through Asia's supply chains are fully justified by the demand that is coming.
Frequently Asked Questions
Why are smaller Asian chip companies raising prices now, and not just the big names?
The AI infrastructure boom has created demand that extends far beyond leading-edge logic chips from TSMC or NVIDIA. The build-out of AI data centres requires enormous quantities of advanced packaging substrates, power management ICs, thermal components, and specialised connectors — products made by hundreds of mid-tier and smaller Asian semiconductor companies. As hyperscalers like Microsoft, Google, and Amazon race to expand AI capacity, these previously overlooked components have become genuine bottlenecks. Smaller chipmakers now have pricing power they rarely possessed in previous cycles, because their customers have no viable short-term alternatives and cannot afford to slow down AI infrastructure deployment.
How will rising semiconductor prices affect Microsoft 365 and Azure AI service costs for businesses?
The cost increases in the semiconductor supply chain will propagate through the technology stack over time. Microsoft's Azure AI services — including the infrastructure underpinning Copilot for Microsoft 365 — run on data centre hardware that is becoming more expensive to build and operate. While hyperscalers have some ability to absorb short-term cost increases, sustained inflationary pressure on hardware will eventually influence cloud service pricing, the pace of AI feature rollouts, and enterprise licensing structures. Businesses should model scenarios where Azure AI compute costs rise 15–30% over the next 24 months and evaluate whether current cloud commitments adequately protect against this risk.
What is advanced packaging, and why is it such a critical bottleneck in the AI chip supply chain?
Advanced packaging refers to techniques that integrate multiple chips into a single package with extremely high-density interconnects — technologies like TSMC's CoWoS (Chip-on-Wafer-on-Substrate), Intel's EMIB, and various 3D stacking approaches. NVIDIA's H100 and Blackwell GPUs, for example, use CoWoS packaging to integrate the GPU die with High Bandwidth Memory (HBM) stacks on a silicon interposer. The substrates and interposers required for these packages are manufactured by a small number of companies — primarily Ibiden, Shinko Electric, and Unimicron — whose capacity cannot be expanded quickly. This concentration creates a genuine supply bottleneck that is distinct from, and in some ways more acute than, the constraints on the GPU silicon itself.
Should businesses accelerate or delay AI infrastructure investments given rising semiconductor prices?
The answer depends heavily on the specific use case and competitive context. For organisations where AI-augmented workflows are delivering measurable productivity gains — such as using Microsoft Copilot for document processing, code generation, or customer service automation — delaying investment to wait for lower prices is likely a false economy, as competitors who deploy earlier will compound their advantage. However, businesses should be strategic about workload placement: not every AI task requires expensive GPU infrastructure, and smaller fine-tuned models running on less exotic hardware can deliver strong results at lower cost. The prudent approach is to invest selectively in high-ROI AI use cases now while building the internal capability to optimise infrastructure costs as the market matures.