In January, our Semiconductor Industry Trends Report 2026 set out five themes we expected to shape the industry this year: AI as the primary growth driver, advanced packaging as an emerging constraint, uneven mature-node supply, inventory as a strategic priority, and sustainability and regional manufacturing shaping investment decisions.
Six months in, we got the big picture right, but the scale has surprised almost everyone. Global semiconductor revenue is now forecast to grow 64% in 2026, according to the technology research firm Gartner, the strongest growth the industry has seen in two decades. AI hasn't just remained the top driver; it's pushed memory and packaging into a shortage of their own.
Meanwhile, the mature-node “correction" we flagged as an ongoing risk in January (industry shorthand for working through stock that built up faster than demand) has largely resolved, only for the imbalance to reappear somewhere new: memory.
This review checks our original predictions against what actually happened, examines what surprised us, and outlines what original equipment manufacturers (OEMs), electronics manufacturing services providers (EMS), and procurement teams should watch in the second half of the year.
|
Original outlook |
Mid-year assessment |
Status |
|
AI would remain the primary growth driver |
Confirmed, and stronger than expected. Revenue growth of 64% forecast for 2026, with AI at roughly 30% of total industry revenue |
Exceeded |
|
Advanced packaging would become a strategic constraint |
Confirmed. CoWoS (an advanced chip-packaging technology) and 2.5D/3D packaging capacity remain tight enough to shape TSMC's own capital spending plans |
Confirmed |
|
Mature-node supply would remain uneven |
Bifurcated. By early 2026, major chipmakers said their stock levels were back to normal after two years spent working through excess supply built up since 2023. A new shortage emerged in memory instead, so the imbalance moved rather than disappeared |
Changed |
|
Inventory management would become strategic |
Confirmed, but the shape changed. Shortage and excess now exist side by side across different categories (memory versus consumer and mature-node) rather than easing uniformly |
Confirmed |
|
Sustainability and regional manufacturing would shape investment |
Confirmed. Regionalisation continued in concrete terms, including new US manufacturing investment from European players |
Confirmed |
Sources: Gartner, Texas Instruments Q4 2025 results, TSMC Q2 2026 capital expenditure guidance
The semiconductor industry is no longer moving through one recovery cycle. AI-linked technologies are attracting extraordinary investment, while other markets recover at their own pace, and supply conditions vary sharply depending on the component, process and end market.
This has real implications for procurement teams. Headline indicators, like total semiconductor revenue or average lead times, no longer tell the full story on their own. Decisions increasingly depend on understanding specific technologies, supplier capacity and product lifecycles rather than reading the market as a whole.
Closer to home, UK dedicated semiconductor companies grew estimated revenue by around 7 to 10% to £10.6 billion this year, according to Perspective Economics' latest sector analysis for the Department for Science, Innovation and Technology, with memory increasingly seen as an under-recognised UK opportunity given the country's compound semiconductor and packaging strengths.
The headline numbers
Equity and ETF performance, H1 2026
Broadcom, the US chip and software company, gave cautious AI chip guidance that triggered a sharp sell-off (many investors selling at once, driving the price down quickly) on 3–6 June, wiping an estimated $1.3 trillion off sector market value in a matter of days (Intellectia).
A second, separate leg of selling hit later the same month: on 22–23 June, semiconductor and tech stocks fell again globally, with South Korea's Kospi down 10% as SK Hynix and Samsung both dropped more than 12% (CNBC). Both recovered relatively quickly, but they're a reminder that even a genuinely strong cycle isn't immune to volatility.
When we published our original report, artificial intelligence was already reshaping the semiconductor industry. Six months of earnings data later, its influence has only grown stronger.
Demand for AI infrastructure continues to drive investment across the whole value chain, from chip designers and foundries to memory manufacturers, packaging specialists and equipment suppliers. While automotive, industrial and consumer electronics recover at a more measured pace, AI has remained the industry's clearest growth engine, and the numbers now back that up more strongly than we expected in January.
Industry snapshot
Earlier technology cycles tended to lift one part of the semiconductor industry at a time. This one is different. Investment is flowing into processors, memory, networking, advanced packaging and manufacturing equipment simultaneously, driven by hyperscalers, enterprise AI adoption and a growing number of sovereign AI programmes (government-led projects building that kind of independent AI capacity, which is one more source of demand for semiconductors, alongside the big cloud companies and ordinary businesses adopting AI).
This breadth is why the rally broadened well beyond Nvidia in H1 2026, with memory, foundry and equipment names doing much of the work behind SMH's 82% and SOXX's 113% gains.
See how Component Sense helps you recover value from your excess stock.
In our view, the AI boom has created two semiconductor markets rather than one. Companies supplying AI infrastructure are running flat out with capacity locked in years ahead. Everyone else is recovering at very different paces, which helps explain why shortages and excess stock can coexist, depending on the technology in question.
Of everything in the original report, this is the theme that's moved furthest from where we expected it to sit.
In January, we said high-bandwidth memory was becoming a constraint on AI performance. By mid-2026, memory has become a defining story in its own right, with a name analysts have coined for it: “memflation”. Gartner's latest forecast captures just how sharp that shift has been.
This theme has drawn huge investor interest. The Roundhill Memory ETF (DRAM), built to give investors direct exposure to global memory stocks, became the fastest ETF in history to reach $10 billion in assets under management, hitting that mark in just 43 days (US News).
Industry snapshot
Figure 4. Growth tracking across the primary memory segments in 2026.
Note on the data: DRAM (temporary working memory) and NAND flash (long-term storage) represent independent market segments rather than a direct head-to-head comparison. High-Bandwidth Memory (HBM) is excluded from the chart as supply remains completely sold out under long-term contracts. Data reflects a mix of actualised first-half results and updated full-year forecasts.
Memory pricing doesn't stay contained to memory chips. It flows into everything built around it, from AI servers to consumer devices to industrial systems that use standard DRAM and NAND rather than AI-grade HBM. Gartner has warned that memflation is likely to delay or destroy non-AI demand into 2028 in some applications, simply because components have become too expensive or too hard to get hold of.
This connects directly to inventory strategy. When a component category swings this sharply on price and availability, forecasting accuracy drops, and the temptation to over-order as a hedge increases.
At the start of 2026, the assumption was that easing chip shortages would lead to a more balanced market overall. Six months of results tell a more mixed story.
The excess stock that built up in mature-node chips between 2023 and 2025 has largely cleared (inventory correction). Several of the sector's biggest analogue and mixed-signal names now say stock levels with their distributors are back to normal, with automotive and industrial demand moving in ways few expected.
Industry snapshot
Back in January, we said some parts of the market would run short of stock while others would be sitting on too much. That's still true today, it's just not the same parts anymore. Shortages have shifted from mature-node automotive and industrial components to memory and advanced packaging, while non-AI consumer categories are now the ones working through slower-moving stock.
Analyst house, iFAST, notes that global semiconductor sales grew nearly 20% year-on-year in Q2, the seventh consecutive quarter of double-digit growth, but that inventory turnover for non-AI consumer chips remains sluggish despite the rise of AI-enabled PCs (iFAST Global Markets).
TSMC reported Q2 2026 revenue of $40.2 billion, up 33.7% year-on-year and at the top of its own guidance range (TradingKey). The figure matters less on its own than what it signals: sustained willingness among hyperscalers to keep funding the expansion of AI computing capacity. TSMC's 2026 capital budget of $52 to $56 billion, the largest in its history, is being run at the high end of that range, effectively a multi-year bet that demand for advanced chip manufacturing, CoWoS packaging (Chip-on-Wafer-on-Substrate, the process that stacks multiple chips together for higher performance) and HBM integration keeps growing rather than levelling off (TechTimes).
The push to regionalise capacity, which we called out as a 2026 theme, has continued in concrete terms. Silex Microsystems' $40 million acquisition of a 200mm fab in Pennsylvania from Onsemi, another US chipmaker, is a useful live example: a European manufacturer buying US capacity specifically to sit closer to customers and reduce geopolitical exposure, rather than building from scratch (Yole Group).
Figure 5. Regional map of major semiconductor manufacturing investments announced or finalised in 2026.
We'd argue supply chain resilience in mid-2026 isn't measured by whether shortages have eased overall. It depends on which technology, which end market and which region you're asking about, the same conclusion we reached in January, now backed by six more months of earnings data.
Not everything in the first half of 2026 followed our original outlook exactly, even where the broad direction was right.
Memory overtook GPUs as the tightest constraint: In January, most of the conversation around AI hardware scarcity centred on GPUs (graphics processing units, the chips originally built for rendering graphics that became the workhorse of AI computing). By mid-2026, memory is arguably the bigger story, with Micron's entire 2026 HBM output sold out under long-term contracts and Gartner forecasting a threefold rise in memory revenue. We said HBM would be a constraint; we didn't expect it to become a bigger issue than processor scarcity.
The stock market fell sharply twice in the same month: The two sell-offs covered above, in early and late June, both recovered fast, but their speed and scale were sharper than the “periodic imbalances” language in our original scenario planning anticipated.
Share prices have risen faster than earnings alone can justify: Analyst house, iFAST puts semiconductor valuations well above their historical average, with much of the good news on earnings, export control easing and tariff exemptions already reflected in the price (iFAST Global Markets). That's a caution flag for the second half that our January report didn't fully anticipate.
The mature-node correction resolved faster than we expected: We expected uneven mature-node supply to persist through 2026. Instead, most major analogue and mixed-signal suppliers describe the correction as largely complete by Q1, with the constraint moving to memory rather than easing across the board.
The theme from January that this year has reinforced most clearly is inventory as a strategic asset class, and the first half of 2026 has made that case even more strongly than we originally did.
Shortage and excess stock are now happening together more visibly than at almost any point since the pandemic, just in different categories than before. Memory and advanced packaging remain scarce enough to be rationed years in advance, with TSMC's own packaging shortfall not expected to fully close until 2027 (TrendForce).
At the same time, non-AI consumer components continue to move through slower-than-ideal inventory, with mobile semiconductor revenue forecast to fall in 2026 on cost pressure alone (IDC). Within mature-node chips specifically, the picture is more mixed than a single trend: Gartner has reported elevated stock levels sitting with suppliers in analogue, discrete and general-purpose microcontroller categories amid weak demand (Gartner, cited by Sourceability), even as other mature-node categories like automotive microcontrollers show signs of tightening again.
This is the same pattern we described in January as a normal part of how the industry works. Long lead times and imperfect forecasting mean shortage and excess stock can exist side by side, and memflation-driven price swings sharpen the incentive to over-order defensively. That's the same dynamic that fed the double-ordering cycles of 2021 and 2022.
For OEMs and EMS providers, this means treating inventory decisions as a set of different choices, not one fixed approach. Components tied to memory-constrained designs need very different handling to components in mature-node categories where supply has normalised.
Redistributing excess stock through the right channels remains one of the more reliable ways to recover working capital without compromising traceability or compliance.
If you're carrying excess stock from either side of this divide, our Consignment vs. Outright Purchase guide walks through which route recovers value fastest, and InPlant™ and Consignment remain two of the more flexible ways to act on it.
For OEMs and design teams: treat memory and advanced packaging lead times as planning constraints rather than variables to solve later. Gartner's own advice, that suppliers should be cautious about signing pricing agreements with unfavourable terms extending beyond 2027, applies just as much to buyers locking in long-term commitments during a volatile pricing window.
For EMS providers: segment inventory by lifecycle stage and category rather than managing it as one pool. Memory-adjacent stock and mature-node stock are moving in opposite directions and need different strategies.
For procurement teams: build flexibility into sourcing rather than betting on a single outcome. With valuations elevated and two sharp market sell-offs already behind us this year, the ability to change course quickly, through using redistributors, matters more than chasing the lowest price in a stable market that may not last.
For finance and supply chain leaders: revisit working capital tied up in excess stock now, rather than waiting for the next inventory swing to force the issue. The organisations best placed for H2 are the ones already treating inventory as an asset to actively manage, not a buffer reviewed once a year.
In the same way as our original report, we've set out three ways the rest of 2026 could play out.
Bull case: AI infrastructure spending continues to outpace even Gartner's upgraded forecasts, memory and packaging capacity expansions come online on schedule, and hyperscaler capital spending commitments extend into 2027 and beyond.
Base case: growth remains strong but increasingly selective. Memory and packaging stay tight through year-end, mature-node markets hold their recent stability, and further ups and downs are likely but shouldn't cause lasting damage. This is broadly the path H1 has already followed.
Bear case: memflation delays enough non-AI demand to drag on the wider industry, valuations reset more sharply than the earlier sell-offs, and capital spending guidance from major players turns cautious rather than confident. Excess stock builds up faster in components tied to paused or delayed projects.
Component Sense's excess stock solutions are built for exactly this kind of environment, where the real risk is genuine uncertainty about which components will be short and which will be sitting in excess six months from now, rather than a single downturn.
Through InPlant™, Consignment and Outright Purchase, we help manufacturers regain control of excess stock whichever scenario plays out.
Looking back, our original Semiconductor Industry Trends Report 2026 got the big picture right over the past six months. AI is still the dominant force. Advanced packaging is still a constraint. Inventory is still strategic.
The details moved further and faster than we expected. Memory became a headline story in its own right. The mature-node correction resolved sooner than anticipated. Two sharp market sell-offs showed how quickly sentiment can turn even in a genuinely strong cycle. Taken together, here's what matters most for the second half: organisations that build flexibility into sourcing, treat inventory as a strategic asset, and stay close to where capacity and demand are actually shifting will be better placed than those relying on a single forecast.