Semiconductor Industry Outlook 2026
Follow the money: top picks across AI, memory, equipment and EDA
Hi, Investor! 👋🏼
I’m Jimmy - welcome back to another edition of Jimmy’s Journal.
In today’s report, we’ll break down the 2026 outlook for semiconductors - and the key names I’m betting on in the sector.
I hope you enjoy the read.
TL;DR (What Actually Matters)
Semiconductors remain the backbone of the tech value chain and the long-term outperformance thesis is still intact.
The growth impulse heading into 2026 is dominated by AI infrastructure, with data center capex still expanding at a pace that’s well above historical norms.
“Bubble” concerns will persist as long as spending stays elevated - but adoption is broadening (consumer + enterprise), and workload intensity is rising, which keeps demand durable.
Supply constraints remain the gating factor (and paradoxically can extend the cycle by pushing deployments out over multiple years).
We expect 2026 to be a “barbell” year: continued strength in AI-linked segments + improving cyclicals as inventories normalize.
Found this content valuable? Share it with your network! Help others discover these insights by sharing the newsletter. Your support makes all the difference!
Why Semis Keep Winning Over Time:
Semiconductors are cyclical, but the long-term track record is hard to ignore:
Over the past decade, semiconductor and semiconductor equipment stocks have outperformed the broader market by ~1,400 bps per year (~14% annually). In other words, patient investors have historically been rewarded for staying exposed to the backbone of the digital economy.
And this isn’t a “one-off” driven by a single product cycle. The outperformance has been supported by structural forces that continue to compound:
Semis are the foundational building block of innovation:
Every major technology wave - cloud, mobile, AI - eventually turns into silicon demand. Semiconductors sit at the base layer of the tech value chain, which makes them one of the most direct ways to ride innovation over long horizons.
The industry has become less cyclical over time:
Semis used to be dominated by a narrow set of end markets. Today, demand is far more diversified across data centers, networking, industrial, auto and consumer. At the same time, supply growth has become more disciplined - meaning fewer “boom-and-bust” episodes driven by reckless overbuilds.
New secular growth engines keep showing up:
The set of demand drivers keeps expanding: cloud data centers, EVs, IoT, and now AI/deep learning. Each wave increases semiconductor content and complexity, raising the long-term revenue floor for the industry.
Profitability and free cash flow matter more than ever:
The sector has shifted from “growth at any cost” to sustainable profitability, free cash flow generation, and capital returns. Stronger margins + stronger shareholder returns tend to translate into stronger long-term stock performance.
Consolidation supports scale, resilience, and valuation:
Industry consolidation has reduced fragmentation, improved pricing discipline, and increased scale advantages. Larger platforms can diversify end exposure, invest through downturns and expand earnings/free cash flow power - creating a more durable long-term setup.
Enjoying the content? Don’t miss out on more exclusive insights and analyses. Subscribe now and stay updated.
What We Saw in 2025:
2025 was a year of two speeds: AI-related demand stayed red-hot, while the rest of the industry quietly moved from “surviving the downcycle” to building the foundation for a broader recovery.
At a high level, total semiconductor industry revenue delivered roughly +21% y/y growth in 2025 (+18% y/y excluding memory), on top of +20% growth in 2024 (+8% ex-memory).
Under the hood, the drivers were very clean:
Units Sold (ex-discretes) up +11% y/y, and
Average Selling Price (also known as ASPs) up ~10% y/y (largely driven my memory chips).
Importantly, industry fundamentals turned positive y/y back in 2023, stayed constructive through 2024, and momentum steadily improved throughout 2025 - likely exiting the year at a high point in 4Q25, even with macro uncertainty still in the background.
Found this content valuable? Share it with your network! Help others discover these insights by sharing the newsletter. Your support makes all the difference!
A Better Setup than 12 Months Ago:
If you zoom out, the starting point for 2026 looks healthier than the setup entering 2025.
A year ago, cyclical segments were fighting end-demand pressure plus customer de-stocking (customers burning down inventories). Today, the picture is more constructive:
Customer inventories are leaner across most markets (autos remain the main exception);
At the end of last year, inventory on semiconductor balance sheets stood near 130 days, modestly above the five-year average of 118 days - a difference largely driven by the AI infrastructure ramp and longer, more complex supply chains (see the table below).
Broad-based and non-AI companies have continued to actively de-risk, pushing inventories down and keeping manufacturing utilization intentionally below normal.
AI-linked companies, on the other hand, are leaning the opposite way - holding higher inventory levels as a strategic buffer to meet strong forward demand and navigate persistent constraints across the AI supply chain.
Outside the AI complex, supply and demand are now much closer to balance, and end-market demand has held up better than feared earlier in the year despite trade and tariff noise.
The industry posted multiple quarters above the 10-year average q/q growth, and by 3Q25, growth printed +15.8% q/q versus a +8.1% 10-year average. That’s not a late-cycle “melt-up” signal by itself - but it does confirm the cycle had real breadth by the second half.
This matters because it reduces one of the biggest cycle risks: shipping into a channel that’s still bloated.
The only data point that really stood out was ADI’s lead time sitting near ~13 weeks, versus a normalized level closer to 5-6 weeks. We’ll be monitoring it closely.
Enjoying the content? Don’t miss out on more exclusive insights and analyses. Subscribe now and stay updated.
Where I’m Optimistic Going Into 2026:
Multiple datapoints across the semiconductor stack suggest the upcycle has room to run into 2026.
The difference versus prior cycles is that this one isn’t being carried by a single end market - it’s being supported by AI infrastructure demand at the top, and a gradual normalization in the rest of the industry underneath it.
Below is how I’d rank the segments I’m most constructive on into 2026 - from highest conviction to lowest - and the key public companies most exposed to each theme.
1. Memory and Storage:
This is the cleanest setup, in my view, because AI doesn’t just increase compute demand - it increases memory demand even faster. Every new generation of AI hardware tends to pull higher memory content, higher bandwidth and more premium SKUs.
DRAM (Dynamic Random-Access Memory) is the “working memory” used for fast, active compute - and AI servers carry meaningfully more DRAM content than traditional servers.
NAND (flash storage used in SSDs) benefits from the storage side of AI: datasets, checkpointing and local caching push more SSDs per rack.
HBM (High Bandwidth Memory) - stacked, ultra-fast memory paired with AI accelerators - remains one of the biggest gating items in the entire AI stack.
The setup heading into 2026 is still tight. One helpful way to see it: memory isn’t behaving like a classic consumer-driven cycle anymore - it’s increasingly being paced by AI infrastructure rollouts and server content growth.
This slide is from Micron Technology’s ($MU) most recent earnings presentation, released late last year. Click here to view the full deck.
Main beneficiaries: Micron ($MU), Western Digital ($WDC), SanDisk ($SNDK), Seagate ($STX).
2. Data Centers:
The heartbeat of this cycle is still data center capex - and the most important signal isn’t just the level of spending, but the direction of revisions.
Over the past few months, expectations have moved sharply higher: 2025 data center spend is now projected to grow +65% y/y, up from roughly +40% earlier last year. That revision alone implies $115B+ of incremental spending in 2025 versus 2024, following an already massive $75-80B step-up the year prior.
Top hyperscalers remain the anchor, but they’re no longer the full story.
Neoclouds and sovereign AI buildouts are increasingly meaningful at the margin.
And most importantly, inference is turning AI from an experiment into an always-on workload - which supports a more durable demand curve.
Main beneficiaries: NVIDIA ($NVDA), AMD ($AMD), Broadcom ($AVGO), Marvell ($MRVL).
3. Custom Silicon (ASICs):
If GPUs are the headline of AI, custom silicon is the long game.
The incentive is obvious: hyperscalers want better performance-per-watt, tighter software integration, and lower total cost at scale.
ASIC (Application-Specific Integrated Circuit) is a custom chip designed for a specific workload/customer - optimized versus merchant, off-the-shelf alternatives.
The reason this trend persists is that the economics compound. Once a platform is deployed at hyperscale, even small efficiency gains translate into massive savings in power, space, and operating cost. That’s why custom programs tend to expand over time, not shrink.
This is also one of the strongest “second-order” beneficiaries of the AI buildout: even when GPU mix shifts, custom programs keep ramping as cloud operators push for differentiated infrastructure.
Broadcom ($AVGO), by the way, has been particularly vocal about the scale of this opportunity. The company recently confirmed Anthropic as Google’s largest TPU customer, with an estimated $21B revenue opportunity in 2026 tied to continued TPU proliferation. Just as importantly, Broadcom is sitting on a massive $73B AI backlog, reinforcing that demand isn’t just strong - it’s still building.
Main beneficiaries: Broadcom ($AVGO), Marvell ($MRVL), Intel ($INTC).
4. Foundries and Leading-Edge Manufacturing:
Foundries are the chokepoint of this cycle. You can design the best AI chip in the world, but you still need a factory that can manufacture it at scale - with high yield and predictable cycle times.
As AI accelerators, networking silicon, and custom programs push deeper into advanced nodes, leading-edge wafer capacity becomes the gating factor.
A few key points heading into 2026:
2nm is the next major inflection. It’s more than a simple “shrink” - it’s a technology transition that raises complexity and increases the value of execution.
TSMC remains the industry’s central “toll booth.” The majority of cutting-edge AI silicon flows through its most advanced nodes, so incremental AI demand ultimately translates into more leading-edge wafers.
Supply constraints can extend the cycle. Tight capacity and long cycle times don’t kill demand - they often stretch deployments and keep pricing power intact.
Main beneficiaries: TSMC ($TSM), Intel ($INTC).
5. Networking and Optical:
This is the segment most investors underweight in their mental model. AI isn’t just compute - it’s compute at scale, and scale requires bandwidth.
As clusters expand, the bottleneck moves from the GPU to the fabric connecting GPUs together. That’s why networking content is rising structurally across AI data centers:
Switching throughput is scaling rapidly
Optical connectivity becomes more critical as speeds step up
Rack-to-rack communication becomes a larger portion of the system
We’re seeing sustained design activity and orders across the key building blocks of the AI fabric:
Broadcom ($AVGO) in switching/routing silicon, PCIe switching and optical
Marvell ($MRVL) in 800G/1.6T PAM4 DSP chipsets (with strong demand tied to next-gen AI deployments)
PAM4 DSP chipsets are the “signal-cleaning brains” inside high-speed optical transceivers.
In practice, they enable hyperscalers to scale bandwidth across data centers (rack-to-rack and row-to-row connectivity) while maintaining signal integrity at these next-gen speeds.
Astera Labs ($ALAB) in Gen5-6 retimers, which are increasingly essential for high-speed connectivity inside AI racks
Marvell has maintained 70%+ market share in PAM4 DSP chipsets for optical transceivers, helping hyperscalers like Google, Amazon, Meta and Microsoft scale their transitions to 200G/400G optical connectivity.
More importantly, this upgrade cycle tends to be self-reinforcing: as switching bandwidth doubles on an aggressive cadence, the rest of the network stack has to follow - driving sustained demand for optical components, signal processing, and connectivity silicon.
Main beneficiaries: Broadcom ($AVGO), Marvell ($MRVL), Astera Labs ($ALAB), MACOM ($MTSI), Arista Networks ($ANET)
6. EDA:
EDA is the “boring winner”, and that’s a compliment.
The core dynamic is simple: chip complexity is rising faster than engineering productivity, and verification demands keep getting heavier as leading-edge design moves toward chiplets, advanced packaging, and tighter power/performance constraints.
EDA demand is anchored to semiconductor R&D budgets - and those budgets tend to be far more resilient than capex in a downturn.
The industry structure is also attractive: the market is highly consolidated, and pricing power tends to be durable.
If you want the most defensive exposure to “more complex chips,” EDA is one of the cleanest ways to express it.
Main beneficiaries: Synopsys ($SNPS), Cadence ($CDNS), Arm ($ARM).
7. Semiconductor Equipment:
Equipment benefits from rising manufacturing complexity, but it’s the most timing-sensitive segment on this list. The reason is that orders can be lumpy and heavily dependent on clean-room readiness, customer capex timing and node transition schedules.
That said, the secular support is real:
Leading-edge logic is getting more complex
Memory is being pushed into more advanced architectures
Advanced packaging is becoming a bigger slice of overall spend
In other words, equipment remains levered to the “hardest” parts of semis - but the path can be less smooth quarter-to-quarter.
Main beneficiaries: ASML ($ASML), Applied Materials ($AMAT), Lam Research ($LRCX), KLA ($KLAC).
Found this content valuable? Share it with your network! Help others discover these insights by sharing the newsletter. Your support makes all the difference!
M&A’s Likely to Continue:
One underappreciated support for the semi complex is that consolidation never really stopped - it simply changed shape.
Deal volume has been volatile year to year, but the direction is clear: strategic buyers continue to use M&A to buy scale, diversify end-market exposure and secure critical IP.
Bloomberg and company reports estimate 2025 semi + semicap deal volume at roughly $47B, up from $30B in 2024 and $25B in 2023 (after a much hotter 2020-2022 window).
What’s changed is the mix. Instead of mega-mergers, we’re seeing more tuck-in acquisitions - smaller and mid-sized companies being absorbed by larger platforms to accelerate time-to-market, strengthen product breadth and defend competitive positioning in an increasingly capital-intensive industry.
Looking into 2026, the setup is constructive: valuations remain elevated in parts of the stack, which can slow deal count - but it also tends to push average deal size higher, especially as AI makes certain assets strategically “must-own” rather than “nice-to-have.”
Here’s the list of notable semiconductor and semicap deals announced over the past few years:
Enjoying the content? Don’t miss out on more exclusive insights and analyses. Subscribe now and stay updated.
What Can Break the Thesis?
The 2026 semi setup is constructive - but it’s not bulletproof.
In my view, this cycle won’t break because “AI collapses.” It breaks if:
spending decelerates faster than expectations, or
supply catches up sooner than the market is positioned for.
Here are the five risks I’m watching most closely:
1. AI capex digestion (the market prices the slope, not the level):
If AI capex goes from “up massively” to “still up, but less massively,” the market can re-rate the group even if fundamentals remain strong. Multiples compress long before revenue actually rolls over.
What could lead to this? In short: below-expected returns for hyperscalers.
If the productivity gains from this wave of investment don’t show up this year - either in revenue growth, margin expansion or clear unit economics - the market will start questioning the capex trajectory and could punish the entire sector.
2. HBM and memory overshoot (supply arrives faster than demand):
Memory is the tightest part of the stack today - which is exactly why it carries the most “overshoot” risk.
If HBM supply ramps too aggressively into 2H26, or if customers front-load builds and then pause, pricing can soften quickly. Even a small change in pricing sentiment can swing investor expectations.
3. Networking timing risk:
Networking is a structural AI bottleneck, but the spend profile can be lumpy. It’s entirely possible that the upgrade cycle plays out more back-half weighted - where deployments don’t vanish, but shift right.
Stocks tend to punish timing mismatches even when the long-term thesis is intact.
4. Geopolitics and export controls:
Policy risk remains a constant headline overhang. Restrictions can re-route demand across regions and customers, while adding friction in the form of compliance, redesigns, and supply chain complexity.
Even if end demand stays healthy, uncertainty alone can pressure multiples.
5. Social backlash risk (jobs, consumption and energy constraints):
AI is not just a tech cycle - it’s increasingly a social and political one. If adoption accelerates faster than society can absorb the disruption, the backlash can become a real macro headwind.
There are three channels I’m watching:
Jobs and wage pressure: if AI-driven productivity gains translate into visible job displacement faster than new roles are created, consumer confidence can weaken - and policymakers tend to respond.
Consumption sensitivity: weaker employment dynamics usually show up downstream as softer discretionary spending, which can hit “non-AI” cyclical demand sooner than investors expect.
Energy bills + regulation risk: AI infrastructure is power-hungry. If electricity costs keep rising (especially in key regions) and grids struggle to keep up, AI capex can become constrained by energy availability - or by regulation. A tighter regulatory stance (for example, around power usage, permitting, or grid prioritization) could slow deployment timelines even if demand stays strong.
Which KPIs should we be watching?
If you want a clean dashboard for 2026, this is it:
Hyperscaler capex revisions: are estimates moving up again, or starting to roll over?
Lead times: especially for broad-based names (ADI was a good tell).
Memory pricing + HBM allocation: DRAM/NAND trends and whether HBM stays constrained.
Utilization + inventory days outside AI: are cyclicals healing without rebuilding excess inventory?
WFE spending guidance: equipment spending is the best real-time proxy for manufacturing intensity.
Found this content valuable? Share it with your network! Help others discover these insights by sharing the newsletter. Your support makes all the difference!
How to Position (The 2026 Playbook):
The reality is that forward multiples across the AI-semi complex are still elevated.
But they’re not “randomly high” - they’re being supported by something very real: earnings growth that keeps getting revised up. In other words, valuation is rich, but it hasn’t become detached from fundamentals.
That said, this is also a crowded trade.
Positioning feels tight, expectations are high, and the market mood is basically: everyone wants exposure, and everyone looks like they’re standing close to the exit. When that’s the setup, small disappointments can create outsized drawdowns - not because the thesis breaks, but because flows reverse.
So the goal for 2026 is simple:
Be positioned - but don’t get reckless.
I don’t believe this is a bubble. AI is a real infrastructure buildout, and semis sit at the center of it. The mistake would be treating it like a lottery ticket instead of a multi-year cycle (just like any other).
Here’s the posture that makes the most sense, in my view:
Stay exposed to the trend, because the demand curve is durable and supply is still the gating factor
Avoid leverage, because crowded trades punish leverage fast
Don’t over-concentrate, because even the best names can drop hard on timing issues
Favor companies that capture revisions, because the winners will be the ones that surprise estimates, not just meet them
This is a “yes, but” year:
Yes, you want exposure - but you want it sized like an investor, not like a speculator.
Enjoying the content? Don’t miss out on more exclusive insights and analyses. Subscribe now and stay updated.
Our Top Picks for 2026:
This is where we get specific.
Semiconductors will likely remain one of the best places to be positioned in 2026 - but stock selection will matter more than ever. Forward multiples are still elevated, the trade is crowded and expectations are high. That means the market won’t reward “good stories.”
It will reward companies that capture the next incremental dollar of AI infrastructure spend - and, just as importantly, companies that can keep executing even if the cycle cools for a quarter or two.
So the goal here isn’t to chase whatever is hottest.
It’s to own the names with the cleanest mix of:
structural tailwinds (AI infrastructure, rising complexity, tight supply);
pricing power and margin expansion; and
the highest probability of positive revisions over the next 12-18 months.
Below are our top picks for 2026, ranked by conviction.
