$1.32 trillion. That’s the number Gartner slapped on worldwide semiconductor revenues for 2026—a 64% leap from 2025’s $805 billion.
And it’s not some vague optimism. Memory alone? Expected to triple, from $216 billion to $633 billion. Non-memory chips chug along at 17% growth, but let’s be real: AI’s the beast pulling this cart.
“Amid high demand for datacentre networking and power, and memory price inflation (memflation), the semiconductor industry is projected to achieve a third consecutive year of double-digit growth in 2026 – a milestone that underscores the sector’s vital role in the AI technology stack,” says Rajeev Rajput, Senior Principal Analyst at Gartner.
Rajput’s words hit hard. But here’s the twist no one’s yelling about yet: this isn’t your garden-variety cycle. It’s memflation—Gartner’s term for memory prices ballooning like they haven’t since the crypto mining madness of 2018. DRAM up 125%. NAND flash? 234%. Relief? Not till late 2027.
Look, I’ve chased these forecasts for years. Back in ‘18, memory spiked on Bitcoin fever, then crashed when rigs went dark. Today’s different. AI datacenters don’t flip off like miners. They’re gobbling HBM stacks for GPUs, custom ASICs—architectural behemoths that demand bandwidth memory no consumer SSD ever will.
Why Memory Prices Are Set to Triple in 2026?
Simple supply crunch meets AI gluttony. Hyperscalers—think Nvidia’s gravy train customers—are pouring cash into infrastructure. Gartner sees their AI spend jumping over 50% next year. GPUs, yes. But also weird custom silicon: Google’s TPUs, Amazon’s Trainium, all thirsting for high-bandwidth memory.
Capacity lags. TSMC’s CoWoS packaging? Bottlenecked. Samsung and Micron ramping HBM3E, but fabs can’t keep pace. Prices soar—profitable for suppliers, nightmare for anyone not riding the AI wave.
Here’s the table Gartner flashed:
| Segment | 2025 | 2026 | 2027 |
|---|---|---|---|
| Memory | 216.3 | 633.3 | 748.1 |
| Non-memory | 589.0 | 686.9 | 806.4 |
| Total | 805.3 | 1,320.2 | 1,554.5 |
(Billions USD. Source: Gartner, April 2026. Note: dates seem off—probably 2025 forecast.)
Non-memory grows steady, sure. But memory’s the fireworks. And Rajput warns:
“Memflation will destroy, or at least delay, non-AI demand into 2028, to varying degrees depending on the application.”
Destroy. Harsh word. Auto chips? PCs? IoT gadgets? They’ll wait in line while datacenter giants hoard.
But wait—my take, absent from Gartner’s spin: this echoes oil shocks of the ’70s. Short-term gouge funds massive capex (new fabs, anyone?), then floods the market. By 2029, we’ll see memory prices crater as AI training plateaus and inference shifts to edge. Prediction: semis stay hot, but memory swings wilder than ever.
How AI Chips Claim 30% of the Pie?
AI semis hit 30% of total revenue in 2026. That’s $396 billion, easy math. Driving it? Hyperscaler buildouts. Nvidia’s Blackwell? Sold out. AMD’s MI300X? Snapped up. Custom stuff from Broadcom, Marvell—quietly raking it in on networking ASICs.
Architectural shift here: gone are monolithic GPUs for everything. Now, disaggregated clusters—racks of accelerators talking via ultra-fast Ethernet or InfiniBand. That chews Ethernet switches (Broadcom wins), power delivery (Infineon), and yeah, stacks of DDR5 or HBM.
CIOs, listen up. Rajput’s advice: dodge long-term contracts locking bad prices past 2027. Smart. Because here’s the rub—memflation’s profound, but not perennial, per Gartner. Prices peak H1 2026, then moderate.
Skeptical? Me too. Suppliers like Micron boast record margins now. Why rush relief? Still, history says they flood supply eventually.
One short para for punch: Non-AI suffers most.
Consumer electronics? Stalled. Industrial? Delayed. Even servers pivot to AI-first designs, squeezing legacy x86.
And power. Datacenters guzzle it—semis for PSUs, cooling chips, all booming. Networking? 400G-to-800G ramps demand retimers, PHYs. It’s a full-stack frenzy.
Will Memflation Backfire on the Industry?
Short answer: yes, for fringes. But core AI? Bulletproof.
Think architecture: AI models balloon—GPT-5 rumors say trillions of params. Training needs exaflops of compute, petabytes of fast memory. Inference? Millions of chips running hot.
Non-AI gets rationed. Smartphones ship with last-gen NAND. PCs? Maybe. But EVs, robots—they’re “AI-adjacent,” so they sneak in.
Bold call: by 2028, Intel, Qualcomm pivot hard to AI-PCs and edge inference, salvaging non-memory. Memory kings (Samsung, SKHynix) feast now, pivot to HBM later.
Gartner’s rosy on total growth—third year double-digit. But that 2027 total? $1.55T, slowing to 18%. AI sustains, memflation fades.
Wrapping the why: semis aren’t just chips. They’re the spine of compute. AI rewires it all—more parallel, more memory-bound. That’s the shift.
🧬 Related Insights
- Read more: Broadcom’s 200T AI Backbone: The Chips Lighting Up Gigawatt Superclusters
- Read more: Asia’s AI Factories: Power Grids, Wallets, and Wars Trump Code
Frequently Asked Questions
What is memflation in semiconductors?
Gartner’s term for explosive memory price hikes—DRAM 125%, NAND 234% in 2026—driven by AI datacenter demand outstripping supply.
How much will global semiconductor revenues grow in 2026?
64%, per Gartner, from $805B in 2025 to $1.32T, with memory tripling on AI and power/networking surges.
Will AI continue driving semi growth after 2026?
Yes—expected to claim 30% of revenues in 2026 and keep leading, though memflation delays non-AI demand till 2028.
