Key Takeaways
AI semiconductors account for more than 95% of an AI server rack’s content value, according to the SIA-Deloitte ‘Powering AI’ report. A virtual teardown of a single rack surfaces over 4,500 packaged chips, and chips absorb more than half of the capital spending needed to build and run an AI data center. The report projects annual AI data-center chip revenue of $1,200 billion by 2028. This piece breaks the numbers down by chip layer and cross-reads them against IDC’s memory-shortage scenarios.
- Chips: over 95% of a leading AI server rack’s content value and over 50% of data-center capex
- AI data-center chip revenue projected at $1.2 trillion annually by 2028 — a near-tenfold jump in four years
- Over $4 trillion in infrastructure investment through 2028, with up to $2.8 trillion going to semiconductors
- IDC: memory supply constraints point to shrinking phone and PC shipments — and rising prices
4,500 chips. That is what falls out when you tear down a single state-of-the-art AI server rack. For AI semiconductors, no other number frames the story this well.
The Semiconductor Industry Association (SIA) and Deloitte published the ‘Powering AI’ report on June 1, built around a virtual teardown — itemizing the parts and value of one rack instead of waving at the whole data center. The verdict is lopsided. More than 95% of the rack’s content value sits in chips.
This piece does not re-run the press release. It breaks the report’s numbers down by chip layer, runs the arithmetic the announcement skips, and cross-reads the result against IDC’s memory-shortage scenarios to map where the bottlenecks and the beneficiaries are.
AI Semiconductors Are 95% of the Rack — the Teardown
The methodology is simple and rare at once. Rather than sizing the data-center market top-down, the report dismantles the basic unit — one server rack — and asks where the value actually sits.
The answer concentrates hard. Chips account for more than 95% of a leading AI server rack’s content value, and more than half of the total capital expenditures required to build and operate an AI data center.
The key here is what ‘data-center investment’ really buys. Not land, not buildings, not even power gear — the center of gravity of the spending is chip procurement.
What $1.2 Trillion Means — Tenfold in Four Years
The market projection is the headline with the longest shadow. Annual revenue for semiconductors used in AI data centers could reach $1.2 trillion by 2028, a near-tenfold increase over the last four years.
The comparison is the striking part. That figure surpasses total global semiconductor sales from 2025 — across all end uses — by more than 50%. Put simply, a single application is projected to outgrow what the entire industry sold for everything.
The investment math points the same way. The report expects over $4 trillion to flow into new data-center infrastructure through 2028, of which up to $2.8 trillion will be spent on semiconductors.
Three Layers in the Rack — It’s Not Just Logic
The report’s second message is about composition. AI infrastructure does not run on accelerators alone; it runs on a full stack of three chip layers.
Advanced logic covers AI accelerators, ASICs, FPGAs, CPUs, DPUs, and networking chips. Memory spans HBM, DRAM, SRAM, and NAND, while analog and foundational chips — power, transceivers, controllers, sensors — hold the floor underneath.
Our read — The table below is an original breakdown that overlays IDC’s supply data on the report’s three-layer chip taxonomy. The goal is to contrast each layer’s role with its current bottleneck signal.
| Layer | Representative parts (SIA-Deloitte) | Role | Bottleneck signal (IDC cross-read) |
|---|---|---|---|
| Advanced logic | AI accelerators, ASICs, FPGAs, CPUs, DPUs, networking chips | Compute and data movement | Demand-driven layer — demand signals dominate supply signals |
| Memory | HBM, DRAM, SRAM, NAND | Holding models and data | 2026 DRAM supply +16%, NAND +17% — below historical norms |
| Analog & foundational | Power chips, transceivers, controllers, sensors | Power, signal, control | Named by the report as an essential full-stack layer |
Why this composition matters is plain. If most of the value is silicon, a jam in any of the three layers jams the whole rack.
Memory Screams First — IDC’s Scenarios
The supply signal cracked first in memory. IDC expects 2026 DRAM supply growth of 16% year-on-year and NAND of 17% — both below historical norms.
When supply tightens at the three manufacturers IDC names — Samsung Electronics, SK hynix, and Micron — the squeeze rolls onto chip prices and consumer devices. IDC frames the fallout as a two-track scenario.
Our read — The table below reassembles the moderate and pessimistic scenario figures scattered across IDC’s blog into one view. Every cell is IDC’s own number, verbatim.
| Metric (2026) | Moderate scenario | Pessimistic scenario |
|---|---|---|
| Smartphone market growth | -2.9% | -5.2% |
| Smartphone ASP | +3-5% | +6-8% |
| PC market growth | -4.9% (vs prior -2.4% forecast) | -8.9% |
| PC ASP | +4-6% | +6-8% |
The two scenarios agree on direction. Shipments fall and prices rise; only the magnitude differs.
The Reverse Math — Three Numbers the Release Doesn’t Say
Restating published figures is not analysis. Make the published figures compute against each other, and implications surface that no press release prints.
Our read — The three numbers below are derived solely from figures cited in this article. None of them appears directly in any coverage we found.
- 2025’s implied ceiling for total global semiconductor sales: roughly $800 billion. If $1.2 trillion in 2028 exceeds 2025’s all-purpose total by more than 50% (SIA-Deloitte), the reverse math caps that 2025 total below $800 billion.
- The implied starting point four years ago: around $120 billion a year. If $1.2 trillion is ‘a near-tenfold increase over the last four years’ (SIA-Deloitte), the base sat near one-tenth of the destination.
- PC exposure to the memory shock runs about 1.7x smartphones. Moving from IDC’s moderate to pessimistic case widens the decline by 2.3 points for phones (-2.9% to -5.2%) and 4.0 points for PCs (-4.9% to -8.9%) — the same supply shock rocks PCs harder.
Our read: put the three together and the conclusion sharpens. While the market’s center of gravity relocates within four years, the cost shock lands asymmetrically across devices.
Our Read — a Map of Bottlenecks and Beneficiaries
SIA president and CEO John Neuffer put the thesis in one line in the release. “The future of AI will be determined by the full array of semiconductor technologies that power it.”
Our read — Connect the report to IDC’s data and one path draws itself. Each step leans on a cited figure.
- Nearly all of a rack’s content value is silicon, and more than half of capex goes to chips — ‘AI infrastructure investment’ is, in substance, semiconductor procurement (SIA-Deloitte).
- That silicon is not one product but three layers — logic, memory, and analog — so the bottleneck can form in any of them (SIA-Deloitte).
- The layer flashing the clearest supply constraint today is memory, with 2026 DRAM and NAND growth below historical norms (IDC).
- If AI data centers absorb tight memory supply, IDC’s scenarios follow: fewer phone and PC shipments at higher prices — the channel through which AI infrastructure costs pass to consumers (IDC).
Our read: the implication is the value chain, not a single accelerator ticker. Memory and the analog-foundational floor are bound to the same demand cycle as the logic layer.
The caveats deserve equal billing. Our read: this is a report from an industry association and a consultancy, which carries a structural incentive to size semiconductors generously. The revenue projection is a forecast, not a print, and it hangs on both supply and demand holding up.
Two trackers are worth a slot on the watchlist. Whether 2026 DRAM and NAND supply growth actually stays below historical norms, as IDC expects, and whether phone and PC average selling prices move into IDC’s scenario bands. If both tilt pessimistic, the chip-centric value shift this report describes will show up in consumer prices first.
Q1. How much of an AI server rack’s value comes from semiconductors?
More than 95% of a leading AI server rack’s content value, according to the SIA-Deloitte virtual teardown. A single rack holds over 4,500 packaged chips, and semiconductors take more than half of the capex needed to build and operate an AI data center.
Q2. How big will the AI data-center chip market get?
The report projects annual revenue of $1.2 trillion by 2028 for semiconductors used in AI data centers. That is a near-tenfold increase over the last four years and more than 50% above 2025’s total global semiconductor sales.
Q3. What does the memory shortage mean for consumers?
IDC expects 2026 DRAM and NAND supply growth below historical norms, with smartphone shipments shrinking by up to 5.2% and PCs by up to 8.9% in its pessimistic scenario. Average selling prices rise by as much as 8% in the same case.
Half of US AI Data Centers Stalled in 2026
KOSPI Hits 6,388 Record: Goldman Raises 12M Target to 8,000
References
- SIA — Powering AI report (2026-06-01) — landing page for the AI data-center semiconductor ecosystem report
- Semiconductor Digest (2026-06-01) — chips account for over 95% of a leading AI server rack’s content value and over 50% of build-and-operate capex
- Semiconductor Digest (2026-06-01) — $1.2 trillion annual AI data-center chip revenue by 2028; near-tenfold in four years; over 50% above 2025 total sales
- DQ India (2026-06-01) — over 4,500 packaged chips per rack; $4T+ infrastructure investment through 2028 with up to $2.8T on semiconductors
- DQ India (2026-06-01) — the three chip layers: advanced logic (accelerators, ASICs, FPGAs, CPUs, DPUs, networking), memory (HBM, DRAM, SRAM, NAND), analog and foundational (power, transceivers, controllers, sensors)
- Semiconductor Digest (2026-06-01) — John Neuffer, SIA president and CEO: “The future of AI will be determined by the full array of semiconductor technologies that power it.”
- IDC (2025-12-18) — 2026 DRAM supply +16% and NAND +17% year-on-year, below historical norms
- IDC (2025-12-18) — smartphone scenarios: moderate -2.9% with ASP +3-5%; pessimistic -5.2% with ASP +6-8%
- IDC (2025-12-18) — PC scenarios: moderate -4.9% (vs prior -2.4%) with ASP +4-6%; pessimistic -8.9% with ASP +6-8%; top three makers named as Samsung Electronics, SK hynix, Micron
Stay Ahead of the Curve
Get the latest insights on AI, tech, and markets — delivered in both Korean and English.
Jugger | Backend Developer
A backend engineer working in enterprise solutions software since June 2021. The hands-on perspective from system design, API implementation, and production operations directly informs this site’s
AI, IT, and cloud infrastructure coverage.
For topics outside direct professional experience—blockchain and macroeconomics—analysis is built on primary-source citations such as SEC filings, whitepapers, institutional research, and government
statistics. Speculation is clearly separated from verified fact, and unverified claims are not published.
AI tools assist with drafting, but every article is reviewed, revised, and verified before publication. Errors are corrected with visible notes when discovered.
Contact: jjunggulee@gmail.com