Key Takeaways
- Delay is already the base case. JLL reports that 57% of data center projects had a construction delay of three months or more in 2025. The statistic does not say what caused the delays.
- Shortages are measured; tariff delays are not. TrendForce puts power-management IC lead times at 35–40 weeks (from 21–26) and PCB and CPU lead times near one year. No public dataset reviewed isolates delay days caused by tariffs.
- Tariff scope is narrower than the headlines. The 25% U.S. duty that took effect on 15 January 2026 applies to specified advanced computing chips and expressly excludes covered products imported for U.S. data center use. Metal duties depend on product, annex and origin.
- Regional spread is large. Transformer lead times read at roughly 100 weeks in EMEA, 43 in the Americas and 20 in APAC. NextMSC calculates that the EMEA figure exceeds the ~78-week window of an 18-month, 50 MW build.
- In India, national power balance hides local queues. FY2025–26 energy shortfall was 0.03% nationally, yet four metros hold about 90% of the ~1,530 MW operating stock and announced pipelines far exceed delivered capacity.
NextMSC assessment
On the evidence available at the 9 October 2026 cut-off, equipment and component lead times are the more directly documented cause of schedule slippage. Tariffs mainly change cost, sourcing and compliance workload. This is an evidence-weighted reading, not a causal ranking: no public source separates delay days by cause.
Executive Summary
Data center schedules are being squeezed from two directions. Tariffs add price and paperwork. Shortages add waiting time. The question for a developer or operator is which one moves the energization date.
JLL's 2026 outlook benchmarks a typical 50 MW facility at about 18 months to build, and notes that some material orders are placed up to 24 months ahead. Against that, average equipment lead time was 33 weeks in 2026, about 50% above pre-2020 levels, and 42 weeks in the United States (83% above 2019). Scaled operators hold 6–12 months of strategic inventory. These are the numbers that shape data center infrastructure procurement.
What the evidence can and cannot say
- Can say: lead times for electrical equipment and server components are long, quantified and rising in the cases reviewed; tariff instruments are documented as duties, exemptions and annexes.
- Cannot say: what share of any project's delay came from a tariff versus a shortage. The public sources reviewed report neither measure, so the comparison below is evidence-weighted, not statistical.
Projects delayed 3+ months in 2025 (JLL).
Year-on-year growth in AI-enabling goods trade, H1 2026, 47% of global merchandise trade growth (WTO, 8 Oct 2026).
2026 AI server shipment growth forecast, against ~13% for general servers (TrendForce).
The WTO and TrendForce figures point the same way. Demand for AI-related hardware is absorbing supplier capacity, and TrendForce attributes its general-server downgrade to extended component lead times and supplier allocation that favors AI products.
Delay Driver Scorecard
NextMSC compared five drivers on the strength of public evidence and the channel through which each acts. The weighting is editorial.
| Driver | Main channel | Quantified evidence | Evidence strength | Relative weight |
|---|---|---|---|---|
| Critical electrical equipment lead times | Schedule and cost | 33 weeks average globally, 42 weeks in the U.S.; transformer ~100 / 43 / 20 weeks (EMEA / Americas / APAC) | Strong for the aggregate, approximate for regional values | Very high |
| Server component allocation | Schedule and availability | PMIC 35–40 weeks, BMC IC 21–26 weeks, PCB and CPU near one year; 2026 general-server growth ~13% from ~20% | Medium-high (analyst forecast) | Very high |
| Grid connection and firm power | Schedule | National supply near balance (FY2025–26 energy shortfall 0.03%); site-level connection timelines are not published | Medium (qualitative at site level) | Very high |
| Approvals, policy and land readiness | Schedule | Policy incentives and MoUs exist; they are not evidence of commissioned capacity | Low-medium | High, market-specific |
| Tariffs on imported materials and equipment | Cost; schedule only indirectly | 25% duty on a narrow chip group with a U.S. data center exemption; metal duties set by product, annex and origin | Strong on rule text, none on delay effect | Medium, scenario-dependent |
Source: JLL, TrendForce, White House, Ministry of Power / CEA, and NextMSC analysis.https://www.nextmsc.com/industry-deep-dive/tariffs-vs-component-shortages
Shortage Evidence: Lead Times Sit on the Critical Path
Electrical and mechanical plant is ordered early and installed late, so its lead time can decide the finish date even when the building shell is on schedule. JLL's regional chart shows how uneven that exposure is, from transformers and generators to cooling equipment.
Weeks, approximate values read from a JLL bar chart (2026 outlook); not a published numeric table
- EMEA
- Americas
- APAC
Dashed line marks ~78 weeks, an 18-month build window (JLL benchmark for a 50 MW facility).
Source: JLL 2026 Global Data Center Outlook and NextMSC analysis.https://www.nextmsc.com/industry-deep-dive/tariffs-vs-component-shortages
Lead time against the build window
To make the regional spread comparable, NextMSC expressed each approximate lead time as a share of an 18-month (about 78-week) build. A ratio above 100% means the item takes longer to arrive than the whole benchmark build lasts, so it has to be ordered before construction starts.
| Equipment | EMEA | Americas | APAC | Reading |
|---|---|---|---|---|
| Transformer | 128% | 55% | 26% | Longer than build window (EMEA) |
| Generator | 105% | 65% | 53% | Longer than build window (EMEA) |
| Switchgear | 68% | 55% | 37% | Over half the window |
| Battery and UPS | 38% | 45% | 28% | Well inside window |
| Bypass panel (ATS) | 24% | 45% | 37% | Well inside window |
| PDU and STS | 18% | 53% | 19% | Well inside window |
| Chiller | 62% | 45% | 27% | Over half the window |
| RDHx / direct-to-chip cooling | 38% | 24% | 19% | Well inside window |
| Cooling tower | 29% | 18% | 27% | Well inside window |
| CRAHU / FWU | 29% | 14% | 21% | Well inside window |
NextMSC calculation: approximate weeks ÷ 78, rounded to the nearest whole percent. Weeks are approximate chart reads, so ratios are indicative only. The 18-month benchmark is JLL's figure for a 50 MW facility; actual builds vary with design, permitting and site.
Source: JLL 2026 Global Data Center Outlook and NextMSC analysis.https://www.nextmsc.com/industry-deep-dive/tariffs-vs-component-shortages
Server components: allocation, not only capacity
The silicon side tightened in parallel. TrendForce, in an analysis dated 15 April 2026, says demand for higher-power AI servers is pulling 8-inch BCD process capacity toward AI-focused PMICs, while limited mature-node foundry capacity and an AI-first allocation policy constrain BMC chips. Its effect is to push general-purpose server supply down the queue.
Weeks; bars show the published range. PCB and CPU lead times are stated only as "nearly one year", with no prior value.
- Prior range
- 2026 range
Solid bar = low end of the range; lighter bar = high end.
Source: TrendForce (15 April 2026) and NextMSC analysis.https://www.nextmsc.com/industry-deep-dive/tariffs-vs-component-shortages
Both ranges moved by the same absolute amount at each end: PMIC lead times rose 14 weeks and BMC lead times 10 weeks. TrendForce's revised view is that general server shipments grow about 13% in 2026, against an earlier expectation near 20%, a cut of roughly 7 percentage points. AI server shipments are projected to grow about 28%.
NextMSC calculation from the published ranges. These are analyst forecasts and supplier assessments, not observed averages or final shipment results.
Tariff Evidence: Real Cost Pressure, Unmeasured Delay
The tariff record is legal text rather than delay data. That is useful, because it shows exactly what is and is not covered. In the United States, trade in semiconductors and servers is where most readers expect the heaviest effect, yet the instrument is narrower than the common shorthand.
| Instrument | Date | What it covers | Key carve-out or limit | Channel to schedule |
|---|---|---|---|---|
| Advanced computing chips (Section 232 action) | Signed 14 Jan 2026; effective 15 Jan 2026 | 25% ad valorem on specified covered products | Expressly excludes covered products imported for use in U.S. data centers, plus other listed exceptions | Cost where not exempt; end-use documentation. No measured delay effect. |
| Aluminum, steel and copper regime | June 2026 proclamation; changes effective 8 Jun 2026 | Covered articles and derivatives, set by annex | Treatment varies by product, HTS annex, metal origin and country | Cost for some enclosures, racks, busbars and cables; classification risk. |
| Reciprocal tariff order | 2 Apr 2025 | Baseline reciprocal rates | Items already under Section 232 or listed in Annex II are excluded; other duties may still apply | Layering with other measures complicates cost modeling. |
| Harmonized Tariff Schedule, 2026 Revision 20 | Published 28 Sep 2026 | Official 10-digit classification and duty lookup | Six-digit HS headings are not enough for a final duty estimate | Classification decides which measure applies. |
Source: White House proclamations, USITC Harmonized Tariff Schedule and NextMSC analysis.https://www.nextmsc.com/industry-deep-dive/tariffs-vs-component-shortages
How a tariff can still become a delay
A tariff becomes a schedule problem only when it changes where a part is bought or how fast it clears customs: a supplier switch after an order is placed, a reclassification dispute, or missing end-use paperwork. The sources reviewed document the rules but not a count of projects held up this way, so NextMSC treats tariffs as an indirect amplifier of the shortage problem.
Jurisdiction check
U.S. tariff actions are used here as a case study of policy uncertainty in global supply chains. They do not set Indian import duty. For goods landing in India, applicable duties, exemptions and origin conditions must be checked on ICEGATE / CBIC and in current Indian notifications.
Power and Grid: A Third Constraint Running in Parallel
National electricity data in India no longer points to a supply gap. FY2025–26 energy requirement was 1,707,493 MU against 1,706,985 MU available, a 0.03% shortfall, with peak demand of 245,444 MW and 245,416 MW met (a 28 MW gap).
Percent of requirement not met (shown as a positive shortfall)
- Peak demand shortfall
- Energy shortfall
Source: Ministry of Power, Government of India (Power Sector at a Glance) and NextMSC analysis.https://www.nextmsc.com/industry-deep-dive/tariffs-vs-component-shortages
Computed by NextMSC from the published requirement, availability, peak demand and peak-met figures. For FY2024-25 the published MW values (a 2 MW gap on 249,856 MW) give about 0.001%, which the supply table lists as 0.01% after rounding; the chart uses the computed value.
The takeaway is about geography. National balance says nothing about substation headroom, a dedicated feeder or an energization date at one site. The same JLL benchmark that shows long lead times for transformers and switchgear also explains why a grid connection can be late even in a surplus system: the equipment that delivers power to a site is itself on a long queue.
India Lens: Concentrated Stock, Fast-Moving Pipeline
CBRE estimates about 1,530 MW of operating capacity at September 2025, and JLL puts India's data center market at about 1.6 GW in mid-2026, with a forecast of 6 GW by 2029. The two figures come from different dates and methods and are not treated as one series. In H1 2026 JLL counts 85 MW delivered and 101 MW absorbed, of which 82% was pre-committed hyperscale capacity.
| Hub (state) | Share of ~1,530 MW stock (Sep 2025) | Implied MW | Share of H1 2026 new supply | Energy not supplied, Apr 2025–Feb 2026 | Pipeline signal |
|---|---|---|---|---|---|
| Mumbai / Navi Mumbai (Maharashtra) | 53% | ~811 | 48% | 0.005% | 500 MW park reported as planned (secondary source) |
| Chennai / Siruseri (Tamil Nadu) | 20% | ~306 | 50% | 0.009% | High new-supply share; policy applicability stated to 31 Mar 2026 |
| Delhi-NCR (cross-state) | 10% | ~153 | n/a | 0.017% (UP only) | UP reports 644 MW proposed state-wide (Jun 2025) |
| Bengaluru (Karnataka) | 7% | ~107 | n/a | 0.013% | No confirmed 2025–26 MW project in source set |
NextMSC calculation. Implied MW multiplies CBRE's city share by the 1,530 MW national figure and is indicative, not state-reported capacity. Energy not supplied is the state's reported MU not supplied divided by its energy requirement, Apr 2025–Feb 2026 (partial year, PIB / Ministry of Power). City shares are not state shares; Delhi-NCR spans several jurisdictions. JLL's two new-supply shares sum to 98%, with the remainder in other locations or rounding.
Source: CBRE India, JLL India, PIB / Ministry of Power and NextMSC analysis.https://www.nextmsc.com/industry-deep-dive/tariffs-vs-component-shortages
What the hub data suggests
- Four hubs account for about 90% of the operating stock, so a single constrained substation or equipment queue can affect a large share of capacity.
- Chennai took about 50% of H1 2026 new supply from 20% of the installed base, which is where simultaneous commissioning could stress local power plant and generators.
- State energy-not-supplied rates are all below 0.02% of requirement in the hub states shown. A statewide energy balance does not measure interconnection capacity at a site.
- Telangana's state portal reports about 200 MW of live IT capacity against about 4.7 GW committed or early-stage. That is a pipeline indicator, not operating stock.
Project tracker: status matters more than size
Announcements, MoUs, groundbreakings and operating benchmarks measure different things and are not additive.
| Project / benchmark | State | Capacity (MW) | Stated investment | Status | Date |
|---|---|---|---|---|---|
| TCS HyperVault campus | Telangana | up to 1,000 | up to ₹70,000 crore | Announced / planned | 5 Sep 2026 |
| CtrlS AI cluster | Telangana | 400 | ₹10,000 crore | MoU / announced | 22 Jan 2025 |
| STT Global Data Centres campus | Telangana | 100 | ₹3,500 crore | MoU / announced | 18 Jan 2025 |
| UPC Renewables–Volt AI data center | Telangana | 100 | ₹5,000 crore | MoU / announced | 22 Jan 2026 |
| Amazon data center, Bharat Future City | Telangana | not disclosed | not disclosed | Groundbreaking | 15 Jul 2026 |
| Google AI Hub, Visakhapatnam | Andhra Pradesh | not specified (“gigawatt-scale”) | US$15 billion, 2026–2030 | Construction phase announced | 28 Apr 2026 |
| Trentar / Dorf Ketal hyperscale park | Maharashtra | 500 | ₹30,000 crore | Planned (media report) | 6 Oct 2026 |
| Uttar Pradesh proposed pipeline | Uttar Pradesh | 644 | ₹21,342.9 crore | Proposed aggregate | Jun 2025 |
| India operational stock (CBRE) | India | ~1,530 | n/a | Operating benchmark | 30 Sep 2025 |
| India market estimate (JLL) | India | ~1,600 | n/a | Operating benchmark | mid-2026 |
MW and investment figures are as stated by each source and are not additive. "Up to" capacity is a ceiling. Google's USD 15 billion covers 2026–2030 and is not data center capex alone; no exact MW is disclosed. Amazon's separate ₹1 lakh crore invitation to invest by 2034 is not this facility's cost.
Source: Invest Telangana, Invest UP, Google, Economic Times, CBRE, JLL India and NextMSC analysis.https://www.nextmsc.com/industry-deep-dive/tariffs-vs-component-shortages
Implications by Stakeholder
Developers and EPC contractors
Long-lead electrical plant belongs on the earliest procurement schedule. JLL reports some orders placed up to 24 months ahead of need. Treat tariffs as a priced line item, and confirm classification and origin before a purchase order is fixed.
Operators and hyperscale tenants
Strategic inventory of 6–12 months is common at scaled operators, which gives larger buyers a buffer that smaller operators lack. General-purpose server supply is the segment TrendForce expects to be crowded out by AI priority.
Equipment and component suppliers
Allocation is shifting toward AI-focused products, with PMIC and BMC supply constrained at mature nodes. Regional capacity differences, such as the transformer spread across EMEA, the Americas and APAC, shape where orders are placed.
Utilities and policymakers
State incentives and MoUs lower development friction but do not guarantee a feeder or a delivery date. National adequacy figures are not a substitute for site-level connection schedules.
Outlook: What to Watch Next
NextMSC does not forecast a delay split. The indicators below would show whether the balance between shortages and tariffs is shifting.
1. Equipment lead-time readings
Track JLL's regional transformer, switchgear and generator values. A sustained EMEA transformer reading above the build window keeps early ordering mandatory.
2. Server component allocation
Watch TrendForce revisions to the ~13% general-server growth forecast and to the PMIC and BMC ranges.
3. Tariff text and exemptions
Follow changes to annexes, the data center end-use exemption, HTS revisions (Revision 20 published 28 September 2026) and Indian customs notifications.
4. Site-level energization evidence
Connection offers, substation milestones and revised delivery dates for named projects, which would allow an actual cause-by-cause comparison.
What would change this assessment
Project-level records showing customs holds, reclassification or rerouting as the binding constraint, or a published dataset that attributes delay days by cause. Either would justify revisiting the weighting in Table 1.
Methodology and Data Note
Approach
NextMSC compiled company, industry, government and intergovernmental sources into an evidence register and compared drivers on documented quantity, date and scope. Data cut-off is 9 October 2026. Planned refresh: quarterly. Each statistic maps to a source in the Source and Reference Sheet.
Current data
JLL 2026 outlook, TrendForce (15 Apr 2026), WTO (8 Oct 2026), Ministry of Power FY2025–26 actuals.
Announced or forecast
JLL's 6 GW India forecast by 2029, TrendForce shipment forecasts, company and state project announcements.
NextMSC analysis
Build-window ratios, implied hub MW, computed shortfall rates and the driver weighting.
Data conflicts and cautions
- CBRE (~1,530 MW, Sep 2025) and JLL (~1.6 GW, mid-2026) differ in date and definition; neither contradicts the other.
- JLL region-level lead times are read from a bar chart and are approximate; the 33-week, 42-week and 50% / 83% values are published aggregates.
- IEA scenarios separate electricity consumed by data centers (about 945 TWh by 2030) from generation needed to supply it (above 1,000 TWh in 2030); they are different measures.
- Tariff rates are classification-, origin- and date-specific. Nothing here is tariff or legal advice.
- Live re-verification of some source pages was pending at publication; the Source and Reference Sheet records the status of each claim.
About the Author
Debashree Dey
— Debashree Dey is Assistant Manager at Next Move Strategy Consulting, where she supports cross-vertical market content and communications across diverse industries for 6 years. Her professional background includes senior content writing, communications, and published manuscript authorship, with experience developing audience-focused business narratives and maintaining clear, consistent messaging. Her role supports research-led content development and editorial quality across NextMSC publications.
About the Reviewer
Supradip Baul
— Supradip Baul is an accomplished business consultant and strategist with over a decade of rich experience in market intelligence, strategy, technology, and business transformation. His work has included rigorous qualitative and quantitative analysis across multiple industries, helping clients shape investment decisions and long-term roadmaps. Earlier in his career, he was associated with Gartner, where he contributed to industry-leading reports and market share analyses. He has worked with leading global companies and holds an MBA with a dual specialization in Marketing and Finance.
Last updated 9 October 2026. Planned refresh: quarterly. Corrections and source challenges: Contact NMSC Research.