The U.S. data center construction boom is facing political headwinds, but Goldman Sachs believes the cycle itself is far from over.

Market sentiment deteriorated sharply after Gina Hinojosa, Texas's governor-elect, called for an immediate moratorium on data centers. The Nasdaq fell 6% in a single week, the AI Momentum Index recorded its fifth-largest decline in history, and data center credit default swap premiums hit a record high.

Yet Goldman Sachs went against the tide by raising its forecasts, lifting its 2026 U.S. data center construction spending estimate from $67 billion to $80 billion, its 2027 estimate from $87 billion to $100 billion, and issuing a first-time 2028 forecast of $125 billion.

In two newly released research reports, Goldman Sachs pointed out that the U.S. data center project pipeline is far larger than the fragility implied by market concerns — 86GW of projects are already under construction, with another roughly $900 billion in high-confidence pending projects, and construction spending growth had already accelerated to 90% year-over-year before entering the election cycle.

Goldman Sachs engineering and construction analyst Adam Bubes and IT hardware analyst Michael Ng reached the same conclusion: recent policy-level uncertainty has caused delays in some project activity but will not reduce the overall market size.

86GW Already Broken Ground, Project Scale Far Exceeds Current Spending Levels

The acceleration of this construction cycle was already evident before the election.

According to U.S. Census Bureau data cited by Goldman Sachs, U.S. data center construction spending growth accelerated from 55% year-over-year in June 2024 to 90% year-over-year in September, with the number of new project starts in the third quarter "increasing noticeably," reaching an annualized scale approaching $100 billion.

Based on the Aterio project tracking database (data as of the end of October), Bubes divided projects under construction into three tiers: projects that have broken ground and are under construction, high-probability announced but not yet started projects, and all remaining projects, valuing each tier at approximately $3.5 billion in construction cost per GW (excluding servers, racks, GPUs, and networking equipment).

Among the approximately 86GW of projects that have broken ground, the corresponding construction spending is about $300 billion — roughly six times the approximately $50 billion in data center construction spending for all of 2025 — and more than $200 billion of that has not yet been invested.

Goldman Sachs believes that adding high-probability announced projects brings the total project scale to approximately $900 billion; if all medium- and low-probability projects are included, the theoretical opportunity size expands to approximately $1.9 trillion.

Capacity-Side Data Strongly Corroborates the Construction Side

From the capacity supply perspective, Michael Ng reached a similar conclusion.

The U.S. currently has 57GW of total active data center capacity, with another 9GW under construction, expected to reach 65GW by the end of 2026. Further, based on the 78GW of projects under construction tracked by Aterio, net capacity additions between 2027 and 2030 are expected to double total U.S. capacity to 143GW.

Ng's core judgment is that "low-risk" capacity already under construction alone is sufficient to support Goldman Sachs's own forecasts. Converted to IT load at a power usage effectiveness (PUE) of 1.5, projects under construction correspond to net additions of 22/24/5 GW-IT in 2027/2028/2029 respectively, closely matching Goldman Sachs's forecasts of 20/24/23 GW-IT.

In other words, Goldman Sachs's 2027 and 2028 capacity forecasts have effectively already materialized, and only 2029 and beyond need to rely on announced but not yet started projects for support.

Ng also flagged an important caveat: Aterio's definition of "under construction" covers early-stage site work such as site clearing, grading, and pipeline laying, and developers often begin such work before signing tenant agreements. If a moratorium is imposed, projects that have broken ground but have not yet obtained power service agreements or environmental permits could still be affected.

Texas Moratorium: Delay Rather Than Termination

The core political risk currently in the market's focus comes from Texas.

Among the five states with cautious stances on data centers (Texas, Ohio, Nevada, Georgia, and Michigan), these five states together account for approximately 42% of capacity under construction and high-probability projects, with Texas alone accounting for 26% of cumulative net new capacity from 2026 to 2030.

However, Goldman Sachs drew an important distinction regarding the nature of the moratorium. In the report, Ng noted that "a moratorium is not a ban or cancellation," and its impact lies in delaying in-service timelines (which in some cases could trigger contract cancellations), raising financing costs, reducing tax incentives, and increasing infrastructure capital expenditures.

Bubes further offered a view that the market has underestimated: even before anyone voted, the 2028 construction schedule was already too aggressive — theoretically, approximately 55GW of capacity would come online simultaneously that year, "far exceeding the pace of capacity coming online historically."

Goldman Sachs had originally expected some projects to experience natural slippage, and Texas's newly elected governor Gina Hinojosa's moratorium merely gives such slippage a political rationale rather than adding an incremental reduction.

Goldman Sachs economists also expect state-level moratoriums to be conditional and relatively short-lived, and any new federal legislation would still require bipartisan Senate support and face veto risk.

As for Texas, according to Wells Fargo analyst Jerry Revich's assessment, the Texas legislature's regular session runs from January 12, 2027, to May 31, 2027, and this period may also roughly match the duration of Texas's moratorium. The next key milestone is the Texas data center audit report due on December 10.

Hyperscaler Spending Shows No Signs of Retreat

The demand side also shows no signs of loosening. Goldman Sachs data shows that hyperscaler capital expenditures are expected to reach approximately $800 billion in 2026, with FactSet consensus forecasting approximately $1.3 trillion in 2027, and the combined revenue backlog of the three major public cloud service providers already exceeding $1.7 trillion.

According to data cited by Revich, H100 compute rental prices rose 26% year-over-year, GB200 rental pricing rose 50% year-to-date and a further 12% in the past month, indicating that compute demand remains robust and the market continues to pay a premium for computing resources.

At the stock level, Ng named IREN as having the highest Texas exposure within his coverage, with 73% to 87% of its data center capacity located in the state; Celestica (CLS) and Arista (ANET) could be affected through servers and data center switches; Digital Realty (DLR) and Equinix (EQIX), with their primary and secondary metropolitan area asset layouts, have extremely limited exposure to development-intensive markets.

TAM May Persist, but Highly Leveraged Developers Face Material Risk

Goldman Sachs's conclusion is relatively cautious, and its exact words were:

The project scale we track is fairly ample and may support continued growth, but we currently lack the visibility to provide a complete endorsement of the future path.

This statement precisely captures the dual nature of the current situation: the sheer scale of 86GW under construction, the thickness of hyperscaler backlogs, and the extreme pessimism that has emerged amid widespread market bearishness all point to a construction cycle that remains resilient.

But Ng's risk warning cannot be ignored: delays mean higher financing costs, lost tax incentives, increased capital expenditures, and in some cases contract cancellations.

For hyperscalers sitting on $1.7 trillion in backlogs, time risk may be merely a rounding error; but for a highly leveraged developer issuing bonds maturing in 2028 at a 12.5% interest rate, it could be a matter of life and death.

The overall market size may be able to endure this period of policy pain, but some borrowers that entered with high leverage may not survive until the clouds part and the moon shines.