2028 — A pit dug by the US midterm election for the AI upstream supply chain

An election has prompted Wall Street to re-examine the pace of AI infrastructure development.

In November 2026, the results of the US midterm elections were released. The Democratic Party not only secured control of both chambers of Congress, but also won gubernatorial seats in Texas and Ohio — two states that happen to have the most concentrated data center construction in the United States.

Morgan Stanley's latest research report Buy the Drop: AI Investment Opportunities Post the US Midterms concludes that following the "Blue Wave" in the US, approximately 5GW of US data center construction capacity will be postponed from 2028 to 2029, equivalent to 20% of the country's new capacity in that year and 10% of global new capacity.

The direct consequence of this delay is that the year-on-year growth of new GW-scale data center capacity in the US will drop by 4% in 2028, before rebounding by 38% in 2029; on the global level, the year-on-year growth will hit 9% in 2028 and 20% in 2029.

The US midterm elections seem to have also "dug a pit" for the upstream AI supply chain — and it is named 2028. Morgan Stanley worries that the large number of postponed data center projects "may create an 'air pocket' in the tech supply chain in 2028, leading to weakened pricing power or inventory accumulation. There may even be order slowdowns, margin normalization and earnings downgrades, even if AI demand itself remains intact."

For the impact of the 2028 "air pocket", upstream cloud semiconductors will bear the brunt. Analysts immediately sharply cut target prices for targets including Asian cloud semiconductor supplier Aspeed, citing the risk of US project delays in the next 2-4 quarters; the legacy memory sector has also been downgraded, as the expectation of pricing power "peaking" has arrived earlier than anticipated.

Cumulative US Data Center Capacity Path Chart: Shows a downward revision from 110GW to 105GW in 2028

Why 2028? How Did the Election Dig This Pit

The direct trigger for the data center delays comes from the US midterm elections.

The Democratic Party won multiple gubernatorial positions including those in Texas and Ohio. Among them, the elected governor of Texas Gina Hinojosa's campaign platform includes suspending approval for new data centers until stronger protection measures are enacted in the Texas state legislature. The elected governor of Ohio Amy Acton has proposed a conditional suspension of new data centers and new permits, requiring projects to pay 100% of natural gas, water and electricity costs, and use union labor.

Analysts judge that projects under construction in 2027 will basically not be affected, as "newly elected governors are unlikely to disrupt projects that are already under construction." The delay risk is concentrated on 2028 projects that have not yet started.

The report points out that Texas and Ohio previously had a combined data center capacity exposure of approximately 4.9GW (on IT metrics) for 2028. This batch of projects will be pushed back due to the new approval process.

At the same time, interconnection applications with ERCOT (Texas power grid) have reached approximately 474GW, more than 5 times its historical peak electricity demand, of which data centers account for about 90% of new power requests. Power bottlenecks have long existed, and the election result has only superimposed policy uncertainty on top of that.

The "Pit" in the Upstream Supply Chain: Cloud Semiconductors Bear the Brunt

What does the 2028 air pocket mean for the upstream AI supply chain?

Morgan Stanley's Asia semiconductor team quickly followed up with the report Cloud Semis: Peak Growth, Less Upside, noting that cloud semiconductors are in the phase of "Peak Growth, Less Upside".

Aspeed is a core supplier of server motherboard management controllers (BMC), directly benefiting from the data center expansion cycle. Montage, a data center semiconductor company, has core products that are memory interconnect chips for server DRAM.

The target prices of both companies have been sharply cut:

Aspeed: Target price lowered from NT$24,000 to NT$14,417, a drop of approximately 40%

Montage: Target price lowered from HK$432 to HK$207, a drop of approximately 52%

Analysts explained that the core logic of the downgrade is: "The US team expects that approximately 20% of new US data center capacity added in 2028 may shift to 2029, which could affect cloud semiconductor suppliers' 2028 earnings."

"Pricing Power Peaking" for Legacy Memory: Arrives Earlier Than Expected

In the same period, Morgan Stanley's legacy memory research team released the report Old Memory: Peak Pricing Power? Reduce exposure., downgrading Winbond, GigaDevice, Powerchip, and Macronix from Overweight to Neutral, and Phison was further downgraded to Underweight.

The report clearly states that the analysis scope is limited to DDR3/4, SLC/MLC NAND and NOR, and does not cover the mainstream DDR5 or TLC/QLC NAND markets.

The core judgment is: "The pricing advantage of legacy memory is no longer sufficient."

The report lists three layers of pressure:

Deteriorated visibility of cloud spending

Weak consumer demand

Supply and demand will normalize by 2028

The most direct trigger is exactly the backward shift in the pace of US data center construction. "A potential air pocket in the tech supply chain in 2028 may lead to lower pricing power and inventory backlogs."

The report specifically emphasizes that the 2026 EPS of these companies is actually still being raised, and the core of the downgrade is "cycle inflection point management" rather than earnings collapse.

Stock prices can start to reprice even if the recent tight situation does not disappear immediately: narrowing shortages are enough to erode incremental pricing power before earnings peak.

Large Cloud Vendors: Capital Expenditure Remains Unchanged, But Timing Is Misaligned

Despite project delays, Morgan Stanley maintains its capital expenditure forecast for hyperscale cloud vendors: Google/Alphabet total US$375 billion / US$400 billion in 2027/2028, Meta US$225 billion / US$250 billion, Amazon US$355 billion / US$365 billion, Microsoft US$238 billion / US$298 billion, all kept unchanged.

The report believes these companies will still "build as much capacity as possible", because new large language model-powered products target billions of users and still deliver a return on capital of over 30%.

But "how much to spend" and "when the construction is completed" are two different things.

The report points out that the impact of delays is more about "postponement rather than cancellation". Some projects may shift from Texas and Ohio to more business-friendly states such as Tennessee and Louisiana. Overseas, Morgan Stanley predicts that 20GW of potential spillover demand may flow to Asia and the EMEA region.

Meta states its goal of achieving a 10x improvement in computing power efficiency by 2027; Google's TPU 8t and 8i deliver up to 2.7x higher training performance and 80% better inference cost-performance ratio respectively compared with the previous generation.

These efficiency gains are seen as a buffer against capacity constraints.

Silicon Valley's Confidence vs Wall Street's Caution

For this 2028 "air pocket", there is more than one voice in the market.

Investor Puru Saxena (Founder of AlphaTarget) posted on social media, offering a judgment that is broadly aligned with Morgan Stanley's but more concise: "The Democratic Party's landslide midterm election victory has indeed raised reasonable questions about AI data center construction, but the reality is more complex than the recent market reaction."

He pointed out that 2027 capacity is "basically intact", as most projects scheduled to go online next year are already under construction, and newly elected governors are not expected to interrupt projects that have already started.

"The AI infrastructure cycle is being extended, not ended," he wrote, citing the investment maxim: "It pays to be greedy when others are fearful."

This is consistent with Morgan Stanley's core stance — but Wall Street's analysis is more detailed: Companies in different links of the supply chain, different geographies, and with different capital structures face vastly different risks.

The scarcity of completed data center assets with guaranteed power supply will rise; the divergence in financing quality will deepen; neoclouds and new projects will face stricter scrutiny, while large cloud vendors will further consolidate their advantages with their scale and capital strength. Morgan Stanley notes:

Financing is still available, but the midterm elections should deepen quality divergence.

Who Will Fill the 2028 Pit

Morgan Stanley's entire analytical framework points to one core conclusion: AI demand has not disappeared, but the misalignment in construction pace will create a real supply chain pressure window in 2028.

This "pit" was not dug by technological failures, but created by state-level elections, a new batch of licensing requirements, and the backlog of 474GW of ERCOT grid interconnection applications.

Electricity — not chips — is becoming the new bottleneck in this AI race.

The report ultimately advises investors to focus on three types of opportunities: large cloud vendors, selected data center supply chain targets, software enablers and AI application adopters.

We believe the market has overreacted, and the partial shift of projects from 2028 to 2029 will not undermine the overall logic of capital expenditure.

This article does not constitute personal investment advice, does not represent the position of the platform. The market carries risks, investment requires caution, please make independent judgments and decisions.

This article is from the WeChat Official Account Wall Street CN, author: Long Yue, published by 36Kr with authorization.