How the midterms could fuel the K-shaped AI trade

A “K-shaped AI trade” describes a widening split between companies benefiting from AI investment and companies struggling to finance it. The upward branch represents firms with capacity, cash flow, and diversified businesses. The downward branch represents firms that depend heavily on borrowing to build new infrastructure. This matters because tighter supply can make existing capacity more valuable.

Microsoft and Amazon each rose more than 6% after the midterm results. AI infrastructure provider CoreWeave fell around 10%. CoreWeave, Nebius, and Iren had already dropped at least 30% in October as credit markets tightened. Investors are separating companies that can fund projects internally from those exposed to expensive refinancing.

The divide could widen if permitting delays slow new construction. Hyperscalers may secure cheaper capacity or acquire assets from weaker operators. Neoclouds face greater pressure from borrowing costs, energy prices, and refinancing conditions.

Data centers are buildings filled with servers, storage systems, networking equipment, power supplies, and cooling systems. They process and store the data behind online services. For AI, they provide the specialized computing capacity needed to train models and deliver responses to users. This makes data centers the physical foundation of AI services.

The article focuses on capacity coming online in different years. Microsoft and Amazon benefit from broader cloud businesses and more existing capacity. Companies such as CoreWeave build infrastructure aimed at serving AI demand. Their projects require large amounts of capital, power, and construction, so permitting and financing can determine how quickly services expand.

The article does not quantify how much computing an individual AI service requires. It does show that construction delays matter. Capacity expected in 2027 is mostly already under construction, while 2028 projects face greater exposure to permitting delays and regulatory uncertainty.

Tighter rules could slow the approval and construction of new data centers. That would not affect every year equally. The timing of a project matters because facilities already under construction are harder to stop than projects still waiting for permits. Delays could reduce or postpone the capacity available for AI services.

Much of the capacity expected to come online in 2027 is already under construction. That limits the near-term impact of new restrictions. Projects planned for 2028 have greater exposure because they may still need permits and face changing regulatory requirements. Nomura identified 2028 as the more vulnerable period.

The article says tighter construction rules could become a larger headwind to AI investment from 2028. BMI warned that a sharp slowdown in data center construction and AI investment could cause a greater-than-expected decline in overall investment. Texas’s proposed moratorium illustrates the potential scale of the policy risk.

When new data centers become harder to build, available capacity becomes scarcer. Scarcity can increase the strategic and financial value of facilities that are already operating or under construction. Companies with that capacity can serve customers sooner, avoid some permitting risk, and face less uncertainty about future supply.

The market reaction showed this logic. Microsoft and Amazon each surged more than 6% after the midterm results, while CoreWeave fell around 10%. JPMorgan’s James Sullivan said the performance split was largely driven by the view that existing capacity had become significantly more valuable because of incremental supply constraints.

Stronger companies may also benefit in another way. JPMorgan Asset Management said financially stronger hyperscalers could secure cheaper capacity or acquire assets if weaker operators scale back. This could further concentrate AI infrastructure among companies with cash flow, diversified revenues, and capacity already in hand.

Hyperscalers are large technology companies with broad cloud businesses, diversified customers, and multiple revenue streams. Neoclouds are newer or more specialized cloud providers focused heavily on AI infrastructure. The distinction matters because companies with broader businesses can rely on more sources of cash when one project or market becomes difficult.

Microsoft and Amazon illustrate the hyperscaler side of the divide. Their shares each rose more than 6% after the midterm results. CoreWeave represents the neocloud side and fell around 10%. JPMorgan said hyperscalers benefit from more diversified customers and revenue streams, while neoclouds face greater exposure to financing conditions.

This structure affects resilience. Hyperscalers can use internally generated cash, secure cheaper capacity, or potentially buy assets from weaker operators. Lower-rated neoclouds must manage borrowing costs, energy prices, refinancing conditions, and project delays more directly.

Borrowing costs matter because data center projects require large upfront investments before they generate revenue. If lenders demand higher interest or wider credit spreads, each new facility becomes more expensive to finance. A company that depends heavily on debt has less room to absorb delays, cost increases, or weaker demand.

JPMorgan’s James Sullivan said borrowing spreads had widened significantly for neocloud companies. He said their cost of additional capital was substantially higher than for hyperscalers. CoreWeave’s five-year credit default swap spread rose past 1,100 basis points, showing how sharply investors’ concerns had increased. CoreWeave, Nebius, and Iren also fell at least 30% in October.

Hyperscalers have broader cloud businesses and more internally generated cash flow. That can reduce their need for expensive borrowing. Lower-rated neoclouds remain more exposed to energy prices, refinancing conditions, and permitting delays.

A credit default swap, or CDS, is a financial contract that provides protection if a borrower defaults. The CDS spread is the ongoing price of that protection, usually expressed in basis points. A wider spread means investors demand more compensation to take or insure against the borrower’s credit risk. It is a market signal, not proof that default will occur.

CoreWeave’s five-year CDS spread surged past 1,100 basis points. The article links that rise to increasing costs for investors seeking protection against a potential default. The sharp move appeared as credit markets tightened and investors questioned how AI infrastructure projects would be financed.

The signal matters because CoreWeave is a debt-funded neocloud with major capital needs. Higher perceived risk can make refinancing more expensive and restrict access to capital. That may force weaker operators to slow expansion, while financially stronger hyperscalers secure capacity or acquire assets.