BlackRock: AI Expansion Will Continue to Drive Up Yields
BlackRock stated that the government's massive borrowing has already intensified capital competition, and the new financing demands from AI infrastructure mean that the upward pressure on yields is not yet over.
The investment research institute estimates that by 2030, the annual financing needs in the U.S. could approach $8 trillion, primarily driven by capital expenditures related to AI expansion. Another estimate puts it at over $7.5 trillion. Bonds related to AI and data centers account for about 14% of this year's U.S. investment-grade issuance, projected to be 5% by 2025, compared to about 1% over the past decade.
Strategy head Vivek Paul noted that even if the market discusses a slowdown in the development of cutting-edge models, it does not mean that physical expansion will decelerate, as most computing power is used for inference rather than training. Institutions maintain a risk appetite, avoiding the weakest borrowers and large-scale issuers in credit, favoring intermediate coupons. Concurrently, the situation in the Middle East and Brent crude oil returning above $100 per barrel are considered additional upward factors, with persistent inflation also factored in.
Major cloud providers issued about $200 billion in investment-grade bonds in the first half of the year, nearly double the total for 2025. The market previously attributed the rise in 30-year U.S. Treasury yields to the supply of tech bonds and growth expectations. Other analyses emphasize that deficits are the main cause, with the corporate sector still being net savers.
BlackRock's power research indicates that to meet data center demands, the U.S. may need to add about 148 gigawatts of power by 2030, compared to about 42 gigawatts of data center electricity use in 2025. Supply is manageable from 2026 to 2027, but tighter after 2027.
This is a game of existing stock. Buyers are issuers that must roll over government bonds and finance data centers, while sellers are the same long-duration funds. Funds are being squeezed between government bonds and investment-grade tech debt; the beneficiaries are credits that can provide intermediate coupons and are not super large issuers, while those under pressure are long-end Treasury holders and large-scale cloud providers expanding their balance sheets. Whether yields continue to rise depends on whether expansion truly halts with the slowdown in model development.
Source: Public Information
ABAB AI Insight
BlackRock titled its report "AI Expansion Meets 5% World," placing the Treasury and major cloud providers in the same funding pool. Over the past decade, tech companies have financed data centers with operating cash flow, and in the first half of 2026, five issuers raised about $200 billion, with AI-related bonds jumping from 5% to 14% in investment-grade over the year. The research institute deliberately separates "slowing model iterations" from "ongoing inference plant operations," signaling to the bond market: the safety debate is moving to the Security Council, while cement and transformers are still being ordered.
The capital path is the simultaneous expansion of national deficits and corporate bonds. Treasury bonds absorb the tax revenue gap, while tech bonds absorb power, chips, and cabinets. BlackRock itself holds stakes in some data center projects, warning of congestion while acting as a landlord. The allocation strategy suggests avoiding both ends: junk credit and giant issuers, aiming to collect coupons in the middle tier, which is about layering in a crowded issuance rather than exiting the theme.
This mirrors the fiscal boosts of 2021-2022 that raised real interest rates and the historical capacity competition in railways and power grids that crowded out Treasuries. The counter-narrative is that deficits are crowding out, while corporations remain net savers. The industry is in a period of expansion where AI transitions from an equity story to a core fixed-income holding: the curve steepens, investment-grade and tech correlations rise, and Treasury hedges become ineffective.
The structural change is a shift in pricing power. The risk-free rate is no longer determined solely by central banks and deficits but must also accommodate issuance windows for data centers. The mechanism is that the same insurance and pension accounts must digest both Treasury supply and hyperscaler supply, raising the clearing rate from both ends. Whoever locks in inference capacity with long-term leases first will pass the rising rates onto late-issuing governments and peers.
ABAB News · Cognitive Law
Models can slow down, but data center financing will not stop together
Treasuries and tech bonds are competing for the same long-duration money
In a 5% world, coupons are priced before narratives