Alphabet issued a hundred-year bond to finance AI data centers. The century maturity reveals what the permanent asset actually is — and it is not what the AI narrative assumes. The infrastructure will be used. The question is whether the investors who financed it will be around to collect.
In February, Alphabet sold a hundred-year sterling bond — one billion pounds, ten times oversubscribed, the first century debt from a technology company since Motorola in 1997. The buyers were pension funds matching long-term liabilities. The seller was a company spending $185 billion this year on data centers. Neither party expects to be around when the bond matures in 2126.
The century maturity is not a financial novelty. It is a confession. The asset being financed will outlive the technology it was built to house.
The Permanent Asset
A data center is three things layered on top of each other. The first layer is the real estate: the land, the building, the grid connection, the water rights, the local permits that took years to secure. The second layer is the power infrastructure: the substations, the transformers, the on-site generation, the contractual right to draw a gigawatt from a grid that cannot build new capacity fast enough. The third layer is the compute: the GPUs, the networking, the cooling systems that will be obsolete within five years.
The hundred-year bond prices the first two layers. Nobody lends for a century against hardware that depreciates in three. Pension funds buying this debt are making the same bet they make on office towers and toll roads — that location and infrastructure access are permanent assets in a way that the equipment inside them is not.
They are right. And the distinction matters more than anything else in the six-hundred-billion-dollar AI infrastructure buildout.
The Financing Loop
The Dallas Federal Reserve published a paper in February titled "How AI debt financing impacts duration supply and interest rates." The finding: AI-related investment-grade issuance could supply $360 billion in ten-year equivalent duration over the course of 2026. Morgan Stanley projects hyperscaler bond issuance will top $400 billion this year, more than double the $165 billion issued in 2025. AI-related issuance now represents roughly thirty percent of net investment-grade corporate bond supply.
This is not a detail. It is a reflexive loop. The companies building AI infrastructure are borrowing at such scale that their debt issuance is changing the rate environment in which they borrow. More supply of long-duration bonds means higher yields on long-duration bonds. Higher yields mean higher financing costs for the next data center. The buildout is making itself more expensive through the act of financing itself.
The century bond is the purest expression of this dynamic. Alphabet borrows for a hundred years. The bond adds duration supply to a market already absorbing record issuance. Yields tick up. The next hyperscaler pays a higher coupon. The infrastructure gets built anyway because the demand for AI compute is real — but the cost of building it rises with every tranche sold.
The SPV
On April 7, Bloomberg reported that PIMCO and Bank of America are arranging roughly fourteen billion dollars in debt financing for an Oracle data center in Saline Township, Michigan. The facility is part of the Stargate project — Oracle's partnership with OpenAI. Blackstone is providing two billion in equity. The total capital package: $16.3 billion for a single campus.
The structure is project finance. The debt sits in a special purpose vehicle, off Oracle's balance sheet. Oracle leases the facility back. The bonds are 144A — private placements sold to qualified institutional buyers. Pension funds. Insurance companies. Endowments. The same investors who buy toll road bonds and airport revenue notes.
This is the tell. When the financing structure for an AI data center is indistinguishable from the financing structure for a highway or a power plant, the market has already decided what the asset is. It is not a technology investment. It is infrastructure. The GPU inside is the tenant. The building and the grid connection are the landlord.
Oracle's total debt is projected to reach $290 billion by 2028. The PIMCO deal is the first pure project-finance SPV for AI infrastructure at this scale — debt structured against the asset's cash flows rather than the sponsor's balance sheet. If it succeeds, it becomes the template. Every hyperscaler with a hundred-billion-dollar buildout and a CFO who wants to keep debt off the books will follow.
The Parallel
In September 1873, Jay Cooke & Company — the most powerful bank in the United States, the firm that had financed the Union's Civil War debt — collapsed under the weight of its investment in the Northern Pacific Railroad. Within two years, eighty-nine railroads had declared bankruptcy. Half of all railroad bonds defaulted between 1873 and 1879. The Long Depression lasted six years.
The track stayed in the ground.
Between 1868 and 1873, the United States had laid thirty-three thousand miles of new railroad. The financing was reckless — leveraged, speculative, built on the assumption that traffic would materialize faster than it did. When traffic came in below projections, the capital structure collapsed. But the physical infrastructure — the right-of-way, the graded roadbed, the bridges, the stations — did not disappear when the bonds defaulted. It could not. The permanent asset was in the ground.
Investors who bought bankrupt railroads at fire-sale prices in the late 1870s built some of the greatest fortunes of the Gilded Age. The network that existed in 1873 — seventy thousand miles — grew to two hundred and forty thousand miles by 1900. The assets that bankrupted their original investors operated at capacity for four decades. The railroad that ruined Jay Cooke was completed in 1883 and ran profitably until World War One.
The lesson was not that the railroads were a bad investment. The infrastructure was transformative — it unified the continental economy, enabled industrialization, and generated enormous productive value. The lesson was that the investors who financed the construction were not the investors who captured the returns. The permanent asset — the right-of-way — outlasted the capital structure that built it.
The Question
The AI infrastructure buildout is the largest capital expenditure cycle in technology history. The five biggest spenders — Microsoft, Amazon, Alphabet, Meta, and Oracle — are collectively committing between $660 and $690 billion in 2026 alone. The demand for AI compute is real. The data centers will be used. The power connections will appreciate. The locations will matter for decades.
None of that answers the financing question.
A hundred-year bond at 120 basis points over gilts prices a century of cash flows. The pension funds buying it are betting that the location and the grid connection will generate rent for a hundred years regardless of what hardware sits inside. That bet is probably correct. But the equity investors funding the buildout are making a different bet — that the returns from AI services will exceed the cost of the infrastructure before the debt comes due. That bet is the one the 1873 investors made about railroads.
The PIMCO deal reveals the fracture line. When debt is structured against the facility's cash flows rather than the operator's balance sheet, the lenders are pricing the asset as real estate. When the equity is provided by Blackstone — a firm built on real assets — the capital markets have spoken. The permanent value is in the dirt and the wire, not in the silicon.
Morgan Stanley projects $400 billion in hyperscaler issuance this year. The Dallas Fed warns the duration supply is structurally rate-positive — the borrowing itself raises rates. The loop tightens with each tranche. And somewhere inside every one of these deals, the same separation exists: the infrastructure that will last, and the technology stack that will be replaced every three to five years.
The railroads that survived the Panic of 1873 operated profitably for decades. The investors who built them went broke. The investors who bought the wreckage got rich. The right-of-way outlasted everything except the ground it was laid on.
The century bond says the same thing is happening now. The question is not whether the AI infrastructure will be used. It is who will own it when the current financing cycle ends.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)