ProShares files for an AI hyperscaler bond ETF

The largest hyperscalers averaged roughly $35 billion of annual bond issuance between 2020 and 2024, a borrowing pattern for businesses that had spent years avoiding the bond market. By 2025 the figure reached $93 billion, and ETF Trends expects AI-adjacent issuance to top $570 billion in 2026, crediting its colleague Kirsten Chang with tracking the wave. The same reporting puts technology debt above 13% of broad investment-grade bond benchmarks, up from 9% in 2024, though the two figures describe different populations: $93 billion for the largest hyperscalers, $570 billion for a wider AI-adjacent category the account does not define issuer by issuer.

The spending behind those numbers is specific—data centers, custom silicon, and the power infrastructure that feeds both. Microsoft, Alphabet, Amazon and Meta have moved away from the asset-light model that defined them, and the paper they are issuing to finance the shift has grown large enough inside the corporate market to be carved out of it. ProShares has submitted an SEC filing for the ProShares AI Hyperscaler Bond ETF, which would track a VettaFi index designed to isolate debt issued by the companies financing global AI infrastructure.

What the account does not supply is most of what a buyer wants from a launch story: no filing date, no ticker, no expense ratio, no listing venue. A registration statement describes a fund that might exist, which is a different thing from a fund with a shelf date. Two construction figures do the describing instead: nearly 75% of the index portfolio carries an AA rating or higher, and nearly half of it matures in 10 years or more. The rating profile is the easier half to understand, since the hyperscalers rank among the highest-rated corporate borrowers anywhere and ETF Trends frames the AA weighting as a degree of credit safety that has become hard to find in yield-focused credit strategies.

Duration carries the risk. A corporate bond portfolio with half its maturity beyond a decade behaves, in return terms, more like a rates position than a credit one, and that inference runs against the way the product will likely be described. Spreads on AA obligations are thin by construction, so an AA-heavy fund has to assemble its yield from term rather than from credit selection, and its price will track the Treasury curve well before it tracks any headline about data-center capital spending.

The rate backdrop sharpens the point. In early-September coverage of WisdomTree's AGZD, the 10-year Treasury yielded 4.78%, its highest since January 2025, with hike odds above two-thirds. A fund filed in October with nearly half its holdings maturing in a decade or more is being brought into a market where the curve does most of the pricing, and ETF Trends makes the advisory case directly, describing the term structure as a long-duration sleeve for advisors looking to lock in elevated long-term yields.

AI exposure, meanwhile, has mostly been sold as equity. The ETF Trends column notes that gaining exposure to the artificial intelligence theme has historically meant accepting equity market volatility, and pitches the bond fund as a route to yield from the same balance sheets that are funding the buildout. What that framing leaves out is the asymmetry: bondholders take a coupon and give up the upside equity investors are bidding for.

Investors in mainstream credit funds already hold a version of this exposure. Our September reporting, citing ETF Trends, found that Alphabet, Amazon, Meta, Microsoft and Oracle issued more than $132 billion of bonds in the first half of 2026, lifting tech's weight in pure corporate bond ETFs by about 300 basis points. Hyperscaler paper has been absorbed into aggregate and investment-grade portfolios for years, so the thematic wrapper has to answer what isolation adds; on the disclosed profile, the answer is quality and duration—the same issuers, weighted and dated differently.

How the index handles issuer concentration is the open question. A rulebook built to isolate debt from the handful of companies financing AI infrastructure can only hold a short list of names, and the coverage does not say how many, whether weights are capped, or whether issuers beyond the four largest hyperscalers qualify; Oracle appears in the issuance tallies cited above. Nor does it say how the index treats new paper, which matters when the tracked borrowers add debt by the quarter.

This publication has argued that the launch machine now outruns the shelf, and a single-theme bond fund is a narrower product than the broad credit funds it would sit beside in model portfolios, one whose uptake likely runs through platform due diligence rather than through a headline. What settles it sits in the paperwork still to come: how tightly the index caps a portfolio that can only be a few names, and what ProShares charges for bonds whose credit quality is the least of their risks.