Fund RadarETF Filings

GraniteShares files for eight AI ETFs that may own private companies

Each Nov. 9 draft would put at least 80% of the fund in one public artificial-intelligence business, and could add private stakes that do not count toward that test.

ETF.net Research

Nov 10, 2026, 8:03 AM ET

COOL+3.87%

A close-up view of a microchip and electronic components resting on a dark surface.

Key takeaways

GraniteShares filed for eight AI ETFs that could own private companies outside the public test.
Private stakes would not count toward the 80% set aside for public stocks.
Expected illiquid swaps and vehicle stakes would share a 15% illiquid-asset limit.
At each rebalance, no one company would be more than 20% of the portfolio.
Tickers and management fees are still blank, and the funds have not started.

GraniteShares ETF Trust filed on Nov. 9 for eight exchange-traded funds that would own narrow public slices of the artificial-intelligence buildout. Each draft would also allow a private-company stake, and that stake would not count toward the 80% of net assets set aside for public stocks. GraniteShares Advisors LLC is the adviser.

A fund could take a passive minority stake in a special purpose vehicle, an outside entity that holds the private shares. The drafts say the fund would not create or control that entity. It could also enter a contract the drafts call a Pre-IPO Perpetual Futures Swap: a private agreement meant to pay the economic return of a futures contract that never expires, tied to the implied market value of those shares.

An eligible private company would be one that would pass the fund's public test if its shares traded, or one the adviser believes is a significant customer of those companies or a significant supplier of AI technology to them. The drafts give developers of large-scale AI models as the example.

The drafts say those swaps are expected to be illiquid, and that they would count toward a limit of 15% of net assets in illiquid investments. An illiquid investment, as the drafts define it, cannot be sold in the ordinary course within seven days near the value the fund has put on it. Stakes in the outside vehicles are also expected to be classified as illiquid, so they would fall under the same limit.

The drafts say the fund does not intend to buy a vehicle that charges an ongoing management fee or a share of profits. The size of each swap would not exceed the assets set aside for it, and the swaps would not be used to magnify the position.

The drafts say the fund would generally try to sell the vehicle interest before the private company went public, because that interest is expected to trade at a significant discount while the new shares are locked up.

What the 80% would own

Under normal circumstances, at least 80% of net assets would go into the stocks of companies that pass the fund's test, or into swaps that pay the return of those stocks. At each rebalance, no one company would be more than 20% of the portfolio.

A figure printed inside square brackets is not final. The 20% cap is not in brackets.

Proposed fund	What would count toward the 80%
GraniteShares Orbital AI ETF	At least 50% of revenue or profit from space work, including computing done in orbit, satellites, launch, or space data. A company without material revenue can qualify on its main business. It must also disclose that it does AI work in orbit, uses AI to run its space business, or supplies those firms.
GraniteShares Korea AI Champions ETF	Headquartered, incorporated, or principally traded in South Korea, with at least 50% of revenue or profit from AI work such as chips, robotics, software, or data-center services. A company that offers its own large-scale AI model, or builds one into its main products, can qualify without the 50%.
GraniteShares AI Data Center Cooling ETF	At least 50% of revenue or profit from data-center cooling. Generating, transmitting, or distributing electric power does not count, and neither does equipment that conditions or distributes that power.
GraniteShares AI Grid & Nuclear Build-Out ETF	At least 50% of revenue or profit from nuclear or grid work, and a public disclosure that the company supplies data centers or that data-center demand is a material driver. Minimums of 30% in nuclear companies and 30% in grid companies are in brackets.
GraniteShares China AI Supply Chain ETF	Organized or based in China, including Hong Kong and Macau, or a majority of revenue from goods or services there, with at least 50% from AI supply-chain work. A company that offers its own large-scale AI model, or builds one into its main products, can qualify without the 50%.
GraniteShares Defense AI Software & Autonomy ETF	At least 50% of revenue or profit from AI work for defense and national-security customers of the United States or its allies. A bracketed alternative would let a firm whose main product is AI software qualify at 25%.
GraniteShares Co-Packaged Optics & AI Interconnect ETF	At least 50% of revenue or profit from technology that connects AI computers, including optics built into the chip package. Long-haul telecom equipment and general enterprise networking do not count.
GraniteShares Advanced Packaging & Chip Stacking ETF	At least 50% of revenue or profit from advanced packaging and chip stacking, including stacked chips and the tools used to package them.

The cooling draft draws a line a listed fund does not. VegaShares AI Thermal, Cooling & Power Management ETF, COOL, says it invests in global companies that generate significant revenue from the cooling, power delivery, and thermal management systems that high-density AI data centers need. Power delivery is what the GraniteShares cooling draft leaves out.

The China draft would bar any company on the Treasury Department's list of Chinese military-industrial complex companies. Wording that would also bar the Defense Department's list of Chinese military companies, and the Commerce Department's Entity List, is still in brackets.

The same draft warns that a company qualifying through its own large-scale AI model may get only a small share of its revenue from AI, and that its results may be driven by its other businesses.

The ticker, the management fee, and the minimum market value a listed company would need to be bought are still blank. The distribution fee is filled in at 0.00%.

Each filing asks to become effective 75 days after Nov. 9. The shares cannot be sold until the registration statement is effective, and the drafts say the funds have not started.

Want more of our stories in Google’s Top Stories? Add etf.net as a preferred source on Google