Marble Raises $11M Seed to Finance AI Compute Operators

Polychain Capital led an $11 million Series Seed round into Marble, a capital markets startup that launched on November 10, 2026, to build private-credit infrastructure for AI compute operators. The company will also run an institutional trading platform, Marble Markets, spanning digital assets, prediction markets and compute.

Maven 11, Ripple, Susquehanna Crypto, Lattice and Blockwall Ventures joined the round, alongside other venture and strategic investors and a group of AI-focused angels. Marble plans to spend the proceeds on hiring across compute financing, trading, risk, legal and operations, and on building the trading, custody and banking connectivity behind Marble Markets.

The founders are Matt Ballensweig and Leon Marshall, who serve as co-CEOs. Ballensweig was most recently BitGo's global head of trading and previously co-headed trading and lending at Genesis. Marshall was CEO of Galaxy Digital's European business and worked earlier at Genesis, Apollo and Trafigura.

Why target mid-market GPU operators?

Marble Compute, the asset-management arm, connects institutional investors with independent AI cloud providers — the market calls them neoclouds — and GPU operators. The founders lay out a gap: the largest neoclouds and hyperscalers can finance growth through public markets and major banks, while smaller operators have fewer and less flexible options even when customers have already committed to purchasing their capacity.

Marble Compute will not lend from its own balance sheet. Its role is to source opportunities, run diligence and underwriting, and structure each financing for investors. The underwriting process examines three things: the hardware itself, the creditworthiness of contracted customers, and the operator's ability to service debt. Marble will also seek diversified customer commitments, so that each financing rests on contracted cash flows rather than a single counterparty.

The rollout sequence, subject to regulatory approvals, comes in stages:

• Loans syndicated directly between operators and institutional lenders come first.
• Single-opportunity off-chain vehicles and on-chain vaults for individual GPU operators follow.
• Diversified multi-opportunity funds arrive as deal flow and a track record develop.

What does the trading platform do?

Marble Markets is planned around principal trading. Clients would face Marble directly, with Marble trading from its own capital. Client assets stay where they are — with existing custodians, brokerages and banks. A high-touch desk would handle execution, supported by a client platform that shows positions, collateral and exposure in one place.

The planned product set is broad:

• Digital-asset spot trading, borrowing, lending and derivatives
• Large-block bilateral prediction-market trading
• Over-the-counter compute forwards
• Tokenized equities and equity swaps over the longer term

The design's distinctive feature is shared collateral. Instead of requiring clients to prefund each counterparty separately, Marble would let them post one pool of eligible digital and traditional collateral across markets. Marble says its Markets lending would be overcollateralized, meaning backed by collateral exceeding the loan's value.

How do the two businesses connect?

Compute forwards are the hinge between the credit and trading arms. An operator could use a forward to fix a rental price for capacity delivered in the future. Marble argues this helps both sides of the market: lenders get clearer cash flows to underwrite, and operators get a better basis for determining borrowing capacity and equity requirements.

Marshall explained the underwriting logic in the launch release: "Two GPU operators can buy identical hardware and raise very different amounts of debt against it. A lender wants to know who's buying the compute, at what price and for how long."

The division of labor is explicit. Marble Compute originates the operator relationships; Marble Markets deepens them through hedging. The founders also frame Compute as a structured route for digital-asset capital to finance AI infrastructure and earn returns from it.

What happens next?

The products remain development plans rather than live services, and every step depends on regulatory approvals. The company has set two near-term targets: Marble Compute expects to close its first loan syndications for mid-market GPU operators in early Q1, and Marble Markets aims to begin digital-asset trading in January.

The launch ties together two requirements for expanding independent AI compute — financing for the hardware and a mechanism for hedging the revenue that hardware generates. The first syndications and the January trading launch will test whether Marble can convert that combined design into an operating business, and whether institutional financing can broaden beyond the largest cloud providers.