Chipmakers now underwrite their customers, from Meta's equity to Anthropic's chip debt

The two ideas are at different stages. The GPU collateral servicer is getting crowded at the data layer. Silicon Data, Compute Exchange and Ornn supply indices and residual curves, and Ornn also sells residual-value protection. Aravolta and Silicon Lien already pitch monitoring and verification to lenders. Banks fall back on traditional appraisers and the assumptions set by rating agencies like KBRA. Nobody I found yet works as a neutral, independent servicer in the way aircraft and railcar lending have them: someone who combines hardware verification, mark-to-market residuals, contract and offtaker credit monitoring, and covenant reporting, and whose marks rating agencies and private-credit valuation committees will accept. A new entrant would win by being independent of the index and exchange businesses, since the Silicon Data and Compute Exchange pair, and Ornn, both have a trading interest in the prices. It would also need to underwrite the offtaker contracts as well as the chips. That matters because commentators argue these deals are really credit substitution against hyperscaler or lab balance sheets, and because facilities run about five years against customer contracts of about three. The risks are that Silicon Data or Aravolta move up into servicing, and that big incumbent appraisers simply add a GPU desk. The circular-financing ledger looks nearly empty as a product. What exists is newsletters, one-off visualizations, sell-side diagrams and blog posts on filings. I found no maintained, structured dataset that links each vendor equity stake, warrant, guarantee, SPV anchor investment or loan (Nvidia's portfolio of roughly $99B, AMD's warrants to OpenAI and Meta, Broadcom's lending to Anthropic) to the revenue and customer concentration that vendor reports. Credit-rating teams and buy-side analysts are the better buyers. Sell-side banks build this internally, and Bloomberg or S&P could add it as a field. That makes the ledger a defensible niche, not a large standalone market, so it probably works best as an entry wedge or companion dataset for the collateral servicer.

• Silicon Data

  Runs daily GPU rental indices, including the H100 index on Bloomberg as SDH100RT, plus B200 and hyperscaler benchmarks. It also publishes residual-value curves built from forward rental curves and is partnering with CME on compute futures. It raised a $30.5M Series A. This is the most likely pricing input for the collateral servicer, and Silicon Data could move into it.

• Compute Exchange

  Sister company to Silicon Data. It has a GPU pricing calculator with forecasts and, since July 2026, a marketplace for used and refurbished GPUs. That gives it a liquidation and resale channel, which a residual-value marking business would want.

• Ornn

  Runs a compute price index meant to be used for settling contracts (OCPI), with contracts reported to clear through ICE. It sells structured residual-value protection and pitches market-based estimates of future GPU values in place of straight-line depreciation. It raised a $33M seed round from backers including Galaxy and a16z crypto.

• Aravolta

  Sells telemetry-based underwriting and continuous monitoring of GPU-backed credit to lenders. USD.AI is a named client. It overlaps directly with the monitoring part of the collateral servicer.

• Silicon Lien

  Building an independent verification layer for compute credit: a baseline at funding, checks during the life of the loan, and exception reports. Its research says current deals rely on desktop appraisals and AUP reports. It is close to the servicer idea, though it appears to be early stage.

• Gordon Brothers (and traditional asset appraisers)

  A secondary source names independent appraisal firms like Gordon Brothers as doing quarterly fair-market-value certifications for GPU lenders. I did not confirm that they do GPU-specific work. Even so, incumbent equipment appraisers are the natural default for banks.

• KBRA and other rating agencies

  KBRA has published on how GPU residual-value erosion weakens collateral coverage. Rating agencies set the residual-value assumptions in rated GPU ABS and private deals. A third-party servicer would have to satisfy them, and they could also compete with it.

• levelbox.ai

  Publishes filing-based analysis of Nvidia's and Broadcom's circular financing exposure. It is the nearest thing I found to a filings-linked circular-financing product, though it appears to be content rather than a structured ledger.

• PlotSet AI money-flow visualization

  A D3 map of money flows between AI companies that labels each flow as verified, estimated or projected. It is a one-off visualization, not a maintained dataset tied to vendor revenue.

• LongYield 'The Great AI Money Loop'

  A paid newsletter that maps the circular deals (Nvidia and OpenAI, AMD's warrants, Oracle). It is commentary, not a structured ledger.

• Morgan Stanley and sell-side research diagrams

  Sell-side banks have published capital-flow diagrams linking OpenAI, Nvidia, Microsoft, Oracle, AMD and CoreWeave. Big banks already do this ad hoc for their own clients, which limits demand from the sell side.

Checked 2026-10-27