Barclays AI Data Center Financing: Insurers Take On the Risk, and Operating Data Comes Into View

What Barclays, HSBC and Deutsche Bank are doing

According to the Bloomberg report by Esteban Duarte and William Shaw, Barclays has spent recent weeks testing whether insurers will provide credit risk cover on more than $10 billion of debt that finances AI infrastructure. The goal is to free capacity for more lending as hyperscaler funding needs grow. One person said Barclays wants insurers to cover more than 50% of the credit risk on some AI-linked facilities. Size and timing are undecided, and Barclays declined to comment.

Other banks are doing the same. A second Bloomberg report, carried by Insurance Journal on 30 October 2026, says HSBC is in preliminary talks with insurers about a series of project finance deals, some tied to AI infrastructure, and is working on a significant risk transfer tied to more than $1 billion of data center linked loans. HSBC called the plans normal portfolio management and said its data center exposure is small. The same report says Deutsche Bank has been working on a risk transfer tied to about $2 billion of project finance debt that includes data center loans.

How credit risk insurance moves the risk

Banks also use derivatives and outright loan sales, but the HSBC report explains the two tools in these stories. In a significant risk transfer, investors deposit liquid assets upfront to cover losses on a first-loss tranche and receive coupon payments. In credit risk insurance, the lender pays premiums and the insurer covers an agreed percentage of losses on specific loans.

Either way, part of the loss on a data center loan moves to someone outside the bank, while the borrower and the building stay the same. The bank gets regulatory capital back and can lend again.

Why it matters for AI data center operators

Insurers have limited room for more of this risk. The Barclays report cites a Willis Towers Watson survey released in May, which found that credit risk insurers had increased their exposure to data center finance. Several non-trade credit insurers in that survey said they were "approaching sector limits or are already full on data center exposure." The HSBC report adds that yields on AI-linked bonds and loans are rising sharply.

Demand keeps rising: the Barclays report cites a McKinsey projection of $6.7 trillion in global data center spending by 2030, and Citadel co-chief investment officer Pablo Salame saying the credit market is "getting jammed" with these risks.

With tight capacity, insurers can choose which loans to cover and at what premium. In our reading, a facility that can show how it runs is easier to discuss than a forecast; the reports do not say how insurers will choose.

Operating evidence worth having ready

None of the reports lists what lenders or insurers will request from borrowers. Some operating records, though, answer the first questions anyone sharing the risk would have: does the AI data center keep running, and does it keep earning?

Availability comes first: outages, their duration and their cause, with a history long enough to show a pattern. Power is next: rack power against the rated limit, which racks are constrained and how the site's PUE has moved over time. Liquid-cooled halls add cooling loop history, meaning supply and return temperatures, flow, pressure and any leak events.

Idle GPUs earn nothing, so allocated versus used card hours and idle capacity by tenant show whether contracted capacity is consumed. Change and incident records, with the tickets, complete the picture.

These records help most when they already exist. Rebuilding a year of history from scattered tools during a refinancing is slow.

What the operator sees in Sensaka

Sensaka for AI Data Centers monitors servers, accelerator cards, networks, storage, racks, power systems, liquid cooling equipment and the environment in one platform. Hardware is monitored agentlessly through the BMC. It shows a rack power density heat map with limit alerts, PUE and WUE for the site and per zone as a live value and a trend, and card level utilization, temperature, power and health for NVIDIA, AMD and other accelerators.

For cooling, liquid cooling monitoring tracks CDU status, supply and return temperature, pressure differential and flow indicators, and leak events, and keeps the history to compare recurring conditions. It monitors; any control action stays with the site's own procedures. GPU usage metering records allocated card hours, idle capacity by project and tenant, and daily or monthly usage summaries.

Alerts become tickets synced with ServiceNow or Jira, and the CMDB keeps change tracking and historical snapshots. Asked "Which racks or zones are near their rated power?", the AI Cockpit builds a board and can render it as a 30-second data video.

Takeaway

Barclays, HSBC and Deutsche Bank are looking for others to share the risk on data center loans, and some insurers say they are near their limits. Nobody has said what operating proof this will require. Operators who keep availability, power, cooling and GPU utilization history in one place will not have to build it under deadline.

Frequently asked questions