Power & Sites · United States

Ensuring uptime, insuring downtime

Credited to Data Center Dynamics · datacenterdynamics.com · 14h ago

Photograph · Data Center Dynamics

Imagine there’s a real estate sector growing globally at a CAGR of 14 percent, with the promise of near-guaranteed returns, stable rent-based cash flows, and enough demand that supply can barely keep up. Investing in such a fertile field would be a no-brainer. In reality, you don’t need to imagine. That fertile land and its promise exists in today’s data center market.

While, in some cases, data centers are already delivering on this dream, no investment is without risk. And, from the pressure of high uptime standards, to uncertainty around AI returns, or even the threat of destruction from enemies and saboteurs, data centers are facing increasing and varied risks across the globe.

The obvious and traditional choice to mitigate that risk is insurance. But with data center hype reaching a fever pitch, and alchemical growth following, the slow-moving insurance markets are struggling to keep up.

In their reticence to miss out on a sure thing, those involved in data center construction and operation are leaving themselves somewhat exposed to new avenues of loss and risk, as an insurance protection gap emerges in the industry shield wall, leaving investors exposed to a hail of jeopardy.

As above, so below

The huge investments in AI infrastructure over the past few years have given all sectors serving the data center industry a significant boost. Insurance is no slouch here, seeing significant growth in global re/insurance for data centers.

Demand for this infrastructure is nowhere near slowing yet, and there are still further opportunities for insurers to carve an even bigger parcel of land in the data center market.

The cell tower climbers risking it all to help us stay connected

Charles-Marie Delpuech, insurance analyst at S&P Global Ratings, estimates a total insurable asset base of over $2 trillion in the data center sector.

“Based on our projection and average annual investments of $300 billion, this asset base could double in six to seven years,” he adds.

In a recent report on insurable risks for data centers, S&P Global Ratings says that demand for data center insurance coverage could generate $10bn in new premiums in 2026. The research firm contextualizes this against another large insurance market, global aviation, which has annual premiums of around $5 billion.

Hyperscale data centers today are significantly larger and more complex than the more traditional deployments that pre-dated the emergence of generative AI and machine learning. This presents a wealth of opportunities for insurers, who will be able to offer services to the entire ecosystem that goes into the construction of such a data center. Developers, builders, utilities providers, investors, and lenders will all have insurance needs going into a project.

As hyperscalers continue to increase their capex, which is forecast to reach $700bn in 2026, they are looking to external funding sources. Delpuech says these investors may require high insured limits, and will likely continue to drive demand for insurance, both to protect assets or cash flows, such as rental income.

However, with such a high level of demand, some insurers are facing more supply than they can handle. One example of this is insurance firm Parametrix, which provides a range of insurance products, including service-level agreement insurance for data centers. Jonathan Hatzor, co-founder and CEO of Parametrix, argues that the insurance market has been slow to adapt.

“You have a risk that is growing, and usually when you have that, the insurance market is there to try to minimize this risk,” he explains. “But in data centers, we still see a lot of demand but not enough supply of insurance, which is also increasing the entire risk factor.”

Hatzor uses the example of investors offering a loan to purchase GPUs for a data center project: “Think about a debt provider or investors that provide a $10bn loan or equity to buy GPUs. The insurance market can not provide a $10bn property coverage for GPUs.

“There is no historical example of insuring properties with equipment worth $10bn inside of it,” he says. “It’s not just the risk that investors are concerned about; it’s also the fact that the insurance market is slow to adapt.”

He jokes: “It’s not just because we’re dinosaurs or because insurance companies are very old, there’s just no historical records of how resilient those GPUs are.”

Hatzor adds that the insurance market needs to catch up and offer a proper insurance ecosystem in order to really “unleash the potential” of the data center sector.

Despite its slowness, the insurance market is rising to meet growing demand on multiple fronts. Delpuech says insurers are increasing available capacity or adapting their insurance products to match.

“Nonetheless, as we may expect from the insurance industry, before taking additional risks we expect underwriting standards and discipline to remain the focus,” he explains.

Many data center companies today are covered by existing general business insurance, but data center-specific exposures are necessitating more defined coverage. When it comes to these specialized coverages, insurers will remain cautious about their offerings, Delpuech says, and in the meantime, hyperscalers may need to look to self-insure uncovered risks.

One way of doing this, for example, is with captive insurance – a form of self-insurance whereby a company creates its own licensed insurance subsidiary, allowing them to better manage hard-to-place risks and their associated costs.

Corroborating Hatzor’s earlier point, Delpuech states that data center projects now have coverage limits that go “beyond historical larger infrastructure projects like bridges and tunnels, which could have required a maximum $5-$10bn in coverage.”

“Some commercial insurers are now offering per-risk limits in the low single-digit billions for data center exposures, still short of the exposure that is coming to the market.”

But he adds: “Collaboration will be key. Given the size and complexity of these projects, insurers are increasingly working together to share risks, and we have seen many broker-led facilities increasing their aggregate capacity in recent months.”

Always on, sometimes covered

Another growing avenue for data center insurance is SLA insurance. Downtime risk is a direct threat to operational incomes and returns because of the various financial penalties that are built into tenancy agreements.

SLA insurance is a solution to this, turning the volatility of an SLA into an insurable risk. Parametrix says solutions such as this are providing the predictability that will attract the capital required to build out the data center sector.

Hatzor says that, in the past, operators were simply trying to convince investors that downtime is extremely rare, especially in a Tier III or Tier IV data center. But with growing interest in the sector from investors, those coming from more predictable asset classes, wanted something more stable.

Institutional buyers are therefore increasingly requiring SLA insurance as part of M&A and refinancing deals.

These sorts of risks and exposures – downtime, power dependency, or operational disruption – are increasingly becoming a focus for insurers as opportunities to provide more coverage. These risks are becoming as challenging, or in some cases more challenging, than risks associated with physical assets.

However, when it comes to AI infrastructure, SLA insurance is still an emerging service. Delpuech predicts that this type of insurance will become an even bigger topic in the coming years, as the facilities being built today start operation.

“Insurers already have a strong presence in the typical data centers property insurance space,” he explains, “what we are seeing is the new complexity and scale of the hyperscale data centers where we expect insurers will expand for risk and limits they are comfortable with.”

“The reality is that the demand for insurance that has not been historically provided is only starting to come.”

Downtime is tricky to insure, especially when you consider interconnectedness with other data centers. Some specialized insurers will likely look to increase their capacity to meet these insurance demands. But, in some cases, it is likely that larger data center operators will look to cover these exposures with captive insurance systems.

In any case, no single insurer will be able to absorb all of the risks associated with data centers, which are in and of themselves collaborative structures. It is likely that insurers will look to partner to share the risk associated with data centers. S&P Global Ratings says it expects collaborative structures to “play an increasingly central role in scaling up the insurance market’s response,” and that alternative capital will start providing capacity as the market develops.”

Exposures and risks

Common concentration risks can often simply come from the geographical footprint of the data center, which can often mean exposure to natural disasters, especially if there are many high-value assets concentrated in a single location.

For example, a recent report from MS Amlin found that half of planned US data centers, which are now increasingly being built in the south of the country, are at risk from severe convective storms, hurricanes, winterstorms, and earthquakes.

Delpuech says it is “pretty standard” for the industry to be able to locate and accumulate property exposure as they would do across all of their insured properties, not just data centers.

However, building a data center requires integration of multiple stakeholders: the end-user, developers, builders, utilities providers, and investors. With so many affected parties in the same space, and as this complexity increases with more complex data centers, providing multiple insurance products to participants in the same data center can cause concentration. Insurers will need to be careful to avoid concentration and aggregation risk.

Delpuech says data quality and tracking will be important in avoiding and dealing with these exposures.

In order to deal with more specialized risks, specialized commercial insurers generally take the lead on data center projects, Delpuech says, with “great expertise” in engineering and construction risk.

“The traditional property risk where most capacity is deployed is well understood by the industry,” Delpuech notes, “where it becomes more complex is everything related to business interruption or cyber. This is where we could expect growing demand over the coming years.”

“These are not new risks for the industry, but the interconnectedness of data centers across the economy means that the financial exposure can be significant. We understand that coverage is provided, but we would expect limits to be relatively modest, with exclusion regarding non-physical damage like power outage being standard. Also, the waiting period for business interruptions means that hyperscalers should be ready to cover the losses in the first hours or days.”

Operators will also need to consider less forgiving exposures when it comes to data centers. For example, as data centers increasingly become military targets as war infrastructure, as seen in the recent US-Iran war, operators may find themselves adrift with no coverage. For example, some reporting has suggested that Amazon Web Services data centers in Bahrain and the UAE that have been hit by drone strikes linked to Iran will not be able to claim insurance on their losses, as existing policies typically don’t cover war.

Standard property and cyber insurance policies generally contain war exclusions, with the insurer note required to pay out. This is obviously more of a problem in high-exposure zones, such as Ukraine and Eastern Europe, or Israel and the broader Middle East. And as these conflicts have become more of a problem for data centers, re/insurers have generally tightened war exclusion language and required explicit carve-outs.

Cautious expansion

While the data center market continues to rapidly grow, insurers are cautiously expanding to match demand, creating a gap in insurance protection. Rapid construction timelines and large asset values will continue to be a mismatch with how the insurance sector operates, and S&P Global Ratings says insurers will maintain “strong underwriting discipline as they assess increasingly large and complex risks,” in the face of the pace of the industry.

Delpuech explains: “Because of this shortage of capacity, third-party capital outside of the insurance industry may be interested in covering the gap but this will require new innovative structures that make economic sense for all parties.”

The data center market is offering an immense opportunity for both investors and insurers, but where insurers are lagging, investors may follow suit. There already exist fears of an AI bubble. If it were to burst, it has the potential to be catastrophic to global markets, bigger than even the dot com bubble. The data center insurance industry is yet another example of an adjacent sector struggling under the pressure of the industry’s growth, as it builds the superstructure before the foundations are set.

If this insurance protection gap persists, investors in the industry might be less likely to throw all their eggs in one AI-shaped basket. And, given the magnitude of investment required to fund this industry, with some predicting around $3trn of required investment over the next five years, wavering now could spell disaster. Equally, as Delpuech argues, third-party capital may look to cover this gap and help to solve this issue, but as of yet those new and innovative structures are yet to be formed, and data center construction shows no signs of slowing.

Power & SitesUnited States

DCD Studio: A year of data center development with Matthew Baynes, Schneider Electric

Credited to Data Center Dynamics · datacenterdynamics.com14h ago

Useful

Seb Moss speaks with Matthew Baynes of Schneider Electric about adapting to AI-driven GPU growth through modular construction, liquid cooling, future-proof designs, and sustainable grid innovations.

View the articleData Center Dynamics

Front Method

Power & Sites · United States

Kioxia Adds Direct Liquid Cooling to PCIe 5.0 Enterprise SSDs

Credited to Converge Digest · convergedigest.com · 14h ago

Photograph · Converge Digest

Kioxia has added direct liquid cooling (DLC) options to its CM9 Series enterprise SSDs, enabling cold-plate cooling in E3.S server designs. The drives also support conventional air cooling. Kioxia is sampling the CM9 Series to selected customers and showcased the new models at OCP Global Summit 2026 in San Jose.

The PCIe 5.0 drives support NVMe 2.0 and dual-port E3.S operation, with sequential throughput rated up to 14.8 GB/s for reads and 11 GB/s for writes. Read-intensive configurations offer capacities up to 30.72 TB with 1 drive write per day (DWPD), while mixed-use configurations offer up to 25.6 TB with 3 DWPD. Kioxia identifies SSD context caching as a target application.

CM9-DLC Specifications

KIOXIA CM9-DLC | PCIe 5.0 Enterprise SSD

Cooling — Cold-plate direct liquid cooling or air cooling

Capacity — Up to 30.72 TB (1 DWPD); 25.6 TB (3 DWPD)

Random I/O — Up to 3.4M read IOPS; 800K write IOPS

Standards — NVMe 2.0, NVMe-MI 1.2c, OCP Datacenter NVMe SSD v2.5

The product supports NVMe-MI 1.2c and the OCP Datacenter NVMe SSD v2.5 specification. Kioxia has not announced general availability or pricing. Sample specifications may differ from mass-production drives.

Direct liquid cooling is expanding from processors and accelerators toward other server components as AI racks become more densely populated. SSD cold-plate support provides system designers with an additional option for managing storage thermals without relying exclusively on airflow. The announcement does not quantify cooling-related energy savings, which depend on overall server design and operating conditions.

AI inference also creates new opportunities for SSD context caching. Large key-value caches can exceed available GPU memory, encouraging tiered memory architectures that place less latency-sensitive data on lower-cost storage. SSD latency remains significantly higher than GPU high-bandwidth memory, so the performance benefits depend on cache placement and software orchestration.

Kioxia’s CM9-DLC introduction connects these storage and thermal requirements. Production qualification, availability and system-level performance data will help determine deployment economics.

Company Profile: Kioxia

Origins — Toshiba Memory separated in 2017; renamed Kioxia in 2019

Technology — BiCS FLASH 3D NAND; PCIe/NVMe enterprise SSDs

Markets — Cloud, AI infrastructure, enterprise storage, mobile and automotive

🌐 We’re tracking AI infrastructure developments. Follow our ongoing coverage at Converge Digest AI Infrastructure.