'Absolutely Ridiculous': Data Center Tenants Wasting Power Capacity As Providers Scramble For More

The data center industry is faced with unprecedented power constraints, particularly in major markets like Northern Virginia and Silicon Valley, yet nearly half of the power already allocated to existing data centers is going to waste. 

Tenants in colocation data centers typically use less than 60% of their leased capacity, multiple experts told Bisnow, with gigawatts of energy going unused even as developers and utilities go to greater and greater lengths to bring new capacity to market. 

This so-called stranded capacity — power contractually guaranteed to be available to tenants that is never used — has generally been regarded as just a fact of life in the data center business, even as new power becomes scarce. But a growing chorus within the industry is pushing both data center providers and their tenants to take steps to use a greater share of the capacity that already exists. They say doing so would help meet surging demand, lower costs for both landlords and tenants, create new revenue streams for providers, and help make an industry that is among the world’s largest power consumers significantly more sustainable. 

“We’re out of power, but there's 800 megawatts in Virginia not used ever — that is absolutely ridiculous,” said Dean Nelson, CEO of IT platform Cato Digital and the chairman of data center industry group Infrastructure Masons, who has been one of the leading voices drawing attention to the issue of stranded capacity.

“If we're going to drive sustainability in digital infrastructure, we cannot have a global portfolio where a minimum of 40% of capacity built is never used. That is a waste of resources everywhere.” 

Northern Virginia is the largest data center market in the world and has been the tip of the spear when it comes to the power constraints facing the industry. The scale of the problem became readily apparent last summer, when utility Dominion Energy announced it would have to delay power deliveries to a number of data center projects in the industry hub of Loudoun County for years due to insufficient transmission infrastructure. It was a warning sign for all major data center markets, where power scarcity has become the norm. 

But of the roughly 2,000 megawatts of electricity already allocated to existing Northern Virginia data centers as of last year, only around 1,200 megawatts were ever used, with tenants in colocation data centers typically using just a fraction of leased capacity, according to Cato Digital.

Chris Downie, CEO of colocation data center provider Flexential, cited similar numbers on data center power usage when speaking at Bisnow's DICE East event in May. 

“The retail customer really only consumes probably 60% of the power that they have to, but they subscribe to the 100% and you have to have the 100% just in case they ever use it, and they never do," he said. "So, maybe not the case in every one of our facilities, but there is 30 to 40% of power availability that goes underutilized, and probably will for the rest of time, if you don't think about the business differently.” 

Tenant agreements with providers guarantee that their full leased capacity will be available to them at all times whether they use it or not, so that extra power can’t be allocated elsewhere. The result is hundreds of megawatts effectively wasted even as developers wait years for new capacity to be added to the grid. 

“It all starts from users, who say 'by gosh, we are leasing 10 megawatts, we want to see 10 megawatts of capacity, and if in fact you don't have it at any point, we're going to impose these [penalties] that are just going to lay you bare,'” said Tim Bright, executive vice president at T5 Data Centers. “And we can say that we know they’re at most going to use eight-tenths of their allocated power, so we could use this capacity go satisfy another tenant, but they just will never go there.”

So why are data center tenants leasing so much more capacity than they need?

A certain amount of excess capacity is typically necessary, experts say. Companies lease extra capacity as a buffer to ensure their IT infrastructure can handle expected growth and that unexpected surges in use won’t exceed their infrastructure’s capabilities. The consequences of exceeding leased capacity can be significant, from lost business to expensive outages and downtime.

But the buffers tenants are building into their data center deployments are beyond excessive, experts say, the result of an inefficient procurement process and a lack of coordination within tenant organizations. Procurement teams calculate needed capacity based on estimates of expected IT load from different teams within their organization, all of whom have incentives to minimize downtime risk and thus build significant buffers into their estimates.

These buffers of extra capacity end up being effectively layered on top of each other, with the procurement team adding an additional buffer of its own. The result is leases of far more capacity than will ever be needed. 

Cato’s Nelson pointed to an incident during his four-year tenure at Uber, where he headed one of the company’s key infrastructure teams, that he said opened his eyes to the inefficiency that resulted from this process. The company experienced a massive failure of half of its data centers, triggering a failover mechanism that meant Uber’s entire global workload was suddenly running only in the company’s East Coast data centers. It was exactly the kind of worst-case scenario — a catastrophic event resulting in a surge in power use — that buffers of additional capacity were meant to account for.

Yet far from being strained by this increased workload, Nelson said power use in those East Coast data centers only jumped 10%, a fact he said demonstrates that a significant percentage of the company’s leased data center capacity was effectively unnecessary.  

"Just think about that: 600-plus cities of food and people moving around, and all that capacity failed over to one site, and we had less than 10% energy increase in the east,” Nelson said. “That means there's just buffers on buffers on buffers and buffers of capacity.”

Although tenants demand more power capacity than they need, Nelson said that colocation providers, for the most part, have also been slow in identifying stranded capacity as a problem that needs to be addressed, even as power constraints plague the industry.

There continues to be a prevailing sentiment in the industry, Nelson said, that as long as every last megawatt at a facility is leased, whether the tenant uses that power or not has no direct impact on the landlord’s bottom line.

But Nelson and others said this perspective is short-sighted. In addition to allowing data center operators to provide solutions to more tenants in more places, reducing stranded capacity would also reduce capital costs and improve margins by lowering the amount of new construction and new infrastructure needed for each new megawatt customers actually use, experts said. This would have the added impact of improving sustainability metrics in an industry increasingly under public scrutiny for its environmental impact. 

Yet none of the experts who spoke with Bisnow foresee any significant decrease in stranded capacity happening anytime soon. 

“No tenant wants to give their extra capacity up, and no operator, in part because they get paid based on that figure, wants to stand up and tell customers that they can't have it,” said Jeffrey Moerdler, a longtime data center and telecom attorney and member at Mintz. “It's an institutional issue: People need to figure out how to be more efficient and to better identify how much capacity they actually need."

Any significant change, experts said, will come from providers making fundamental changes to how they lease data center capacity. At Cato, Nelson has implemented an alternative model in which available power is oversubscribed to tenants, with different price points determining which tenants have access to extra power when they want it. It’s a leasing model that Nelson compares to how airlines sell tickets. 

“Tenants doing AI, machine learning and that kind of technologies, they can be like standby passengers where they get a discount, but if there’s a power problem they’re going to be shut off, because that’s all they need,” Nelson said. “Right now, we're treating everyone as a first-class customer where you’ve reserved a seat and you get it whether you use it or not.”

But Nelson acknowledged that this kind of alternative model will be slow to catch on, with few data center providers offering any alternatives to traditional leases that could help eliminate stranded capacity. 

“We have a chicken-and-egg scenario,” he said. “[Colocation providers] are not offering it because they're too conservative, and they're saying the customers don't want it, while the customers are not asking for it because they don't know it exists."