How Rising Data Center Costs Are Driving SASE & SSE Price Increases

The data center market, in 60 seconds

The US data center market isn’t “getting tight.” It’s full.

Vacancy rates in primary North American colocation markets dropped to 1.4% at the end of 2025. That’s the lowest anyone’s ever recorded. At the same time, demand is accelerating: US data center power consumption hit 61.8 GW in 2025 and is on track to reach 75.8 GW in 2026. That’s a 23% jump in one year.

What’s driving it? AI. Every ChatGPT query, every Claude conversation, every Copilot suggestion runs on GPU-dense racks that chew through 3x to 10x more power than regular cloud workloads. AI is expected to eat up 44 GW of US data center demand in 2026, nearly matching the 38 GW that everything else uses combined.

Here’s the kicker: AI companies are willing to pay whatever it takes and lock in long-term deals for every available rack. That pushes everyone else out, including the cloud-proxy security vendors whose whole business model depends on having more and more data center space.

Why this is your SASE/SSE vendor’s problem

Cloud-proxy vendors (Zscaler, Cisco Umbrella, Netskope, Forcepoint) all have the same dependency baked into their architecture: they route all your traffic through their data centers for inspection. Zscaler runs 150+ data centers globally. Netskope has its NewEdge backbone. Cisco Umbrella pushes everything through Cisco’s global PoPs.

Every new customer they land needs more data center capacity. Every user they add generates more traffic that needs more PoP infrastructure. It’s basically a straight line: more users, more traffic, more racks, more money.

Back in 2021, the average colocation rate was about $120 per kW per month. By the end of 2025, it hit $196. That’s 63% more in four years. And the curve is getting steeper, not flatter.

When your SWG vendor’s infrastructure costs jump 63%, that money doesn’t just vanish. It probably ends up on your renewal.

Here’s the math that should keep you up at night

This is how the cost pass-through works:

Electricity prices are up 40% nationwide since February 2020. In Virginia (the biggest data center market on the planet), electricity costs have spiked up to 267% over five years. Utilities asked for $29 billion in rate increases in just the first half of 2025 to pay for grid upgrades driven by data center demand.

Those electricity costs get rolled into colocation pricing. Colocation providers pass their infrastructure and power costs to tenants through lease increases. That $196/kW-month number? It’s heading past $220 by 2027.

Colocation costs get rolled into vendor operating costs. Zscaler’s gross margin already squeezed from 78% to 77% because of rising data center expenses. That pressure is only going to pick up as electricity and colocation keep climbing.

And vendor costs get rolled into your subscription price. That’s where the 35%+ increase came from.

The things driving all of this (AI demand, power grid bottlenecks, construction delays) aren’t letting up. They’re getting worse. Almost half of all US data centers planned for 2026 have been canceled or pushed back because the power grid can’t keep up. The projected supply gap hits 10 GW by 2028. Electricity is expected to climb another 40% by 2030.

Your vendor’s next price hike isn’t an “if.” It’s already baked into the numbers.

Building more data centers won’t fix it

You’d think with all the money flowing in, the industry would build its way out. Not quite.

The investment numbers are massive. The four biggest hyperscalers (Alphabet, Amazon, Meta, Microsoft) committed around $650 billion to AI infrastructure in 2025-2026. The Stargate Project announced $500 billion over four years. Construction starts hit $25.2 billion in January 2026 alone.

But the actual pipeline? It’s cracking under real-world constraints.

Getting a new power grid connection takes 3 to 5 years. The US only makes about 20% of the large power transformers it needs domestically. China controls roughly 60% of global production. A single transformer takes 18 to 24 months to build and ship. On top of that, communities in places like Northern Virginia are fighting new construction, adding permitting delays.

The bottom line: about 7 GW out of 12 GW of data center capacity announced for 2026 won’t show up on time. The Stargate Project, despite all that $500 billion press, has barely broken ground as of April 2026.

Money isn’t the problem. Physical infrastructure is. And that means the shortage (and the price increases it causes) will stick around through at least 2028-2030.

What your SSE/SASE bill looks like over the next 3-5 years

Here’s how this plays out if you’re on a cloud-proxy SSE stack:

2027: AI workloads pass non-AI workloads for the first time. Colocation rates blow past $220/kW-month. If you’re not willing to pay AI-level premiums for data center space, you get squeezed. Mid-market SSE customers feel the pinch first.

2028: Some of the delayed construction from 2026 starts coming online, but it’s not enough. The 10 GW supply gap is still there. Your SASE vendor has a choice: eat the costs (more margin compression) or pass them to you (another price hike). History says they’ll pass them to you.

2029-2030: Data centers now consume 10-12% of all US electricity, up from 4.4% in 2023. Power prices are 40%+ above 2025 levels. The market settles into a new normal, but it’s a permanently more expensive normal. If your security architecture depends on scaling data center infrastructure, you’re paying double or more for that capacity compared to 2024.

Do the math for a mid-market company paying $300,000 a year today. Cumulative increases of 40-70% over five years push that to $420,000 to $510,000. Same product. Same thing it was doing before. Just pricier, because the stuff underneath it costs more.

There’s an architecture that doesn’t have this problem

dope.security’s Fly Direct architecture doesn’t need more data centers when it gets more customers.

When your SSE stack runs on the device itself, there are no PoPs to lease. No colocation contracts ticking up every year. No straight-line relationship between customer growth and data center spend. SSL inspection, URL filtering, policy enforcement: it all happens right on the endpoint, using less than 100 MB of RAM. Traffic goes straight from the device to wherever it’s going.

The economics work completely differently:

Cloud-proxy vendor: 1,000 new users = 1,000 users’ worth of new PoP capacity at $196+/kW-month (and rising).

dope.security: 1,000 new users = 1,000 agent deployments on devices your company already owns. Cloud management is shared. Inspection compute is distributed.

As data center costs go up, the cost advantage of running security on the device compounds. Every year colocation rates climb, the gap gets wider. And legacy vendors can’t just flip a switch to fix it. Zscaler can’t move inspection to the endpoint without rebuilding their entire platform from scratch. Netskope can’t walk away from NewEdge. They’re locked into a model that gets more expensive every single year.

dope.security deploys in minutes through MDM. SSL inspection works right away, no certificate headaches, because it all happens on-device. Outreach Health got 99% of devices secured within a week. A Fortune 100 rolled out to 18,000+ devices in record time.

And your renewal next year? It won’t come with a surcharge for the data center capacity crisis.

One question to ask at your next renewal

When your SASE or SSE contract comes up, ask your vendor this: “What percentage of my subscription cost is driven by your data center infrastructure, and how are rising data center costs going to affect my pricing over the next three years?”

If they dodge it, that tells you everything.

The data center crunch isn’t temporary. AI demand isn’t slowing down. The power grid isn’t catching up overnight. Construction isn’t keeping pace. Every single one of those forces pushes cloud-proxy SSE pricing higher.

Your security architecture shouldn’t be a bet that data center economics will get better.