Flex: The Data Center Demand Response Broker

You know the conversation. You've been in it. Every utility exec, every hyperscaler PM, every energy VC is circling the same question: where do we find more power for data centres? New generation. Nuclear. Geothermal. Stranded connections. It's all supply-side thinking. All of it.

Then Duke dropped a study that reframes the whole thing. If data centres flex their load for 50 hours a year - about two hours scattered across peak days - the US grid can absorb 100 GW of new demand without a single new power plant. That's $150 billion in infrastructure nobody needs to build. (Google figured this out already. Signed 1 GW of demand response contracts. But they own the full stack - facilities, workloads, utility relationships. Everyone else is stuck.)

100 GW

unlockable with flex

$150B

in avoidable infrastructure

50 hrs

annual curtailment needed

The orchestration layer between utilities, colo operators, and tenants. The thing Google built internally that nobody else has.

Here's why Google can pull off demand response at 1 GW scale: they own everything. The facility, the workloads, the utility relationship. One company, one decision. But over 70% of data centre capacity is multi-tenant colocation - Equinix, Digital Realty, CoreSite, hundreds of smaller operators. These buildings house thousands of tenants running workloads the operator can't see, let alone control.

The colo operator can cycle HVAC and fire up backup generators. That gets you 10-15% load shed. Fine for a press release, not for a grid programme. The real flexibility - the compute itself - sits one layer up, with enterprise tenants who have no relationship with the utility and zero incentive to participate.

So you've got three parties who all want the same thing and can't coordinate. The utility needs controllable demand. The colo operator wants faster interconnection and cheaper rates. The tenant wants a lower power bill. The economics line up perfectly. The product connecting them? Doesn't exist.

The play: a three-sided flexibility broker for multi-tenant data centres. Utilities get a controllable load resource they can dispatch. Colo operators qualify for flex tariffs and jump the interconnection queue. Tenants earn credits on their power bills for workloads they didn't need running at 3pm on the hottest Tuesday in August.

Critical Uncertainty

Will data centre demand flexibility become an economic requirement enforced through tariff design - or will it stay a voluntary sustainability exercise that most operators ignore?

Our Belief

Flex is about to become mandatory, not optional. There are 77 large-load tariffs pending across 36 US states. A year ago that number was 19. Xcel just proposed a new rate structure in Colorado that ties interconnection speed directly to load flexibility. DOE directed FERC to fast-track flexible load interconnection with a 60-day study timeline. The regulators have picked a side. Data centres that flex will connect faster and pay less. Those that can't will wait in line and pay a premium. Within two years, flex capacity will be priced into every colocation lease.

The Crux

Testable in 30 days. Will colo tenants actually opt into programmatic workload flexibility for a 15-25% power cost cut? The test: deploy a lightweight flex agent with 5 enterprise tenants at one colo facility. Trigger 3 demand response events during peak grid stress. If 3 of 5 participate and shed 20%+ of compute load, the multi-tenant problem is solvable and this is a multi-billion-dollar platform. If they won't flex at any price, the business is colo-operator-only - HVAC and backup gen cycling - and the TAM shrinks by an order of magnitude. Either answer is useful. That's the point.

1 GW

of demand response signed by Google with US utilities

Google / DCD, Mar 2026

$150B

in avoidable grid infrastructure if data centres flex

Duke Nicholas Institute, Feb 2026

large-load tariffs pending across 36 US states

Utility Dive, Apr 2026

EPRI DCFlex demo sites testing geo-shifting and load flex

EPRI / PRNewswire, Feb 2026

Trace the thread. Google signed demand response contracts with Entergy Arkansas, Minnesota Power, DTE Energy, TVA, and Indiana Michigan Power - all targeting ML workloads specifically. EPRI's DCFlex is testing compute load shifting from Ashburn to Chicago during grid congestion, with Compass, NVIDIA, Google, and Oracle at the table. Xcel Energy proposed a Colorado tariff (April 2026) that creates explicit rate tiers for flexible data centre loads. And DOE directed FERC to finalize large-load interconnection rules by April 30, 2026, floating a 60-day fast-track for flexible loads. The regulatory, economic, and technical pieces are converging. The product that stitches them together for multi-tenant environments? That's the gap.

We built it. Click through.

A 4-step walkthrough - from the broken value chain to the product to the 30-day crux test.

Explore the Demo

Regulatory

Utility Dive, Apr 2026

77 large-load tariffs pending in 36 states. A year ago the number was 19. Utilities aren't asking data centres to flex. They're telling them.

The shift from "how do we serve this load" to "on what terms" happened faster than anyone in the room expected. Xcel's Colorado proposal is the template: flex your load, get faster interconnection. Don't flex, pay a premium and wait. Rate design as industrial policy. The data centres that treat flexibility as an operational capability - not a favour to the grid - will have a structural cost advantage over everyone still treating it as optional.

Signal

Google / DCD, Mar 2026

Google signs 1 GW of data centre demand response. The interesting part isn't the number. It's that nobody else can replicate it.

Google built this because they control the full stack - facilities, workloads, utility relationships. One company, one decision. Equinix can't do this. Digital Realty can't do this. Their tenants' workloads are invisible to them. The 1 GW milestone proves the economics work beautifully. It also proves the gap: the orchestration layer between colo operator and tenant doesn't exist outside Google's walls. That's the whole opportunity in one stat.

TECHNICAL

EPRI / PRNewswire, Feb 2026

EPRI's DCFlex tests geo-shifting compute from Ashburn to Chicago during congestion. Works technically. Has no business model.

Nine demo sites. Compass, NVIDIA, Google, Oracle at the table. EPRI confirmed flexibility exists in compute, HVAC, and on-site power - but "not as a single, uniform capability." The engineering is done. What's missing is who pays whom, how flex gets measured, and how value flows back to the tenant who actually shed the load. That's a product problem, not a research problem. And nobody's building it.

RESEARCH

Duke Nicholas Institute, Feb 2026

Duke's study: 50 hours of curtailment per year unlocks 100 GW of grid headroom. The math is settled. The incentive structure isn't.

The $150B in avoidable infrastructure is mostly gas plants and transmission lines that utilities would otherwise bill to ratepayers. Think about that for a second - utilities have a $150B reason to not make flexibility easy. The venture that cracks this has to build a model where the utility gets paid for flexibility value, not penalised for building less. That's the real design challenge, and it's economic, not technical.

OpenGridWorks

Public data for public use

Energy infrastructure data and visualization that sparks joy. Map of US transmission, data centers, substations, and rights-of-way.

Open the tool

Sometimes the most contrarian move isn't finding more of what everyone's chasing - it's making better use of what's already there.

Talk to Limineer

One opportunity. Every two weeks.

Pulse delivers one researched venture opportunity in energy - with evidence, methodology, and a clickable prototype. No roundups.

Share this issue

Pass it to someone working on the same problem.

LinkedIn X Email