![[headshot] image of customer giving a testimonial (for a ai biotech company)](https://cdn.prod.website-files.com/690cdf62418694e59d87cccf/695d40f0cb2a2d98b5650aa7_cat-bw-outline-profile.png)
Energy is suddenly everyone's problem. US electricity prices rose 6.9% last year — more than double inflation — and Goldman Sachs projects data centers will drive roughly 40% of electricity demand growth through 2030. Wholesale power costs in PJM, the largest US grid market, jumped 76% year-over-year in the first quarter. Communities are blocking new data center builds, regulators are stepping in, and "where will the power come from" has become a board-level question across the entire digital economy. Most of that conversation is about AI. But there's a part of it that's about you — specifically, about how your analytics stack is built.
Over the past decade, millions of companies built the same machine.
A cloud data warehouse, provisioned and configured in-house. A set of ETL pipelines, scheduled to run whether or not anyone reads the output. A BI server kept warm around the clock. Staging environments. Dev environments. The dashboard someone built in 2023 that still refreshes hourly and that nobody has opened since.
Every business genuinely has its own data. But the machinery — the warehouses, the pipelines, the visualization servers — is structurally identical from company to company, duplicated millions of times over, and mostly idle. That was always expensive. It's now becoming something else: a liability, because the resource all of that machinery runs on is getting scarce.
Compute Just Became a Constrained Resource
For most of the cloud era, compute was treated as effectively infinite — you could always provision more, and power was someone else's problem. That assumption no longer holds.
According to the International Data Center Authority's May 2026 report, data centers now consume roughly 6% of all US electricity, with annual global data center spending approaching $1 trillion. A peer-reviewed study published in Environmental Research Letters in May found that data centers' share of US electricity use more than doubled between 2018 and 2023 — from 1.9% to 4.4% — and modeled wholesale electricity costs rising as much as 29% nationally by 2030.
The constraint is already showing up in prices. Goldman Sachs analysts reported that US electricity prices rose 6.9% in 2025 — more than double the rate of inflation — and project data centers will account for roughly 40% of electricity demand growth through the end of the decade. In PJM, the largest US grid market, the independent market monitor reported wholesale power costs jumped nearly 76% year-over-year in the first quarter of 2026. And the buildout itself is hitting walls: in Northern Virginia's "Data Center Alley," developers face waits until 2032 for new project permits, while "time-to-power" delays are stretching data center launches by 1.5 to 2 years industry-wide.
The headlines focus on AI training runs and hyperscale campuses, and that's where the biggest numbers are. But there's a detail in the IDCA report that should interest anyone who runs a business, not a data center: smaller server rooms embedded in corporate buildings and regional offices account for at least 15% of total data center power consumption — and they're considerably less efficient than the large facilities, precisely because they're fragmented and underutilized.
That's the analytics duplication problem, measured in watts. Lots of small, separately owned machinery. Mostly idle. Collectively enormous.
Idle Infrastructure Is the Hidden Cost
The quiet weakness of self-managed analytics infrastructure is utilization. A data warehouse provisioned for your Monday-morning reporting peak is wildly overpowered for the rest of the week. A self-hosted BI server can't tell the difference between a Tuesday full of executives refreshing dashboards and a Sunday night serving nobody — it draws power, and bills, either way.
So the real question for any data stack isn't "how much compute does analytics need?" It's "how much of the compute we've provisioned is doing useful work right now?" For most standalone stacks, the honest answer is: not much.
Shared, managed infrastructure changes that math. When many teams run on one well-operated platform, peaks and valleys average out, compute scales to actual demand instead of worst-case provisioning, and capacity that would sit idle across a thousand separate deployments gets used. It's why a city bus moves people more efficiently than a thousand idling cars — not magic, just pooled capacity that actually gets utilized.
Why Nockpoint Runs as One Platform, Not a Thousand Stacks
This is one of the quieter reasons we built Nockpoint the way we did.
Nockpoint runs on Snowflake — a deliberate architectural choice we've written about earlier in this series. One of Snowflake's most underrated design decisions is the separation of storage from compute: your data sits cheaply at rest, and compute spins up when there's a query to run and suspends when there isn't. You don't pay — in dollars or in watts — for a warehouse to exist. You pay for it to work.
On top of that, Nockpoint is managed and multi-tenant by design. Instead of every customer provisioning their own warehouse, building their own pipelines, and keeping their own BI servers warm, everyone runs on one platform that's operated, tuned, and right-sized continuously. Your data stays yours. The machinery is shared.
Customers feel that directly: no infrastructure to babysit, no separate Snowflake bill, one platform with both Power BI and Apache Superset built in as visualization layers. The part that never makes a feature list is just as real: dramatically less duplicated, idle compute per company served.
We didn't set out to build a sustainability product, and we won't pretend a business intelligence platform is what fixes the grid. We set out to build the most efficient path from raw data to answers — and it turns out efficiency is efficiency, whether you measure it in dollars, engineering hours, or kilowatt-hours.
The Era of Right-Sized Analytics
Here's where we think this goes.
For a decade, the prestige move in data was to build big — more pipelines, more tooling, more infrastructure, more headcount to run it. That era assumed compute was cheap and unlimited. The numbers above say otherwise, and as power constraints reshape what data centers cost to build and run, the cost of compute flows downstream to everyone who buys it.
In that world, the advantage shifts to right-sized analytics: stacks that scale with actual usage, share infrastructure where sharing makes sense, and spend compute on answering questions rather than on existing. The payoff isn't abstract — it's a lower total cost of ownership, fewer moving parts to maintain, and an analytics stack that gets faster to change instead of heavier every year.
The most efficient query is the one you don't run twice. The most efficient warehouse is the one that sleeps when you do. And the most efficient stack is the one you never had to build at all.
That's the one we run for you — starting at $50/team/month.
Sources:
International Data Center Authority (IDCA) US data center consumption report, May 2026
Environmental Research Letters, May 2026
Goldman Sachs Research, February 2026
Monitoring Analytics, PJM State of the Market Report Q1 2026
Congressional Research Service, Data Centers and Their Energy Consumption (R48646, updated May 2026)
