The same fleet, at every company.
Why Baseload Systems is named after a power-grid term, and why we think VM capacity should be treated the same way utilities treat electricity demand.
A decade in cloud operations, seeing the same shape of waste everywhere.
Marisela Fontaine spent a decade running cloud infrastructure teams for a series of mid-sized companies. At every single one, an audit of the Azure VM fleet turned up the same pattern: a handful of machines, sized during a rushed launch or a nervous scaling decision years earlier, quietly consuming a disproportionate share of the compute bill while running at a fraction of their provisioned capacity. Nobody had touched them since, because resizing meant a scheduled restart, and a restart meant a conversation nobody wanted to start.
Baseload is a term borrowed from power utilities, on purpose.
A power grid's baseload is the minimum, steady level of demand it has to meet around the clock — not the rare peak on the hottest day of the year. Utilities plan their steady-state generation around baseload and handle spikes separately, with peaking capacity brought online only when actually needed. Most companies do the opposite with their VM fleet: they provision every machine for the peak that might happen and pay full price for that peak capacity every hour of every day, spike or not.
She founded Baseload Systems in 2018 to bring that utility-industry discipline to Azure compute — size for steady-state, handle genuine spikes deliberately, and stop paying peak prices for baseload workloads.
Four things that haven't changed since 2018.
We size for real usage, not fear.
Every rightsizing recommendation is backed by at least thirty days of actual utilization data — never a guess, and never sized around a hypothetical worst case that hasn't happened in years.
Downtime for a resize gets scheduled, not avoided.
A brief, planned restart during a maintenance window is a manageable cost. An oversized VM running at 7% utilization for two more years because nobody wanted to schedule that restart is not.
Rightsizing is ongoing, not a one-time project.
A fleet that's perfectly sized today drifts within months as workloads change. We review utilization on a standing monthly cadence, not as a single engagement that ends at handoff.
We work exclusively in Azure.
Every recommendation accounts for Azure-specific mechanics — Hybrid Benefit, Reservations, Savings Plans, Spot pricing — rather than generic cloud advice loosely adapted to fit.
Meet the team applying this daily: Our Team, or see exactly how an engagement starts on the How It Works page.