The Company

PubNub runs the real-time messaging infrastructure behind a ton of apps you use without knowing it: think delivery tracking, smart locks, live leaderboards. Trillions of API calls a month, five global regions, all on AWS. A small FinOps team, and huge variability in traffic, since one customer's app can go viral overnight while another runs a scheduled product drop.

The Triangle of Pain

Hamilton called it a "triangle of pain," made up of three compounding problems:

• Workloads that are basically unforecastable more than a few months out
• A small team that can't spend its time debating whether 3,000 instances will be needed for 15 days or four years
• The two-sided fear of wasted spend: pay full price on-demand, or commit long-term and risk sitting on a six-figure reservation nobody uses

That last risk wasn't hypothetical. Right after PubNub locked in a multi-year commitment, engineering rebuilt a service in Rust and made half of it unnecessary.

What PubNub Actually Needed

Not another one-to-three-year cloud provider commitment. PubNub wanted flexibility in the form of shorter terms, risk transfer so someone else absorbs the downside if usage drops, a process that didn't route every decision through engineering, real savings rather than 1-2%, and something simple enough for a lean FinOps team to run.

Where Insured Commitments Fit In

Archera's Insured Commitments gave PubNub 30-day guaranteed terms, all at the billing layer, so no infrastructure changes and no engineering tickets. If usage drops and a commitment goes underused, Archera's guarantee absorbs that risk instead of PubNub eating it, through either a Rebate Guarantee or a full Release.

The pricing mechanic: Archera buys the standard 1-3 year reservation, then charges a risk premium (roughly 9-13%, lower for longer terms) for the ability to break out early, and passes the rest of the discount to the customer. On a 30-day Guaranteed Reserved Instance (GRI), that nets out to about 30% savings even after the premium.

See how a guaranteed-term structure could work for your own commitment mix. Schedule a demo →

The Three-Layer Framework

PubNub landed on a three-layer stack for matching commitment type to confidence level:

• Base layer (~70-75% of infrastructure): Steady, predictable workloads, locked into one-year GRIs or native long-term commitments
• Middle layer (~15-20%): Known-but-uncertain spikes, like seasonal events, new product rollouts, and parallel deployments, covered with 30-day guaranteed terms since the risk is fully transferred
• Top layer: True on-demand. The stuff nobody can predict, like viral moments and surprise traffic, left uncommitted on purpose

As patterns become clear, for example a customer sticks around or a product proves durable, PubNub converts middle-layer 30-day commitments into base-layer one-year terms.

The Results

On-demand spend dropped from 20-15% down to a sub-5% target. Over half a million dollars saved in year one alone. And maybe the bigger win: the multi-hour "how much should we commit" meetings involving the CEO, CFO, and engineering leads mostly went away, since smaller, reversible commitments don't need that level of deliberation anymore.

What This Means for Your Team

You don't have to choose between overpaying on-demand and gambling on a long-term commitment. Layer your commitments by confidence level, and let a shorter guaranteed term absorb the risk on anything you're not 100% sure about.

‍

Read the full PubNub story

Or 

Watch the full session

‍