A 90-day reading across fifteen dev infra providers, told through one number, one chart, and one prediction.
The story of Q2 2026 in dev infra is not the story the pricing pages tell. Headline prices on most providers' flagship tiers held flat or moved within a percentage point. The dashboard a developer screenshotted in February would look unchanged today.
But the bill, in many cases, is materially different. Across 15 tracked providers and 1075 detected changes, the most consequential moves happened in the small print: per-GB bandwidth, per-million-request rates, free-tier quotas, and the spread between included pools and the prices charged the moment you exceed them.
This is the second quarter we have observed this pattern. In Q1 we called it “the back-of-the-bill quarter.” In Q2 it is starting to look less like a moment and more like a posture. We expect Q3 to make this explicit — and offer two specific predictions at the end of this report.
“The cheapest way to raise prices on the back-of-the-bill economy is to raise the back of the bill.”— OBSERVATION, FROM OUR 90-DAY LOG
If you tracked headline per-seat prices alone on the seven hosting providers we monitor, you would conclude that hosting got slightly cheaper in Q2. Vercel held Pro at $20/seat. Netlify held Pro at $19/seat. Render shipped a small drop on Standard.
The overage and feature picture is the opposite. Of the seven hosting providers, five moved at least one overage rate upward in Q2. The mean increase, weighted by the included-quota size for that line item, was 14.6%. Bandwidth overage was the single most-changed field in the entire dataset.
The May 17 move on Vercel is the obvious headline, but it isn't the most interesting datapoint. The median has been creeping up for nine months; Vercel just brought their rate back to the top of the pack after lagging for two quarters.
For self-serve customers, overage rates set the marginal cost of one more user, one more deploy, one more viral moment. Headline price is the price of being here; overage is the price of winning. Q2's data suggests providers have learned that raising the price of winning is less visible — and therefore less politically expensive — than raising the price of being here.
The database picture is, in some ways, the inverse of hosting. Headline prices on Pro tiers held steady across Supabase, Neon, PlanetScale, and Convex. But included pools on free and entry tiers shrank — sometimes substantially, almost always quietly.
| Provider | Tier | Field | Q1 close | Q2 close | Δ |
|---|---|---|---|---|---|
| Supabase | Pro | DB compute hours / mo | 730 | 540 | −26% |
| Neon | Launch | Headline $/mo | $19 |
Two specific predictions, with confidence calibration. We will grade these in the Q3 report. We graded Q1's predictions in the previous issue: 3 of 4 correct.
A "bandwidth pack" SKU — pre-paid, sold separately from the per-seat Pro tier — is the cleanest way to recover margin without touching headline price. Fly.io's May 16 "Launch" addition is plausibly the first move in this direction.
CONFIDENCE 0.72Railway has already done it on hosting. Supabase's compute-hour tightening suggests they are reading the same internal numbers. The first DB provider to make this move will define the language others copy.
CONFIDENCE 0.58If you find yourself bill-shocked by either of these moves in the next 90 days, you would not be alone. The point of this report — and of the dashboard it accompanies — is to make sure you are surprised on the day of the change, not on the day of the invoice.
| $24 |
| +26% |
| Neon | Launch | Included storage | 10 GB | 15 GB | +50% |
| PlanetScale | Hobby | Reads / mo | 1B | 500M | −50% |
| Upstash | Pro | $ / M requests | $0.20 | $0.25 | +25% |
| Convex | Free | Function calls / mo | 1M | 1M | flat |
“Headline pricing on databases is a marketing surface. The pricing that matters is the limit table.”