Render's $1.5B Valuation and a 20× Bandwidth Cut in the Same Quarter: Who $0.15/GB Egress Is Actually Priced For

On February 17, 2026, Render announced a $100 million extension of its Series C at a $1.5 billion valuation — total funding $258 million, led by Georgian with Addition, Bessemer Venture Partners, General Catalyst and 01 Advisors — to build what it calls the cloud for AI-native software. Sixty-five days later, on April 23, it shipped new workspace plans that cut included bandwidth on the Hobby tier from 100GB to 5GB, trimmed Pro from a generous effective 100GB-era allowance to 25GB, and set a uniform $0.15/GB overage with a hard August 1, 2026 deadline forcing every legacy workspace onto the new terms.

A company posting its best growth numbers in the same quarter it makes its free tier 20× smaller looks like a contradiction until you price the bandwidth. It is not. And the number that settles the argument is mercilessly simple: on Hobby, the 95GB you used to get for free now costs $14.25 a month in overage alone. On a Hetzner CX22 it costs $0 — because the same box already includes 20TB flat for €3.79 a month whether you use 5GB or 5,000.

The full curve is below, early, so you can find your own row before you read the why.

The math, first: $0.15/GB vs 20TB flat​

This table is all-in cost per month: workspace fee + overage. Hobby is $0 with 5GB included, Pro is $25 with 25GB included, Scale is $499 with 1TB included. Overage is $0.15/GB past the included allowance. Hetzner is a single CX22 (€3.79/mo) or CPX42/CAX-range equivalent at roughly €15.99/mo — both include 20TB with no per-GB meter. For context, €3.79 is about $4.10 at current rates; the gap below is not a currency trick.

Two breakevens matter:

• Hobby is cheaper than a Hetzner box only at ≤5GB. That is a personal site, a staging preview, a webhook receiver — the bandwidth a Hobby project can actually stay inside without thinking about it. At 25GB — one modest Next.js site with images and an API — Hobby's $3 overage is nominally cheaper than €3.79, but you are already metering; one viral post and next month is $14.25.
• At 100GB — the old Hobby allowance — the self-hosted box is 3–4× cheaper on bandwidth alone, before you count that the box also runs the compute. At 500GB (a small SaaS docs site, a mobile API with image responses) the gap is 19×. At 1TB it is 40× vs Hobby and still 3× cheaper than Scale's $499 anchor plan.

The sensitivity is the point. Double your egress and the Hetzner number does not move. Double it on Render and you add $15, $60, or $150. The variable that growth dials is the one your flat box already amortized.

A note on what this table is and is not: it prices bandwidth + workspace fee. Render compute is billed separately — Hobby services start at $7/mo per instance, autoscaling is Pro-only — so a real deploy adds that line to every Render column. The Hetzner side is also not free: you pay the VM, you operate the cluster, you patch, you run backups. The delta above is strictly the egress tax. For bandwidth-heavy workloads it is already the largest line; for bandwidth-light ones the Hetzner ops burden can outweigh €3.79 in saved overage. Section 5 maps that trade honestly.

What actually changed on April 23​

Render's April 23, 2026 changelog and New Workspace Plans docs are precise and worth reading line by line, because the cut is not just a number — it is a plan-architecture change with a migration deadline.

Before: legacy workspaces ran on a seat-based model where Hobby included roughly 100GB of outbound bandwidth before any overage logic mattered for most hobby projects. Teams paid per seat, not per workspace, and the bandwidth number was generous enough that a docs site or side-project API rarely thought about it.

After, April 23 onward:

Plus: legacy workspaces that do not migrate by August 1, 2026 are force-migrated to the new tiers, and new Hobby workspaces can create multiple projects to group services — a nicer ergonomic wrapper around a materially smaller free allowance.

The $1.5B part needs its own dated source because the TODO's "same quarter" claim is only interesting if the two events are actually contemporaneous and not quarters apart:

• February 17, 2026 — Render announces a $100M Series C extension at $1.5B valuation via Business Wire, picked up by VentureBeat, PYMNTS, and SiliconANGLE. Lead investor Georgian (which also led the original Series C), with Addition, Bessemer Venture Partners, General Catalyst and 01 Advisors participating, total funding to $258M.
• February–April 2026 — the raise is framed around building the cloud for AI-native software and long-running agent workloads, on the back of what the company describes as triple-digit year-over-year growth. The April 23 repricing lands 65 days later, inside the same Q1-to-Q2 reporting window the banner quarter refers to.

Same quarter is not narrative embroidery. The two moves happened 65 days apart, under the same funding announcement that valued the company on its growth rate.

Why record growth and a 20× cut are not in tension​

The instinct is to read the two headlines as opposed — if Render is growing over 100% year-over-year, why would it squeeze hobby bandwidth? The answer is that the growth is in spite of Hobby bandwidth, not because of it. Hobby bandwidth at 100GB was a below-cost loss leader.

Put numbers on the subsidy. Transit runs roughly $0.05–0.09/GB at retail before CDN; PaaS egress at $0.15/GB covers that plus margin. A single Hobby project pulling its full old 100GB free allowance cost Render roughly $5–9 in raw bandwidth — $15 at the price it now charges. Multiply by thousands of hobby projects serving image-heavy portfolios or uncached API JSON and the free tier becomes the largest uncapped variable cost on the statement. It subsidizes bytes, not seats.

When a company raises at $1.5B on triple-digit growth, that growth is underwritten by paying workspaces — Pro and Scale. Every uncapped Hobby gigabyte that does not convert is margin those paying tiers cover. Cutting Hobby from 100GB to 5GB and Pro's buffer to 25GB (with $0.15/GB thereafter) does three things at once:

1. Caps the uncapped line. 5GB keeps only sub-threshold experiments free; anything real hits the meter in week one.
2. Makes the funnel pay. A team shipping 80–150GB on Hobby discovers on its first invoice that 80GB already costs $11.25 in overage — and that Pro at the same 80GB costs $33.25 ($8.25 of Hobby overage vs $8.25 + $25). Pro is positioned for autoscaling and unlimited services, not bandwidth savings.
3. Protects the valuation thesis. At $1.5B, value accrues to being where AI agents and apps run, not where they park free static assets. Bandwidth is the worst wedge for that thesis.

None of this is cynical. It is how a generous PaaS discovers the price at which generosity scales. The 20× cut is the moment Render priced Hobby like a funnel, not a slab of infrastructure.

Who is it actually priced for?​

The table in section 2 answers "how much." This section answers "for whom" — three archetypes that actually map to the GB rows.

1. The experiment (<5GB) — Hobby at $0 is still the best deal​

A personal site, a weekend API, a cron that pages you, a preview for an interview take-home. Under 5GB of egress a month — HTML and tiny JSON payloads, no images served from the PaaS, no download endpoint — Render Hobby costs literally $0 plus whatever compute you attach. A Hetzner box at €3.79 is more expensive here and you have not yet needed the 20TB you bought. If your project lives here, stay on Hobby. The metering change does not touch you.

Ops burden: none. Git push. Done.

2. The early SaaS landing page and docs (10–40GB) — the awkward middle​

A marketing site with optimized images, an OpenAPI docs site, an early product with 1,000–5,000 active users fetching 200KB–1MB per session. This workload used to be comfortably free under the old 100GB Hobby allowance. Now:

• Hobby: 10GB → $0.75, 25GB → $3.00, 40GB → 35GB × $0.15 = $5.25
• Pro: flat $25 at any of those sizes (still under 25GB or just over), but $25 is 5–30× the Hobby overage at the same sizes
• Hetzner CX22: €3.79 flat

Add compute: one Starter web service ($7/mo) plus Hobby Postgres ($7/mo) turns a $3 Hobby bandwidth bill into ~$17 all-in. The Hetzner side runs the same stack on the same €3.79 box, with headroom.

When Render still wins: bursty traffic that would be mostly idle on the box, or a team with zero appetite for patching. A CX22 at 90% idle still costs €3.79; a Hobby service that sleeps 20 hours a day pays only for CPU-seconds actually drawn plus bandwidth. Truly sleepy workloads can undercut the always-on box.

When self-hosting wins: steady low-rate traffic with predictable image or API egress. At a sustained 30GB, Hobby is $3.75 in overage plus ~$14 of compute — call it ~$20–25 all-in; Hetzner is still €3.79 with 19.97TB of headroom. One evening to stand up the cluster, then minutes per month.

Double that from 30GB to 60GB and Hobby goes $3.75 → $8.25 (+$4.50) while Hetzner stays €3.79. That fixed-vs-variable shape never stops mattering.

3. The bandwidth-heavy app (100GB–5TB) — the PaaS meter's worst fit​

Image-heavy portfolios, file exports, CSV/PDF generation, mobile APIs serving 1–5MB responses, AI image endpoints, public dataset mirrors. These are not edge cases; they are the reason PaaS bandwidth pages exist at all.

• 100GB: Hobby $14.25, Pro $36.25, CX22 €3.79 → self-host 3.8×–9.6× cheaper on egress alone
• 500GB: Hobby $74.25, Pro $96.25, CX22 €3.79 → self-host ~20–25× cheaper
• 5TB: Hobby $767, Scale $1,117, CX22 €3.79 → self-host 200×+ cheaper

Sensitivity to 2× and 10× is the killer detail. If your API payload doubles from thumbnails, Hobby egress at 500GB → 1TB goes $74 → $153 (+$79) while Hetzner stays €3.79. If a Hacker News hit 10×'s traffic for a day, the Render invoice multiplies with it; the fixed box absorbs the burst until CPU saturates — a scaling decision you control, not a bill that arrives after the cache.

Self-hosting is not free. Budget 2–4 hours a month for patches, backups, and cert renewals, plus $5–20 for object storage or managed Postgres if you do not co-locate the database. Even at $50/hr, 3 hours is $150 — still cheaper than the $74–153 you avoided at 500GB–1TB, and dramatically cheaper at 5TB. At 25GB and below, that labor line is exactly what Render's ops-free Hobby was priced to eliminate.

The honest map:

• <5GB: Hobby $0 — PaaS wins on both price and ops.
• 5–30GB, steady: roughly parity on price; decision is ops appetite and whether you want a box to grow into.
• 5–30GB, bursty/idle: Hobby/Pro can undercut the always-on box.
• 100GB+: flat wins on price by multiples that make ops cost rounding error; the PaaS meter is paying for convenience, not infrastructure.

That is what "who it is actually priced for" means in practice. The 20× cut did not raise prices for everyone; it raised them for the workloads PaaS bandwidth was never subsidized to carry — the ones where owning the box was always the cheaper physics, now laid bare on the invoice.

Before August 1: what to do with a legacy workspace​

If you have a legacy workspace that has not yet migrated to the April 23 tiers, August 1, 2026 is not a suggestion. Render's docs mark it as a forced migration. Do three things before then, each taking about ten minutes:

1. Check your actual egress. In the Render dashboard, open your workspace → Usage → Bandwidth for the last 30 days. If the number is under 5GB on Hobby or 25GB on Pro, the migration is cosmetic. If it is 40GB, 100GB, or 400GB, multiply the excess by $0.15 and put that number next to your workspace fee. That is your new baseline for next month.
2. Decide on tier vs topology. Under 5GB, stay Hobby. In the 10–60GB steady band, price Hobby overage vs a single Hetzner CX22/CAX11 at €3.79 with Postgres co-located, and be honest about how much you want to operate. Past 100GB, price the Hetzner box as infrastructure you already need to own — because at that volume you do.
3. If you self-host, do it on infrastructure you control with the API shape your agents already speak. Bex.co's Cluster-API-backed fleet provisions real Hetzner machines — not a VM slice on someone else's control plane — and exposes a Render-compatible API plus an open MCP server so git-push and agent-operated deploys work the same day you cut over. The egress that cost $74 on Hobby or $96 on Pro at 500GB is still €3.79 flat because the bandwidth is the box's, not a line item. You trade a per-GB meter for a machine you own and a control plane you can grant an agent access to without granting it your PaaS account.

Bandwidth has become the PaaS margin lever the way storage was five years ago — zeroed on the marketing page, metered on the invoice, grown quietly until a repricing makes the subsidy visible. Render's repricing made it visible on April 23. August 1 makes it unavoidable.

The $1.5B valuation and the 20× cut are not opposite stories. They are the same one: the more a PaaS scales on paying workloads, the less it can afford to give away the bytes those workloads ride on. The question is not whether the cut is fair — it is the rational way a high-growth hosted platform prices. The question is whether your bytes still belong on a per-GB meter at all.

Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own. Star the repo on GitHub or deploy your first app today.