Build Not yet confirmed elsewhere 1 publisher 3 min read Published Aug 22, 2026, 3:00 AM PDT

The weekend YAML platform funds an eighth of the org the working version needs

A vendor-adjacent staffing model prices a do-it-yourself internal developer platform at 60 people and $7.5M a year. The arithmetic it leaves out sits on both sides of the trade.

The Engineer · Build desk

Send a correction

What happened

The New Stack puts the honest cost of running your own internal developer platform at about 60 people and $7.5 million a year, with no end date.
The staffing follows the CNCF platform reference architecture: seven product teams covering infrastructure through developer enablement, at seven to nine engineers each.
At a conservative $125,000 per engineer that is $37.5 million of payroll over five years, before anyone ships an application.
Shops that buy instead report much thinner operate teams: 6,500 developers to 16 ops, 2,500 to 5, 1,200 to 6.

Why it matters

What it costs cost The glue person sitting in each development group is unaccounted for in the business case, so $7.5 million is a floor rather than a total, and the overrun is invisible because it is already hired.
Who has to decide decision Signing the cheap pitch is a commitment to fund six more functions later or to own an orphan, and that choice arrives long after the sponsor has been promoted.
What it limits constraint Because the build total never lands in one place, there is no review at which it can lose on price, which is the only forum where a license number wins.
Who's exposed exposure The recommendation to build usually comes from the people whose promotion packet improves if it is accepted, so the business case and the career case are written by the same hand.

Next story · Product · 3d ago

Akuity's Kargo Enterprise beta promotes one change through a Kubernetes fleet in waves

One report1 publisher

The 60 is an org chart before it is an engineering estimate. Seven product functions, at seven to nine engineers each, with a couple of scrum masters and product owners on top ^[2]^[3]. Divide the payroll by the heads and you get $125,000 on the nose ^[9], which tells you what the model does not carry: no benefits or employer tax load, no tooling budget, no cloud bill, and no premium for the coordinators over the engineers. The real annual number is above $7.5 million, and this model cannot tell you by how much.

The version executives approve is one team, weekend work, a control plane with some YAML over it ^[6]. Against the scoped organisation, that is 12 to 15 percent of the headcount ^[8]. One of seven functions gets funded and the other six do not stop existing. They land on the person in each development group who does the glue between the platform and the applications, which is the line the business case spreadsheets do not have ^[19].

The buy side of the comparison is softer than the build side. The collected ratios include 45 app teams to 5 operators in one shop and 300 app teams to 4 in another ^[18], which is 9 teams per operator against 75, a factor of 8.3 between two samples measured the same way ^[10]. Per developer it runs 200 to one at the tightest and 500 to one at the loosest ^[11]. Sizing an operate team off these means deciding which sample you believe before you start.

The piece stops one line short of the arithmetic that settles it. Five operators at its own $125,000 figure cost $3.125 million over five years, against $37.5 million of build payroll ^[16], leaving roughly $34.4 million of room for licenses, support and migration before buying stops winning ^[12]. That is the number the vendor paper would have to fill in, and the published version does not ^[13]. The ceiling is generous enough that most contracts fit under it, which is presumably why the calculation runs in this direction.

Both sides have a thumb on the scale. The figures come from a VMware Tanzu Platform paper the author helped update, and he flags it in his second paragraph ^[4]. The counterweight is that engineers get promoted for shipping platforms, and "we shipped a platform built on Kubernetes" reads better in a packet than "we onboarded everyone to a thing we paid for" ^[20]. What survives both incentives is the structure. Seven functions have to be staffed whether or not anyone named them in the proposal, and the glue people are already on payroll under other titles ^[19].

What to watch

A license price published against the same staffing model, which would turn the $34.4 million of five-year headroom into an actual comparison rather than a ceiling.

Clarity's read

What the record supports and how the coverage leans. The claims behind it follow.

Reality

Evidence32
Adoption: Insufficient
Hype gap+44
Incentives84
Confidence58

Why these scores

Claim ledger

Ranked by verification strength, evidence, and original report placement.

[1]
Most executives signing off do not see the total, because it is spread across multiple cost centers under "engineering".
Reported Supported2 sources— create a free account to open them View cited source
[2]
Aligning a platform team with something like the CNCF platform reference architecture produces about seven product teams: infrastructure, operations, deployment, runtime and middleware, database, security, and coaching/developer enablement.
Reported Supported2 sources— create a free account to open them View cited source
[3]
Each of those teams is a two-pizza team of 7 to 9 engineers, and adding a couple of scrum masters and product owners brings the total to about 60 people.
Reported Supported2 sources— create a free account to open them View cited source
[4]
The author states the numbers come from a recent VMware Tanzu Platform paper he helped update, and tells readers to flavour them with vendor salt.
Reported Supported Source: thenewstack.io author2 sources— create a free account to open them View cited source
[5]
The visible cost of a commercial platform is one license number on one purchase order, while build-your-own headcount is spread across cost centers, often done in the shadows, and looks like normal hiring, so nobody sums it.
Reported Supported2 sources— create a free account to open them View cited source
[6]
The cheap version of the pitch executives hear is one team, weekend work, and a control plane with a bit of YAML on top; it ships v1 and becomes a maintenance burden nobody wants to own.
Reported SupportedView cited source
[7]
Buy-side teams are smaller because they do not build the platform, avoiding continuing work on APIs, integrations, developer frameworks, security pipelines, dashboards, upgrade tooling, the next AI capability developers want and the next compliance regime auditors want.
Reported SupportedView cited source
[8]
A single two-pizza team of 7 to 9 engineers is 12 to 15 percent of the 60-person organisation the working version is said to require.
Derived Supported2 sources— create a free account to open them
[9]
The $7.5 million annual figure works out to exactly $125,000 per head across 60 people, so the model prices scrum masters and product owners at engineer base salary and includes no benefits load, tooling or infrastructure cost.
Derived Supported1 source— create a free account to open them
[10]
Measured per app team, the collected buy-side ratios range from 9 app teams per operator to 75, a factor of 8.3 between two samples using the same unit.
Derived Supported1 source— create a free account to open them
[11]
Measured per developer, the collected buy-side ratios range from 200 developers per operator to 500, a factor of 2.5.
Derived Supported1 source— create a free account to open them
[12]
Five operators at the source's own $125,000 salary cost $625,000 a year, or $3.125 million over five years, leaving about $34.4 million of the $37.5 million five-year build payroll available for license, support and migration before buying loses on cost.
Derived Supported1 source— create a free account to open them
[13]
The published material quantifies build staffing and payroll to the dollar but supplies no license price for the commercial alternative it recommends.
Derived Supported1 source— create a free account to open them
[14]
Building and operating your own internal developer platform the way most enterprises end up doing it costs about 60 people and $7.5 million a year, every year.
Reported Insufficient Source: thenewstack.io2 sources— create a free account to open them View cited source
[15]
At a conservative salary of $125,000 per year per engineer, the 60-person platform organisation is $7.5 million in payroll every year, indefinitely.
Reported Insufficient2 sources— create a free account to open them View cited source
[16]
Five years in, that is $37.5 million spent on payroll alone for the team that builds the platform, not the applications.
Reported Insufficient2 sources— create a free account to open them View cited source
[17]
Platform team ratios collected by the author for organisations that buy: 6,500 developers to 16 ops, 2,500 developers to 5 ops, 1,200 developers to 6 ops.
Reported Insufficient2 sources— create a free account to open them View cited source
[18]
Further collected ratios: 45 app teams to 5 ops, 300 app teams to 4 ops, 350 apps to 7 ops, 300 apps to 8 ops.
Reported Insufficient2 sources— create a free account to open them View cited source
[19]
Organisations that built their own platform have a shadow platform group: each development group tends to have at least one person doing glue work between the platform and the application teams, a staffing cost completely unaccounted for in the business case spreadsheets.
Reported Insufficient2 sources— create a free account to open them View cited source
[20]
People get promoted for shipping platforms, and "we shipped a platform built on Kubernetes" is a better promotion-packet line than "we onboarded everyone to a thing we paid for"; the pattern is called resume-driven development and there is a study about it.
Reported Insufficient2 sources— create a free account to open them View cited source

Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.

Add to ChatGPT Add to Claude

Sources

1 independent publisher whose own reporting we read for this story.

thenewstack.io
1 article · August 21, 2026
Platform Engineering ROI: What it costs to build your own platform

Ask Clarity

Have a question this piece left open? The answer comes from its reporting and claim ledger — nothing else.

Clarity analysis

How the coverage leans and what the record actually supports. Read the methodology

Perspective Coverage

How much of the coverage speaks to builders, operators and investors. It measures emphasis, not whether the story is right.

Builder

For builders the useful content is the shape of the thing, not the price tag: seven CNCF-reference-architecture-aligned product teams (infrastructure, operations, deployment, runtime/middleware, database, security, enablement) at 7-9 engineers each, plus coordination roles. Read against that, a single team is 12-15 percent of the stated requirement, which is the concrete warning to anyone scoping a control-plane-plus-YAML v1: the roadmap after v1 is APIs, integrations, frameworks, security pipelines, dashboards, upgrade tooling and each new AI or compliance demand, indefinitely. The piece offers no example of a deliberately narrower platform that held, so it argues shape without arguing scope.

Reality

Evidence, adoption, hype gap, incentives and confidence, each read independently from the current record on the scale shown beneath it. How these are measured.

Evidence32: 0100
Single vendor-adjacent source; arithmetic checkable, inputs not
One article carries the entire cluster, and it discloses that its central numbers come from a VMware Tanzu Platform paper the author helped update. The internal arithmetic is verifiable (60 x $125,000 = $7.5M; $37.5M over five years; a 7-9 person team is 12-15 percent of 60), but the inputs are not: the operator ratios are anonymous and undated, the shadow-platform claim rests on unquantified conversations, and the resume-driven-development study is referenced without title or authors. No independent data corroborates the 60-person model.
AdoptionInsufficient: No dated deployment or usage disclosure in the cluster
The reporting contains no release, deployment, benchmark, pricing or named usage disclosure. The only quantities resembling adoption are anonymous developer-to-operator ratios collected by the author over an unspecified period, with no organisation, product or date attached, which cannot be treated as observed adoption.
Hype gap+44: −100 understated0 aligned+100 overstated
Precision on the build side, silence on the buy side
The claims run ahead of the evidence in a specific, measurable way: build cost is stated to the dollar ($7.5M/year, $37.5M/five years) while the recommended commercial alternative carries no price at all, leaving roughly $34.4M of five-year headroom unexamined. The salary-only model excludes benefits, tooling and infrastructure, and the operator ratios offered as the counterfactual vary by up to 8.3x within the same unit. The vendor provenance is disclosed, which restrains the score from going higher.
Incentives84: 0100
Disclosed vendor origin behind a buy recommendation
The author states the figures come from a recent VMware Tanzu Platform paper he helped update, and the same paper supplies the 12-18 month velocity-gap claim; the article's conclusion favours buying a commercial platform. The commercial interest is therefore direct and load-bearing on both sides of the comparison. Disclosure is explicit and pre-emptive, which is why this is scored as a transparent rather than concealed incentive.
Confidence58: 0100
High confidence in the gaps, low in the magnitudes
What is checkable is checkable with certainty: the provenance disclosure, the absence of any license price, the salary-only construction of $7.5M, and the dispersion inside the operator ratios all read directly off the text. What cannot be assessed from one vendor-adjacent source is whether 60 people is the real cost of a working platform for any given organisation, or whether the buy side clears the roughly $34.4M of headroom the article leaves open.

Topics and entities

Follow any of these and your For You feed starts watching them — no settings page required.

Topics

Platform Engineering Follow
Engineering IncentivesFollow
Engineering Cost ModelingFollow
Engineering Org StaffingFollow
Build vs Buy Follow

Entities

CNCF Platform Reference ArchitectureFollow
The New Stack Follow
David RicardoFollow
VMware Tanzu PlatformFollow
Résumé-Driven DevelopmentFollow
Kubernetes Follow