Microsoft's pitch deck for Power Platform almost always opens with a chart that makes custom development look like a museum piece: 74% faster delivery, 3.2x ROI, citizen developers everywhere. The numbers are not wrong. They are just Year 1. And almost no enterprise application is retired at the end of Year 1.

The honest question is not whether Power Platform is cheaper than custom development. It is whether the Power Platform total cost of ownership over three years beats the alternative for your specific workload, integration depth, and user base. That question has a defensible answer. It rarely matches the sticker price.

This article lays out a vendor-neutral, seven-category TCO framework we use at BabyBots to pressure-test buy-vs-build decisions with clients. Every figure is current for 2026 and cited to primary sources. Every assumption is labeled illustrative. The goal is not to pick a winner. It is to give you a model you can defend in front of finance.

TL;DR

Power Platform and custom development have opposite cost shapes: Power Platform front-loads almost nothing and then charges monthly per user, while custom development front-loads the build and costs relatively little to maintain. A Year 1 comparison is meaningless. Only a three-year, seven-category TCO model reveals where the real crossover sits, and that crossover has moved sharply in the last 18 months because of premium connector enforcement, AI-accelerated custom delivery, and the AI Builder credit sunset.

Key Takeaways

• Sticker-price comparisons systematically underestimate Power Platform TCO by 30-50% because they omit governance, premium connector reclassification, and citizen developer sprawl.
• Microsoft ended premium connector grandfathering on October 1, 2024 and began in-product license enforcement on April 1, 2025, with stricter enforcement scheduled for February 2027.
• AI-accelerated custom development has compressed initial build hours by 20-45% of addressable time, which flattens the historic buy-vs-build curve dramatically.
• A defensible governance overhead line item runs 15-25% of licensing spend at scale, covering Center of Excellence staffing, ALM, environment strategy, and DLP administration.
• Break-even between Power Platform and custom development typically lands between 150 and 400 users for standard workloads, and shifts sharply based on integration depth and app lifespan.

Why Sticker-Price Comparisons Fail

Every Power Platform TCO article that ranks on Google was written by a Microsoft partner. That is not a conspiracy. It is a structural bias. Partners have an incentive to steer the reader toward more Microsoft spend, more licensing consulting, or more managed services. The math always resolves toward Redmond.

The result is a body of content that models two things well: licensing and initial build. And that omits five things badly: premium connector economics, governance overhead, talent, exit reserve, and the compounding cost of citizen developer sprawl. Any TCO model that stops at licensing and build will miss 30-50% of the real three-year cost. We have watched this play out inside enterprise clients often enough to treat it as a pattern, not an anecdote.

The other systemic error is treating custom development as a 2015 waterfall project with a $150,000 upfront invoice and a six-month timeline. That model is obsolete. Generative AI has changed the delivery economics of custom code more in the last two years than the previous decade combined, and any TCO comparison that ignores it will overweight the low-code side.

The Seven-Category TCO Framework

Every workload has seven cost categories over a three-year horizon. Model all seven or you are not doing TCO. You are doing licensing math with a spreadsheet border.

The BabyBots Seven-Category TCO Model

• Category 1: Platform licensing. Per-user, per-capacity, or per-bot fees paid to the vendor for the right to run the workload.
• Category 2: Premium connectors and capacity add-ons. The uplift triggered when an app touches SQL Server, Azure SQL, Dynamics 365, Salesforce, or exceeds Dataverse, request, or credit limits.
• Category 3: Build and implementation labor. Design, development, testing, and initial deployment. Higher for custom, lower for low-code, but never zero for either.
• Category 4: Maintenance and enhancement. Bug fixes, minor features, platform version upgrades, dependency patching. Runs 15-20% of build annually for custom; 10-15% for well-governed Power Platform estates.
• Category 5: Governance and Center of Excellence overhead. DLP policies, environment strategy, ALM tooling, tenant monitoring, license optimization, and CoE staffing. Frequently zero in vendor models. Rarely zero in reality.
• Category 6: Talent acquisition and retention. Fully-loaded cost of the humans required to keep the workload alive, including scarcity premiums for PL-600 architects and senior .NET engineers.
• Category 7: Exit reserve. The cost you would incur to migrate off the platform. Usually modeled as zero. Almost never zero in practice.

Category 1: Current 2026 Power Platform Licensing

Start with the numbers on Microsoft's live pricing pages. These are the illustrative rates for U.S. commercial as of Q3 2026. Verify against Microsoft's Power Platform pricing page before signing anything.

2026 Power Platform Rate Card (Illustrative)

• Power Apps Premium: $20 per user per month; drops to approximately $12 per user per month at 2,000+ seats.
• Power Automate Premium: $15 per user per month for attended flows.
• Power Automate Process: $150 per bot per month for unattended RPA and cloud process automation.
• Power Automate Hosted Process: $215 per bot per month, which includes the hosted Windows machine.
• Power BI Pro: $14 per user per month.
• Power BI Premium Per User (PPU): $24 per user per month.
• Power Pages authenticated users: $200 per pack per month for 100 users.
• Power Pages anonymous users: $75 per pack per month for 500 users.
• Copilot Studio message packs: $200 per month for 25,000 Copilot Credits.
• Dataverse database capacity: $40 per GB per month above the seeded allotment.

The trap is not the headline rate. It is the multiplication. A modest workload with 200 users on Power Apps Premium alone runs $48,000 per year at list price. Add Power Automate Premium for 40 approvers and two unattended process bots and the licensing bill crosses $60,000 annually before a single premium connector is touched. Multiply by three years and licensing alone approaches $180,000 for a workload most sponsors described in the pitch as "basically free because we already have Microsoft 365."

Category 2: The Premium Connector Reclassification Event

The single most expensive line item most enterprises miss is premium connector economics. Standard connectors (SharePoint, Outlook, Teams, OneDrive) are included with Microsoft 365. Premium connectors are not.

The moment an app touches a premium connector: SQL Server, Azure SQL, Dynamics 365, Business Central, Salesforce, ServiceNow, SAP, or any custom connector, the app is stamped Premium and every end user needs a paid Power Apps or Power Automate license. This is not new. What is new is enforcement.

Microsoft's grandfathering of legacy SQL Server access ended on October 1, 2024. In-product license enforcement began on April 1, 2025, and a further stricter enforcement wave is scheduled for February 2027. The practical effect is that thousands of apps that ran for years on standard licenses are now generating license-required errors and triggering emergency procurement conversations.

We have seen this play out at three clients in the last twelve months. In every case, the sponsor's original TCO model assumed standard connectors would carry the workload. In every case, real integration requirements forced at least one premium connector into production within nine months. In every case, licensing costs jumped 40-120% overnight when enforcement kicked in. Any TCO model built on "we will stay on standard connectors" is not a model. It is a hope.

Any TCO model that stops at licensing and build will miss 30-50% of the real three-year cost, because governance, premium connector reclassification, and citizen developer sprawl are not optional line items.

Category 3: Build Labor in the Age of AI-Accelerated Delivery

The custom development side of the ledger has changed more in two years than in the previous decade. Ignoring that shift is the single biggest mistake in every partner-authored TCO article on Google today.

The evidence is now overwhelming. McKinsey's controlled study found that developers using generative AI completed common tasks up to twice as fast, with particularly strong gains on code documentation, new code generation, and refactoring. GitHub's randomized controlled trial of Copilot found a 55% reduction in task completion time on a scoped programming task. Microsoft's own internal analysis showed developers using Copilot merged 12.92% to 21.83% more pull requests per week. The addressable share of developer time these tools can accelerate lands between 20% and 45%.

Two caveats matter for honest modeling. First, junior developers see smaller gains than seniors, because senior engineers are better at prompting, reviewing, and rejecting bad AI output. Second, the productivity gains show up in coding time, not requirements gathering, stakeholder alignment, security review, or user acceptance testing, which still consume most of a project's calendar.

The net effect for TCO modeling: a mid-complexity internal business application that would have cost $150,000 and taken six months to build in 2022 now defensibly costs $90,000-$120,000 and takes three to four months in 2026 using AI-accelerated senior engineers. That flattens the historic buy-vs-build curve in ways every three-year model needs to reflect.

Categories 4-6: The Costs Vendor Models Omit

Maintenance and Enhancement

The industry rule of thumb for custom software maintenance is 15-20% of the build cost annually, covering bug fixes, minor features, dependency updates, and security patching. Well-governed Power Platform estates run 10-15%, because the platform handles infrastructure and runtime patching. But this is only true if the estate is well-governed. Ungoverned Power Platform estates run higher, because every citizen-built app is a maintenance liability nobody owns.

Governance and Center of Excellence Overhead

This is the category vendor models pretend does not exist. In practice, running Power Platform at enterprise scale requires: a Center of Excellence team, the CoE Starter Kit deployment and ongoing operation, environment strategy across dev/test/prod for every solution, DLP policy administration, tenant-level monitoring, ALM tooling (Azure DevOps or GitHub Actions pipelines), license optimization cycles, and a maker community program.

A defensible rule of thumb from what we observe in the field: governance overhead runs 15-25% of total licensing spend at scale. Below 100 users, governance is typically absorbed into existing IT overhead and looks free. Above 500 users, it becomes a dedicated function with real headcount. Any TCO model that omits this line will understate three-year costs by six figures.

Talent Acquisition and Retention

Custom development requires senior .NET or full-stack engineers, currently commanding fully-loaded costs of $180,000-$260,000 in most U.S. markets. Power Platform requires PL-600 Solution Architects, Dataverse specialists, ALM engineers, and a CoE lead, with fully-loaded costs in the $140,000-$220,000 range. The gap is narrower than the low-code narrative suggests, because scarcity for certified PL-600 architects has driven salaries up sharply since 2024.

Illustrative 3-Year TCO Scenarios

The following scenarios are illustrative. They use the 2026 rate card above, assume mid-complexity workloads with moderate integration depth, and are labeled clearly. Do not paste them into a board deck without adjusting to your context.

Scenario A: 50 Users, Mid-Complexity Internal App

• Power Platform 3-year TCO: Licensing $36,000 + premium connectors and Dataverse $18,000 + build $25,000 + maintenance $10,000 + governance $8,000 + talent allocation $30,000 + exit reserve $8,000 = approximately $135,000.
• Custom development 3-year TCO: Build $95,000 (AI-accelerated) + maintenance $42,000 + hosting $12,000 + governance $4,000 + talent allocation $18,000 + exit reserve $5,000 = approximately $176,000.
• Verdict: Power Platform wins by roughly $41,000 at this scale.

Scenario B: 200 Users, Mid-Complexity Internal App

• Power Platform 3-year TCO: Licensing $144,000 + premium connectors and Dataverse $54,000 + build $45,000 + maintenance $22,000 + governance $30,000 + talent allocation $60,000 + exit reserve $18,000 = approximately $373,000.
• Custom development 3-year TCO: Build $130,000 (AI-accelerated) + maintenance $58,000 + hosting $28,000 + governance $10,000 + talent allocation $45,000 + exit reserve $12,000 = approximately $283,000.
• Verdict: Custom development wins by roughly $90,000 at this scale.

Scenario C: 500 Users, Mid-Complexity Internal App

• Power Platform 3-year TCO: Licensing $360,000 + premium connectors and Dataverse $110,000 + build $65,000 + maintenance $32,000 + governance $85,000 + talent allocation $130,000 + exit reserve $45,000 = approximately $827,000.
• Custom development 3-year TCO: Build $180,000 (AI-accelerated) + maintenance $85,000 + hosting $55,000 + governance $18,000 + talent allocation $95,000 + exit reserve $25,000 = approximately $458,000.
• Verdict: Custom development wins by roughly $369,000 at this scale.

The crossover typically sits between 150 and 400 users for standard mid-complexity workloads. Below the crossover, Power Platform's low upfront cost dominates. Above it, custom development's flat maintenance curve dominates. This is not a moral judgment about either platform. It is arithmetic.

The Citizen Developer Tax

Forrester's 2026 low-code data shows enterprise adoption at 77%, with an installed base of roughly 16.2 million citizen developers globally. The same research shows 61% of IT leaders cite shadow IT as a top governance risk from ungoverned no-code platforms.

The citizen developer tax has four components: orphaned apps whose maker has left the organization, duplicate workflows that solve the same problem three different ways, shadow Dataverse tables that bypass data governance, and unused Premium seats assigned to makers who built one app and moved on. Every one of these costs money. None of them appear in a partner's TCO model.

At one client with 850 Power Platform makers, an audit revealed 43% of Premium seats were assigned to users who had not opened Power Apps in 90 days. That was $82,000 per year of pure license waste. This is not exotic. It is the median outcome of ungoverned adoption.

The AI Builder to Copilot Credits Transition

Any 3-year TCO forecast built without accounting for the AI Builder sunset is already obsolete. Microsoft has confirmed that AI Builder capacity add-ons reached end of sale on November 1, 2025 for new customers, and that seeded AI Builder credits inside premium licenses will be removed on November 1, 2026. AI Builder features are migrating to Copilot Credits, at $200 per month for 25,000 credits.

The practical effect is that document processing, prediction, and form automation workloads that ran on "free" seeded AI Builder credits will start metering against Copilot Credits inside the forecast window. For high-volume document processing workloads, this transition alone can add $30,000-$150,000 to a three-year forecast that assumed AI Builder economics would persist.

Break-Even Sensitivity: What Actually Moves the Crossover

The three-scenario model above assumes moderate integration depth and a five-year app lifespan. Real decisions are more sensitive than that. Four variables move the crossover point aggressively.

• User count: Below 150 users, Power Platform's low upfront cost almost always wins. Above 400 users, custom development's flat maintenance curve almost always wins. The 150-400 range is where honest analysis matters.
• Integration depth: Every additional premium connector shifts the crossover downward by roughly 30-50 users, because premium licensing kicks in for the entire user base.
• App lifespan: A two-year workload almost always favors Power Platform. A ten-year workload almost always favors custom, because Power Platform's recurring cost compounds while custom's build cost amortizes.
• Change velocity: Workloads that change monthly favor Power Platform's rapid iteration. Workloads that stabilize favor custom's lower run-rate.

Decision Framework: When Each Wins

Choose Power Platform when

• User count is under 150-200 for the specific workload.
• Integration is limited to standard connectors or one premium connector.
• Expected lifespan is under five years, or the requirements are genuinely unstable.
• Business ownership and rapid iteration outrank pixel-perfect UX.
• You already have mature governance, a functioning CoE, and disciplined environment strategy.

Choose custom development when

• User count exceeds 300-400 for the workload.
• The application is customer-facing, high-scale, or latency-sensitive.
• Domain logic is complex enough that low-code abstractions fight the requirements.
• Expected lifespan is seven years or more.
• IP ownership, portability, and vendor concentration risk matter to the board.

Choose both when

• You are building a portfolio of workloads and can allocate each to the right platform.
• Power Platform handles internal workflows and approvals while custom handles customer-facing and mission-critical.
• You have the organizational maturity to run Fusion teams that blend citizen and professional developers.

Frequently Asked Questions

Is Power Platform really cheaper than custom development?

It is cheaper in Year 1 for almost every workload. It is cheaper across three years for workloads under roughly 150-200 users with limited integration depth. Above that threshold, custom development typically wins on three-year TCO, especially with AI-accelerated delivery compressing initial build cost.

How much should we budget for Power Platform governance?

As a defensible rule of thumb, budget 15-25% of licensing spend for governance and Center of Excellence overhead at scale. This covers CoE staffing, ALM tooling, environment strategy, DLP administration, and license optimization. Below 100 users you can often absorb this into existing IT. Above 500 users it becomes a dedicated function.

What changed with premium connector licensing in 2024 and 2025?

Microsoft ended grandfathering for legacy standard-connector access on October 1, 2024. In-product license enforcement began on April 1, 2025, and a stricter enforcement wave is scheduled for February 2027. Apps that touch SQL Server, Azure SQL, Dynamics 365, Salesforce, ServiceNow, SAP, or custom connectors now require every end user to hold a paid Power Apps or Power Automate license.

How does AI change custom development cost in 2026?

Controlled studies from GitHub, McKinsey, and Microsoft show developers completing coding tasks 20-55% faster with generative AI assistance, addressable to 20-45% of total development time. In practice, this means a mid-complexity internal application that cost $150,000 to build in 2022 defensibly costs $90,000-$120,000 in 2026. Senior engineers capture more of the gain than juniors.

What is the citizen developer tax and how do we quantify it?

The citizen developer tax is the compounding cost of ungoverned low-code adoption: orphaned apps, duplicate workflows, shadow data stores, and unused Premium seats. In our engagements, 30-45% of Premium seats are typically assigned to inactive makers within 18 months of unmanaged rollout. Quantify it by auditing seat activity quarterly and reclaiming licenses.

Should exit costs really be a line item in TCO?

Yes. Any three-year commitment to a proprietary platform carries migration risk. A defensible exit reserve runs 5-10% of cumulative platform spend, funded to make a future migration decision possible without a crisis-driven rewrite. Most organizations skip this line and regret it when strategic direction changes.

Sources

• Licensing overview for Microsoft Power Platform, Microsoft Learn: https://learn.microsoft.com/en-us/power-platform/admin/pricing-billing-skus
• Power Platform Pricing, Microsoft: https://www.microsoft.com/en-us/power-platform/pricing
• SQL Server grandfathered licensing model ends October 1, 2024, Power Apps Guide: https://powerappsguide.com/blog/post/sql-server---grandfathered-licensing-model-to-end-on-october-1-2024
• End of AI Builder credits, Microsoft Learn: https://learn.microsoft.com/en-us/ai-builder/endofaibcredits
• Unleashing developer productivity with generative AI, McKinsey and Company: https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/unleashing-developer-productivity-with-generative-ai
• 2026 software development price guide, FullStack Labs: https://www.fullstack.com/labs/resources/blog/software-development-price-guide-hourly-rate-comparison
• Low-Code Statistics 2026, ToolJet compiling Gartner, Forrester, McKinsey, and G2 data: https://blog.tooljet.com/low-code-statistics-market-ai-trends-2026/

The Strategic Bottom Line

Power Platform is not cheaper than custom development. It is differently priced, on a different curve, with different risks. The crossover between the two depends on seven variables most TCO models omit and on three market shifts that have accelerated in the last 18 months: premium connector enforcement, AI-accelerated custom delivery, and the AI Builder credit sunset.

The organizations that will win the next three years are not the ones that pick a side. They are the ones that build a portfolio-level architecture, allocate each workload to the platform where its three-year math wins, invest in the governance discipline that keeps low-code from becoming shadow IT, and refuse to sign anything based on a Year 1 comparison. The math is defensible. The discipline is the hard part.