“Custom adtech renaissance” is the kind of phrase that gets ahead of its evidence. Read it quickly and it suggests brands are standing up their own demand-side platforms, replacing the licensed stack with something built in-house. That is not what is happening, and the gap between the phrase and the reality is worth being precise about — because the thing that is happening is more consequential than a wave of DIY DSPs would be.

The Build That Almost Nobody Is Doing

Start with the option that gets talked about most and executed least. Building a mid-scale DSP is a capital project estimated in the range of $10–25 million once infrastructure, integrations, and compliance are counted, and it buys you a product category that already has well-capitalized incumbents operating at scale.

It also does not end when the platform ships. Even a fully owned platform still needs live integrations with exchanges, identity providers, and verification vendors — which means the build-versus-buy question rarely resolves into a clean binary and instead becomes a permanent commitment to headcount, cloud spend, and legal review. Very few advertisers have a strategic reason to own bid-request processing.

So if buyers are not building DSPs, what are they building?

The Layer That Actually Moved: Log-Level Data

The ANA’s programmatic media transparency work is the most useful evidence here, because it measured the thing rather than surveying opinions about it. The study covered $123 million in ad spending across 35.5 billion impressions between September 2022 and January 2023, and it was built specifically on log-level data — the impression-by-impression record held by the tech vendors.

Its findings explain the in-housing pattern better than any vendor thought-leadership does:

  • Made-for-advertising sites took 21% of impressions and 15% of ad spend — an estimated $13 billion a year across the industry.
  • The average campaign ran across roughly 44,000 websites, when a few hundred would have reached the majority of the intended audience.
  • Only 31% of participating advertisers could get access to their own log-level data at all. Of 67 advertisers who wanted to take part, 21 could actually obtain impression-level records.
  • Advertisers who did connect and match their log-level data saw a modeled 20% improvement in ad spend productivity.

Read those together and the strategy becomes obvious. The waste was not hiding in the bidding algorithm. It was hiding in the absence of a record — and the record was held by whoever operated the infrastructure. A 20% productivity gain available to the third of advertisers who could see their own data is the entire business case for owning a data layer, and none of it requires building a DSP.

Subsequent ANA work found MFA spend down substantially as buyers acted on these findings, while also concluding that programmatic remains far from transparent — including the “cost waterfall” analysis finding that only a minority of programmatic spend reaches the consumer as working media. The problem got better where advertisers could see it.

What “Owned Infrastructure” Means in Practice

The buyers making real progress are not making a build-or-rent decision about the whole stack. They are picking one or two layers where ownership creates a durable advantage and continuing to license everything else. In practice, the layers being taken in-house cluster tightly:

The customer data platform and identity resolution. This is first-party data joined to outcomes. It is the asset that does not transfer when you switch DSPs, and the one that determines whether any measurement work downstream is possible.

The clean room and data-matching layer, particularly for retail and commerce media, where the whole negotiation between brand and retailer is about who sees what at which grain.

Measurement and incrementality, which is consistently the hardest of the three to staff. This is the honest constraint on the whole trend: measurement is harder to build than an ad server because it needs data science capability rather than engineering capacity, and a majority of marketers report limited internal analytics or data science resources as their principal measurement barrier. Ambition here regularly exceeds hiring.

What stays rented is almost everything else — bidders, exchange connectivity, verification, fraud detection, and the supply-path plumbing.

Why This Is Structural, Not Cyclical

In-housing has cycled before. The ANA found 35% of marketers had expanded in-house programmatic capability as far back as its 2017 survey work, up from 14% the prior year, and Adobe was forecasting 62% by 2022. Those projections consistently overshot, because the thing being in-housed was the buying — and buying is exactly the function where scale, tooling, and specialist labor favor an outside partner.

This cycle is different in what it targets. Owning the data layer is not an attempt to out-execute an agency at media buying. It is an attempt to hold the evidentiary record — so that supply-path decisions, MFA exclusion, and incrementality claims can be checked rather than accepted. That is a governance motive, not an efficiency motive, and governance motives do not reverse when the next efficiency argument arrives.

There is also a durability argument. Strategy and campaign direction were already substantially held by advertisers — ANA’s earlier work put strategic control in-house for roughly 69% of respondents, with only about a quarter outsourcing it. Extending control from strategy into the data substrate is a shorter step than extending it into execution ever was.

The Honest Caveats

Two things are worth holding onto before treating this as a settled trend.

First, the phrase itself travels ahead of the data. The “custom adtech renaissance” framing has circulated most strongly in market-specific coverage — notably around Indian brands building their own stacks — and generalizing it into a global pattern is a claim nobody has actually evidenced at scale. What is well-evidenced is narrower: log-level data access improves outcomes, and a minority of advertisers have it.

Second, ownership is not the same as capability. An owned clean room that no one is staffed to query is a line item, not an advantage. The 20% productivity figure in the ANA work attached to advertisers who connected and matched their data — the connecting and matching is the value, not the possession.

The practical question for a buyer in 2026 is not whether to build. It is narrower and more answerable: can you obtain impression-level records for your own campaigns today, and if not, which contract renewal is the moment you can require them? That is where the returns in this shift are actually located, and it is a procurement conversation before it is an engineering one.