The Infrastructure Bet

The Infrastructure Bet

The deceleration in The Growth Engine has a proposed fix. Management is building four infrastructure layers — the Kokai AI buying engine, the OpenPath and Sincera supply path, Unified ID 2.0, and the Ventura streaming operating system — meant to reaccelerate growth by owning more of the open internet's plumbing. On the company's own markers the layers are gaining traction. But their payoff is largely unmeasurable from outside the company, and two of them pull against the neutrality that the independence case rests on.

This is the mechanism behind management's claim that the slowdown is fixable rather than structural. It deserves to be examined on its own, because if the bets work they widen a moat the prior chapter called narrow, and if they stall the structural-ceiling reading gets stronger.

The stack management is building

Each layer targets a different part of the transaction between an advertiser's budget and a publisher's impression: the buying decision (Kokai), the path to inventory (OpenPath/Sincera), the identifier that makes an impression targetable (UID2), and, in connected TV, the device layer itself (Ventura).

Sources: Q2 FY2025 earnings call [1]; FY2025 10-K, Item 1A Risk Factors [2]; Q4 FY2025 earnings release [3]; Q4 FY2024 earnings release [4].

Kokai, and the performance case

Kokai is the platform overhaul at the center of the reacceleration argument. By the second quarter of 2025 roughly three-quarters of client spend was running through it, and management expected all clients on it by year-end [5]. The commercial case is that better buying tools deepen spend: management says clients who moved the majority of their budgets onto Kokai grew their overall Trade Desk spend about 20% faster than those who had not [6].

Source: Q2 FY2025 earnings call [7].

The named results are real cases the company chose to cite, not audited averages. Samsung reported a 43% improvement in reaching its target audience on an omnichannel European campaign; Cashrewards reported a 73% improvement in cost per acquisition in Asia; in aggregate management points to a better-than-20-point lift across key campaign metrics under Kokai [8].

Source: Q2 FY2025 earnings call; the two cases measure different KPIs (audience reach vs cost per acquisition) [9].

These are the strongest data points in the chapter, and also the ones to hold at arm's length. They are hand-picked, self-reported, and expressed as relative improvements without a base. They support the direction of the bet — Kokai is being adopted and clients who lean in spend more — without proving its magnitude.

OpenPath and Sincera: reaching into the supply

OpenPath lets publishers connect directly to The Trade Desk, bypassing a supply-side platform if they choose [10]. Management describes a "material" amount of spend now flowing through it and cites Hearst Newspapers seeing a fourfold improvement in fill-rate after adopting it [11]. The 2025 acquisition of Sincera, a supply-chain metadata company, feeds transparency data into the platform, with a free "Open Sincera" version offered to the wider industry [12].

The logic is efficiency: a cleaner, cheaper path from budget to impression puts more of each dollar into working media, which should lift spend and retention. Jeff Green frames OpenPath as "a canary in the coal mine and a stalking horse" — a way to see where the supply chain is inefficient and to keep exchanges honest [13].

Two facts temper this. First, OpenPath is explicitly not meant to be the primary supply path; the company still sources inventory from over 430 directly integrated exchanges, publishers and supply-side platforms, and management notes SSPs have a better competitive chance after the Google antitrust verdict, not a worse one [14] [15].

Source: FY2025 10-K, Item 1 Business (reported as "over 430") [16].

Second, the company itself flags that the effort may not work: it launched OpenPath to improve supply-chain efficiency but "cannot guarantee that this or future offerings will prove attractive to our clients or otherwise be successful" [17]. That admission matters because the backdrop in 2025 was a large increase in ad supply that outpaced demand, lowering both CPMs and fill-rates for non-premium publishers and reducing the transparency of the metadata OpenPath and Sincera are meant to clean up [18].

Identity without a forcing function

Unified ID 2.0 is the layer the open internet needs most and controls least. It converts logged-in signals such as email addresses and phone numbers into pseudonymous identifiers, giving advertisers a way to target authenticated audiences without third-party cookies; the related OpenPass single sign-on and the European EUID extend the same architecture [19]. Adoption on the sell side is tangible: FreeWheel, Index Exchange, Magnite and PubMatic have integrated EUID, and the company continues to add UID2 partners across streaming [20] [21].

The complication was set out in The Independence Moat: Google announced the deprecation of third-party cookies, then abandoned it in April 2025. That removed a disruption, but it also removed the forcing function. UID2 adoption is voluntary, and a durable cookie makes the case for switching to it less urgent for publishers who were dragging their feet. The identity layer is therefore the bet with the clearest strategic logic and the least company control over its own timeline.

Ventura, and the neutrality question

Ventura is the most ambitious layer and the one that most tests the independence thesis. It is a streaming-TV operating system, positioned to fix "inefficient advertising supply chains and content conflicts-of-interest" in CTV, deployed through TV manufacturers and distribution partners [22]. In early 2026 the company broadened it into a "Ventura Ecosystem" that lets other CTV participants use its monetization tools [23].

Here the through-line to the moat becomes two-sided. The independence case rests on The Trade Desk owning no media and taking no side but the buyer's. OpenPath already moves it into functions historically performed by SSPs — management openly says publishers can "act as their own SSP" through it [24], and elsewhere that OpenPath keeps exchanges and SSPs "in check" so it need not wait on regulators [25]. Ventura goes further, taking an operating-system position at the device layer. The more of the chain the company occupies, the more it starts to resemble the vertically integrated players whose conflicts it criticizes.

The counter to that concern is concrete. The Trade Desk still owns no inventory and sells no media; OpenPath is optional and free to publishers; and Ventura is pitched as removing the middleman's toll rather than collecting a new one. Occupying more of the plumbing is not the same as owning the audience — the specific conflict that defines a walled garden. The distinction is real, but it narrows as the footprint widens, and it is worth watching rather than assuming.

What the disclosure lets you measure

The honest constraint on this chapter is that none of these layers has a revenue line. The Trade Desk earns the same platform fee on gross spend whether a dollar is bought through Kokai or the legacy stack, sourced through OpenPath or an SSP, targeted via UID2 or a cookie. The bets are funded through operating expense and the stock-based compensation examined in Stock Comp and Cash, and their return shows up only indirectly — in retention, in spend growth per client, and in whether the ~21% take rate holds. The markers management gives are genuine but selective: an adoption percentage here, a client case study there, a "material" share of spend without a number [26] [27].

The read this chapter lands on: the infrastructure bet is the most credible case that the deceleration is fixable rather than structural, because it is a coherent, well-funded attempt to control the open internet's identity, supply and device layers rather than merely ride them. The strongest fact against it is that the one layer that matters most for competing with walled gardens — identity — lost its forcing function when Google kept cookies, and its adoption runs on other companies' timelines, not The Trade Desk's [28]. What would move the read is evidence the market cannot yet see: a disclosed OpenPath share of spend, an independent measure of UID2 authenticated reach, or take-rate expansion the company attributes to these layers. Until then, the bet is credible and unproven in equal measure.