OpenX published a company blog post on September 2, 2026, arguing that the "cloud tax" - a term rival executives have used to describe the hidden cost of running programmatic infrastructure on public cloud servers - misdiagnoses where inefficiency actually sits in the advertising supply chain. Written by chief technology officer Joel Meyer, the post claims the supply-side platform supports 3 million queries per second and can add regional capacity in 24 hours, positioning public cloud infrastructure as an advantage rather than a liability for buyers.

In short

OpenX published a blog post on September 2, 2026, rejecting the idea that running its advertising exchange on public cloud servers adds a hidden cost passed on to brands and agencies. The argument matters because rival executives, including Index Exchange chief executive Andrew Casale, have spent 2026 making the opposite case, that moving decisioning off public cloud and into proprietary data centers cuts latency and expense. For media buyers, the dispute is really about where their custom bidding logic runs and how fast it responds, not about which cloud provider issues the bill.

A rebuttal aimed at a specific industry argument

Every millisecond in a programmatic auction carries consequences that most buyers never see directly. Processing programmatic advertising at global scale requires enormous computing power, making the infrastructure behind every auction more consequential than most buyers realize, and while they may never see it directly, they feel the effects in how supply surfaces, how well their signals are used, and ultimately, the results their campaigns deliver. That framing opens OpenX's blog post, and it sets up the company's target: a phrase that has circulated through supply-side platform marketing and executive commentary for much of 2026.

The cloud tax is the argument that running an omnichannel supply-side platform on a public cloud adds a hidden infrastructure cost that eventually reaches brands and agencies. OpenX's post does not treat that argument as settled. The company states plainly that the cloud itself is not the issue, and that what actually matters to buyers is the architecture underpinning the exchange and how that design affects latency, the time it takes to process and respond to a bid request.

The phrase "cloud tax" did not originate with OpenX. Index Exchange chief executive Andrew Casale used the term at IAB's 2026 Annual Leadership Meeting on February 12, 2026, when he outlined a shift toward impression-level AI decisioning, describing operational costs from external infrastructure dependencies that consume resources and constrain innovation. Casale's argument called for decisioning logic to move upstream, off public cloud servers and into an exchange's own infrastructure, to eliminate that cost entirely. The framing has since been repeated across trade coverage, including in PPC Land's explainer on ARTF, the IAB Tech Lab standard for running partner code inside an auction as a container, where Casale's cloud tax argument is credited as one catalyst for the wider containerization movement across the sell side.

OpenX's response is notable because it does not dispute the goal Casale described. It disputes the diagnosis. Where Casale's argument locates the tax in public cloud infrastructure itself, OpenX locates the equivalent cost somewhere else entirely: in the alternative of running a private, self-managed data center.

What OpenX says its architecture actually does

The technical core of OpenX's post rests on a single operational claim. Aside from the requests OpenX sends to bidders, the company states it has never called a service running outside its own cloud-native infrastructure, and that every step in the ad request pipeline, whether the logic belongs to OpenX or its partners, runs inside its environment under strict latency service level objectives that it monitors continuously.

That distinction matters technically because it describes where a decisioning model physically executes relative to the auction. According to OpenX, that architecture means custom logic can run within the auction itself, reducing latency and enabling faster, more consistent decisioning at scale, and every engineering hour devoted to provisioning capacity, maintaining systems, or troubleshooting infrastructure, which the company says would be required if it maintained its own private cloud, is an hour not spent developing new capabilities for buyers.

This is essentially the same architectural principle that has driven the wider containerization movement across the sell side: logic running closer to the data it needs to evaluate produces faster, more informed decisions than logic that has to travel outside the auction and back. OpenX's post frames its own seven-year-old cloud-native infrastructure as having reached that same destination through a different route, without the container standardization work IAB Tech Lab formalized in November 2025.

OpenX describes public cloud infrastructure as ideally suited to the demands of real-time bidding, where traffic is variable and every millisecond matters, because capacity can scale as demand changes, compute can stay close to the data it acts on, and redundant-by-design systems recover quickly when something goes wrong. By contrast, the company argues privately managed systems can perform well, but that growth is slower and more hands-on, because new capacity, regions, and redundancy still depend on managing physical infrastructure directly.

For buyers, OpenX frames the real distinction as not simply where a platform runs, but whether its infrastructure can keep pace with the auction or becomes another constraint, arguing that a self-managed model requires more resources to maintain infrastructure while working with a cloud provider leaves more available for the speed, stability, and scale buyers expect from their adtech partners.

The logic-versus-data principle

A second technical argument in the post concerns where computation should physically sit relative to signal data. OpenX states that running logic in containers is not a new capability for the company, that it is how OpenX has run its own platform for years, and that this experience is why it is positioned to operate containers on behalf of its partners.

The efficiency argument rests on a simple principle, according to the post: it is better to execute logic where all of the data already is than to move a subset of that data out to wherever the logic happens to live, because a model sitting outside the auction can only ever evaluate what fits in the request sent to it, while a model running inside the auction can see the full set of signals and influence a decision without data in the bidstream having to travel, with the primary benefits being better performance at lower cost and lower latency.

This is close to a restatement of the argument Amazon, Bedrock Platform, and Index Exchange have made in separate contexts throughout 2026 about the limits of queries-per-second throttling. As PPC Land's explainer on demand-side platforms documented, no bidder can evaluate every request sent to it, so demand-side platforms apply pretargeting rules that discard traffic before scoring, and exchanges cap queries per second. Bedrock Platform and Index Exchange addressed that constraint directly on April 21, 2026, when they ran a containerized demand-side platform bidder inside exchange infrastructure for the first time, removing the queries-per-second ceiling that had forced buyers to evaluate only a sampled slice of the bid stream. OpenX's post makes a parallel case for its own infrastructure without referencing that specific deployment.

Numbers OpenX attaches to scale and speed

The post includes several specific figures that quantify the company's infrastructure claims, though these are OpenX's own reported metrics rather than independently audited figures.

OpenX states it supports 3 million queries per second and beyond, giving partners the capacity to apply decisioning logic at scale without limits based on status or priority, contrasting this with unnamed platforms that the company says cap capacity at a fraction of that figure, often with a lower ceiling for standard partners and a slightly higher one for partners who qualify for preferred status.

By comparison, OpenX says it can expand capacity in minutes or deploy to a new region in 24 hours instead of months, without making that level of support contingent on partner size or status, a design the company frames as giving every partner the same treatment instead of reserving speed and flexibility for a select few. The post's separately published frequently-asked-questions section adds a further figure: OpenXBuild, the API suite the company launched on January 6, 2026, lets partners run decisioning logic inside secure containers in approximately 5 milliseconds.

That 5-millisecond figure sits close to the execution window Index Exchange has reported for its own containerized deployments. According to PPC Land's report on the Bedrock and Index Exchange deployment, Index Exchange reported a container execution window of under five milliseconds for Bedrock's bidder running inside Index Cloud. Neither figure has been independently verified by a third party, and OpenX's post does not claim independent verification for its own timing.

Sustainability, security, and operational history

The post ties its infrastructure argument to a broader set of claims spanning environmental impact, security posture, and seven years of operating history. OpenX states that when it moved from physical servers to a fully cloud-based infrastructure seven years ago, that shift freed engineers to spend less time maintaining hardware and more time building the company's products and other custom solutions, while adding that the benefits extend beyond engineering efficiency to clear operational advantages in scalability, sustainability, and security.

On environmental impact, the company says moving off its own hardware made OpenX the first company across advertising, media, or technology to achieve CarbonNeutral certification and meet the Net-Zero Standard requirements, and that different hardware profiles let it match workloads to the most efficient equipment instead of retiring hardware on a fixed schedule, keeping older machines in service where they still perform well, using ARM processors where they fit, and placing workloads near low-carbon power sources like hydropower when possible.

That claim extends a position OpenX has repeated consistently across 2025 and 2026. Most recently, on August 20, 2026, OpenX published a separate statement backing the IAB UK Gold Standard, in which the company again cited its 2022 CarbonNeutral certification and framed unnecessary auctions as carrying "a real computational and environmental cost." The recurrence of that argument across multiple company statements over the past several weeks suggests OpenX is building a consistent public position tying infrastructure efficiency to environmental credentials, ahead of the Gold Standard's newly added sustainability pillar, which addresses inefficient ad refresh rates and bid floors that limit unnecessary computational load.

On security, the post argues that a public cloud lives or dies by its ability to guarantee and manage security, providing what OpenX calls excellent and holistic security controls that allow the company to lock down roles and permissions consistently and systematically, a level of rigor it says is near impossible to achieve when managing a variety of hardware and software from different vendors with different permission models, particularly when that management is a fraction of someone's full-time job.

On operational history, the company states that seven years of running containerized workloads at real auction volume has given OpenX deep expertise in operating cloud-native infrastructure at scale, showing up, it says, in reliability and in faster product development.

OpenXBuild as the product expression of the argument

The post connects its broader infrastructure argument to a specific commercial product. OpenX describes OpenXBuild as a software suite that lets buyers and partners run their own logic inside secure containers on OpenX's cloud-native infrastructure, for faster audience matching, more actionable auction insights, and a cleaner signal path for effective bidstream shaping and filtering, framing that foundation as giving the company the scale and flexibility to keep building what comes next.

OpenXBuild is not a new announcement. OpenX first unveiled the suite on January 6, 2026, the same day major publishers including AccuWeather, BuzzFeed, the Guardian, and Hearst committed to The Trade Desk's rival OpenAds platform, according to PPC Land's report on Matt Sattel's appointment as OpenX chief executive that February. At launch, PPC Land reported that OpenXBuild's three application programming interfaces let advertisers deploy custom logic directly into supply-side infrastructure, with early adopter tvScientific reporting a 70 percent reduction in cost-per-conversion. That figure is a vendor-reported result from a single early customer, not an independently audited benchmark, and OpenX's own company profile compiled by PPC Land notes that performance figures the company publishes are largely unaudited.

The pattern of containerized decisioning that OpenXBuild represents is not unique to OpenX. Zillow became the first brand advertiser to pilot containerized real-time bidding with Chalice Custom Algorithms and Index Exchange, a test reported on August 6, 2025, months before OpenXBuild's launch. PubMatic later built its own version, called Decision Fabric, which PPC Land covered in June 2026 as a proprietary implementation of the same architectural shift at exchange scale. Magnite, too, set out deployment routes for outside models running inside containers, according to trade coverage of the standard. The industry-wide convergence on this pattern, moving decisioning logic into the exchange rather than leaving it in an external demand-side platform, is one point where OpenX's post and its rivals' public statements largely agree, even as they disagree about whose infrastructure caused the problem the pattern is solving.

Context: a debate that predates this post by months

The broader argument OpenX is entering did not begin with this post, and it has moved through several distinct phases across 2026. Casale's February 12 remarks framed the cloud tax as a reason to move intelligence upstream, closer to the impression. Two months later, on April 21, 2026, Bedrock Platform and Index Exchange acted on that framing directly, running what PPC Land reported was the first containerized demand-side platform bidder operating inside an exchange's own data center infrastructure rather than on external public cloud servers.

IAB Tech Lab formalized the underlying technical standard on November 13, 2025, when it released the Agentic Real-Time Framework version 1.0 for public comment, with named contributors including Index Exchange, Chalice, The Trade Desk, and OpenX itself. That detail is significant for reading OpenX's post: the company was a named contributor to the same standards effort that grew out of the cloud tax argument it is now pushing back against, even as it frames its own approach as having already reached the same destination through different, longer-standing infrastructure.

The framework's central claim, a 90 percent reduction in latency, is one IAB Tech Lab reports rather than an independently measured figure, according to PPC Land's coverage. That caveat applies equally to the specific numbers in OpenX's own post, none of which carry third-party audit or verification cited in the source material.

Why this matters for the marketing community

For media buyers, agencies, and publishers, the practical stakes of this dispute are narrower than the rhetoric surrounding it. Both OpenX and Index Exchange are converging on the same technical destination: decisioning logic that runs inside the auction rather than calling out to external servers, reducing round-trip latency and giving models access to a fuller set of bidstream signals. Where they disagree is on attribution, whether the previous generation's inefficiency came from public cloud infrastructure itself or from the alternative of maintaining private, self-managed data centers.

That disagreement carries commercial implications. If OpenX's argument holds, the meaningful signal for evaluating a supply-side platform is not which cloud provider it uses, but rather its published latency service level objectives, its queries-per-second ceiling, and the extent to which partner logic actually executes inside the auction rather than through an external call. Vendor-reported figures like the 3 million queries-per-second claim and the roughly 5-millisecond container execution window are useful starting points for that evaluation, but they remain self-reported, and the source material itself does not claim outside verification.

The dispute also arrives at a moment when the supply side is investing heavily in public positioning around infrastructure efficiency, tying technical architecture to environmental credentials and operational reliability. OpenX's repeated citation of its 2022 CarbonNeutral certification, first raised in its Gold Standard statement on August 20, 2026 and now repeated in this post, suggests infrastructure marketing is becoming a competitive front in its own right, one that buyers evaluating supply-side partners will likely encounter more often as rival platforms publish their own infrastructure claims in response.

Timeline

Summary

Who: OpenX, an independent supply-side advertising platform headquartered in Pasadena, California, and its chief technology officer, Joel Meyer, who authored the post.

What: OpenX published a company blog post arguing that public cloud infrastructure is not the source of hidden costs in programmatic advertising, disputing the "cloud tax" framing that Index Exchange chief executive Andrew Casale introduced in February 2026. The post claims OpenX supports 3 million queries per second, can deploy new regional capacity in 24 hours, and runs partner decisioning logic inside its auction in approximately 5 milliseconds through its OpenXBuild product.

When: OpenX published the post on September 2, 2026.

Where: The statement was published on OpenX's company blog and applies to the company's global cloud-native infrastructure supporting connected television, mobile, and desktop advertising inventory.

Why: The post matters to media buyers and publishers because it responds directly to a cost argument that has shaped supply-side platform infrastructure decisions across 2026, including Bedrock Platform's April 2026 containerized deployment inside Index Exchange and PubMatic's own Decision Fabric product. Whether buyers should weigh a platform's cloud provider or its published latency and capacity figures more heavily is a question this dispute leaves open, since neither side's specific performance numbers carry independent verification.