Amazon Advertising runs on a mechanism that would, if Datonics LLC has its legal theory correct, constitute direct infringement of U.S. Patent No. 10,984,445. The patent — titled "Causing Delivery of Electronic Advertisements Based on Provided Profiles" and granted April 20, 2021 — describes a system where a central entity collects visitor profiles from publisher sites, segments those profiles by behavioral signal, and distributes matching audiences to multiple advertisers simultaneously. This is, in simplified form, exactly what Amazon's DSP does when it ingests browsing data from Fire TV apps, Amazon.com behavioral signals, and Twitch viewing patterns, then makes those audience segments available to any brand willing to bid.
Amazon's position on this, formally stated in an Inter Partes Review petition filed April 18, 2025, is that the mechanism described in the patent is obvious. That prior art — specifically Gilmour, Merriman, and Frauenhofer — rendered the claims invalid before Datonics ever filed its continuation application in 2018. Amazon was not infringing, in Amazon's legal theory, because the invention was never novel enough to deserve a patent in the first place.
The petition was IPR2025-00873, submitted to the Patent Trial and Appeal Board. The PTAB institution decision, issued October 31, 2025, found sufficient merit to proceed to trial on all 14 claims. Amazon's challenge was live. The case was live. And Amazon continued to run its ad network the entire time.
The '445 patent, filed originally as a continuation application tracing back to a 2006 provisional, claims a specific architecture: a "profile owner" computer system that aggregates visitor data across a network of publisher sites, maintains behavioral profiles on each visitor, and then — when a "media property" requests an audience matching specific characteristics — delivers those profiles for ad targeting purposes. The media property never sees individual user data; it receives a matched segment. The profile owner handles the matching. The targeting is mediated through a layer of abstraction.
This is not a patent on behavioral targeting generally. It is a patent on a specific distribution architecture: the profile aggregator as intermediary between publisher data and advertiser demand. The model underlies what the industry calls "people-based" or "identity-based" programmatic targeting. LiveIntent uses it for its email-based identity graph. LiveRamp uses it for its RampID onboarding and distribution network. The Trade Desk's UID2 operates on a similar structural principle.
Datonics, now a subsidiary of AlmondNet alongside Intent IQ, built its licensing business on this portfolio — 170-plus granted patents covering targeting, behavioral advertising, and identity resolution. Roy Shkedi, AlmondNet's CEO, described the portfolio's scope in April 2026: a multi-year licensing program covering publishers, DSPs, data platforms, and technology companies operating across the programmatic stack. The $136 million jury verdict in the Western District of Texas — AlmondNet Inc. et al v. Amazon.Com Inc. et al., Case No. 6:24-cv-00234 — was the enforcement action for one tranche of that portfolio. Amazon owed money for using the patented architecture without a license.
Amazon's response was the IPR: challenge the patent's validity at the USPTO while litigating the infringement case in federal court. If the claims fell at the PTAB, the district court case became significantly weaker. If the claims survived, Amazon had at least delayed the payment and built leverage for a settlement on better terms.
This is standard practice among large technology companies. The dual-track IPR-and-litigation strategy is not unique to Amazon; it is the playbook. Apple runs it. Google runs it. Meta runs it. The sequence is familiar: get sued for patent infringement, immediately file IPRs challenging the asserted patents, use the PTAB proceedings to build leverage, settle from a position of calculated weakness rather than admit infringement at full sticker price.
What makes Amazon's case unusual is the specificity of the overlap. Amazon was not accused of infringing a tangentially related patent while its own product worked differently. Amazon was accused of running an ad network whose core audience-targeting architecture — the profile collector and distributor — operated on the same structural principle as the Datonics patent. Amazon's own DSP product was the infringing machine. And Amazon's PTAB petition was arguing that the machine should never have been patentable in the first place.
The argument had legal merit. Gilmour, published in 2004, taught a "profile service provider" system receiving queries from requesting entities and providing user profile information for targeted advertising. Merriman, a 2014 patent, taught the use of cookie-based tags to link visitor devices to profile information and real-time redirects for ad delivery. Frauenhofer — a research paper, not a patent — taught a system storing user requests and automatically performing ongoing matching to select users for content delivery. Individually, none of these taught every element of the '445 patent. In combination, Amazon's petition argued, they rendered the claims obvious to a person having ordinary skill in the art.
The PTAB agreed the argument had sufficient merit to proceed to trial. That trial never concluded. On April 7, 2026, AlmondNet and Amazon announced a settlement. Amazon took a license to the AlmondNet IP portfolio. Both parties jointly dismissed Amazon's appeal of the $136 million verdict. The IPR proceedings terminated. The industry's most aggressive challenger to the profile-distribution targeting model had negotiated its way to the same outcome as every other company that had tried to use the technology without a license: a paid-up right to continue.
The targeting paradox is not simply that Amazon used what it argued shouldn't be patentable. The deeper structural condition is that the entire programmatic advertising ecosystem — every DSP, every data platform, every identity resolution service — runs on mechanisms that look, structurally, like the Datonics patent. The profile aggregator collecting behavioral data from publisher sites and distributing matched audiences to advertisers is not one company's innovation. It is the foundational architecture of the open-display ad market as it has existed since the mid-2000s.
The reason the patent was worth defending — and worth challenging — is that this architecture generates real revenue. Datonics' licensing program covered not just Amazon but multiple companies in the ad tech stack. LiveIntent filed its own IPR against the same patent in July 2025, arguing obviousness over Beyda and Herz. The LiveIntent petition made a similar case: the profile-distribution targeting model was established prior art by 2012, and Datonics' claims should not have issued.
Two separate challengers. Two separate obviousness arguments. Both procedurally viable. Neither completed. The patent remains alive. The licensing program continues. The layer nobody sees — the profile aggregator in the middle of every programmatic transaction — continues to function as both a legal asset and an operational necessity.
What changes if the claims had fallen? The answer is structural. Dozens of companies operating the profile-distribution layer would have faced reduced licensing costs and lower litigation exposure. New entrants would have faced lower IP barriers. Amazon's own $136 million liability would have been vacated. The open web's targeting infrastructure would have been slightly cheaper to operate. It would not have changed how the ads are bought or sold. It would have changed who pays whom for the right to do it.
What changes because the claims survived? The patent holder maintains leverage. The licensing program continues. The industry's cost of doing business includes a line item for the profile aggregator's cut. And Amazon — having negotiated a license — now has the right to keep running exactly what it was already running, having paid for it retroactively.
There is a second paradox embedded in this story, one that does not appear in the legal filings. The profile-distribution targeting model — the mechanism at the center of both the patent dispute and Amazon's own ad network — has never had a clean empirical account of its actual effectiveness. The economic literature on behavioral targeting's CPM premium has been contested since Beales published "The Value of Behavioral Targeting" in 2010, finding meaningful price premiums for well-targeted display inventory. Subsequent work, including the Perla et al. meta-analysis, has found that end-to-end effectiveness of behavioral targeting is modest once you account for data leakage, selection bias, and measurement artifacts.
The profile aggregator layer sits in the middle of a market whose participants cannot agree on whether the thing being sold is actually more valuable than its contextually targeted alternative. Amazon's DSP growth — from $37.7 billion in advertising revenue in 2022 to a projected $70 billion in 2026 — suggests buyers believe the targeting is worth the CPM premium. Amazon's own patent challenge argues the mechanism producing that premium was never a novel invention. Both things can be true simultaneously. The market pays for a targeting premium that the legal system has not yet determined required any actual innovation to produce.
This is the targeting paradox as a structural condition, not merely a legal one. The entity that built the targeting infrastructure needed to argue, for legal and competitive reasons, that the infrastructure was obvious. The market that pays for the infrastructure continues to behave as though the infrastructure has proprietary value. Both Amazon and its competitors operate inside this contradiction. The settlement did not resolve it. It added a line item.
The April 7, 2026 announcement was three paragraphs long. Roy Shkedi's statement called it "a years-long journey" and said it "feels good to have this dispute behind us." No financial terms were disclosed. Amazon did not issue a statement. The joint dismissal of the appeal was filed April 6. The district court case closed. The IPR terminated.
What the settlement preserved: the patent, the licensing program, Amazon's right to operate its targeting infrastructure, and the industry's ongoing obligation to the profile aggregator layer as a cost of doing business.
What it did not resolve: the underlying question of whether the profile-distribution targeting model constitutes a patentable technical improvement or an abstract idea implemented on generic computer hardware. The Federal Circuit's February 2026 remand of a related Datonics patent application sent the issue back to the USPTO for further consideration without ruling on eligibility. The Alice/Mayo doctrine — which has held, in case after case, that targeting advertising via computer networks is an abstract idea insufficient to survive step one of the patent eligibility analysis — remains applicable and unresolved on this patent family. Customedia Technologies v. Dish Network (2020) held exactly this. Chewy Inc. v. IBM (2024) extended it. Integrated Advertising Labs v. RevContent (2024) affirmed it by Rule 36.
The '445 patent survived the IPR challenge because the proceedings ended before a final written decision. It survived the district court litigation because Amazon settled rather than appeal an adverse verdict it had already exhausted its post-trial options on. It has not survived the consistent doctrinal trend in the Federal Circuit, which holds that this category of invention — matching consumer profiles to advertisements using a computer network — does not constitute a patentable technical improvement under the controlling precedent.
The targeting paradox does not resolve. It reproduces at the next patent, the next challenger, the next company that discovers it has been running an ad network on someone else's claimed invention. Amazon is not the last company to find itself in this position. The structure of the programmatic advertising ecosystem — with its layered intermediaries, its profile aggregators, its distributed data collection — was built over twenty years by hundreds of companies making independent decisions about how to match advertisers to audiences. The intellectual property framework is attempting to assign ownership to a set of structural choices that, in aggregate, constitute the industry's operating assumptions.
Whoever holds the enforceable patent on the profile aggregator wins the right to charge rent on the infrastructure the entire industry built.