Which AI Ad Products Would the Sanders Bill Force Apart?

This is probably not a 2026 compliance deadline. The American AI Sovereign Wealth Fund Act was introduced by Sen. Bernie Sanders on June 18, 2026, and Roll Call’s read was that passage is highly unlikely in the current Republican-controlled Congress.[1] That caveat belongs at the top, because the wrong version of this story turns into “your Performance Max campaigns are about to be split off next quarter.” They are not.

The useful version is narrower and more operational: if the bill’s categories were applied to advertising platforms, the major AI ad products buyers use every day would likely be treated as AI services, not just campaign types. For buyers trying to understand the advertising impact of Sanders’ AI companies bill, the key point is that the mechanism here is not a routine monthly fee on ad accounts. It is an equity-linked AI tax and a structural-separation requirement aimed at large AI activity.[2]

That table is an analytical mapping, not a regulator’s classification. No platform compliance scenario from Google, Meta, Microsoft, TikTok, The Trade Desk, or Criteo has been cited here, and this assessment relies on published analysis rather than a direct review of the bill-text PDF. The legal mechanics below come from those published analyses; the product consequences are the practical read from those mechanics.

The tab that breaks first is structural separation

The bill’s hard trigger is not “uses AI somewhere.” Companies with more than $200 million in annual AI-related gross receipts would have to structurally separate AI from non-AI operations within 90 days.[2] The separated entities could not share equity, directors, or management.[2] That is the part that turns a tax-and-industrial-policy proposal into an ad-account problem.

The bill defines covered AI trades or businesses to include AI data centers, AI computing infrastructure, AI services, and advanced robotics.[2] The advertising question sits inside “AI services,” which includes the development, distribution, and sale of an AI model trained using computing power above a threshold set by the Secretary of Commerce.[2] Commerce setting the compute threshold is important because the bill does not need to name Performance Max, Advantage+, AI Max, or Symphony Agent. It can sort the underlying model class, then pull the ad product along with it.

For campaign managers, “structural separation within 90 days” would not feel like a white paper category. It would feel like a platform changing where the product lives. If an AI ad product had to sit in a separated entity with no shared management or equity, the operational seam could appear in account access, billing, support ownership, reporting continuity, conversion-data sharing, audience availability, creative libraries, API endpoints, and product permissions. Those are scenario-based implications, not announced compliance plans.

The $200 million threshold also matters because it is not limited to the handful of companies usually described as frontier AI labs. American Action Forum’s analysis argues the plan would reach a broader set of AI firms, and its description of the 50% equity tax illustrates the severity: a company would effectively mint 100 new shares and hand them to Treasury, doubling total shares to 200.[3] That equity-transfer mechanism is not the central advertising issue, but it explains why any covered platform would have a strong incentive to challenge, delay, or redesign around the rule.

Why ordinary campaign labels do not settle the legal classification

Ad platforms package AI as familiar media-buying workflow. Google calls Performance Max a campaign type. Meta frames Advantage+ as automation. Microsoft presents AI Max as a search-ad enhancement. TikTok’s Symphony Agent is closer to the label regulators would understand because the agentic layer is explicit. Those product labels help buyers know where to click. They do not answer whether the platform is distributing or selling access to an AI model service trained above a Commerce-set compute threshold.

Performance Max is the easiest example because it bundles so many decisions into one product surface: bidding, channel allocation, asset combinations, audience expansion, and conversion-seeking delivery. A buyer may only see a campaign settings panel, but the value being sold is model-mediated decisioning across Google inventory. If the model layer clears the compute threshold, the practical classification risk is that PMax becomes part of an AI services business even though it is purchased inside Google Ads.

The same logic applies to Meta Advantage+. Its buyer-facing promise is fewer manual inputs and more automated discovery across audience, creative, and placement decisions. That makes it attractive when performance is stable and hard to audit when performance degrades. Under the Sanders framework, the audit question would be different: not “can the buyer control the automation,” but “is this product distributing AI model capability as a covered service?”

Microsoft AI Max adds another wrinkle because it sits in search advertising, a workflow many buyers still mentally separate from generative-AI products. The Copilot-integrated layer narrows that distance. If the product depends on model-driven query, creative, landing-page, or matching expansion, it is hard to argue the AI component is merely incidental. Again, that is a product-mapping inference, not an official classification.

TikTok Symphony Agent is the cleanest fit because the product language already points toward an AI agent assisting campaign creation and optimization. If regulators start from function rather than dashboard placement, an agent that creates, optimizes, or manages advertising work is less likely to be treated as a normal media-buying control and more likely to be examined as an AI service layered into advertising.

The edge cases are not small enough to ignore. The Trade Desk’s Koa and Criteo’s AI-engineered commerce media do not look like consumer AI labs, but the bill’s threshold is based on AI-related gross receipts, not cultural visibility. A large DSP or commerce-media company using AI-powered bid optimization could plausibly enter the same conversation if the covered receipts and model-threshold tests were met. That is where the $200 million line starts to look less like a frontier-lab screen and more like a serious ad-tech screen.[3]

What separation would mean inside a campaign workflow

The clean legal phrase is “no shared equity, directors, or management.”[2] The messy implementation question is whether an AI ad product could still share the same ad account, identity graph, conversion tags, billing profile, product support team, reporting interface, and API permissions as the non-AI advertising business. If the answer were no, the campaign manager would inherit the breakage.

Start with access. A buyer running search, shopping, YouTube, display, and PMax from one Google Ads account today expects shared users, shared permissions, and a shared account history. A structurally separated PMax entity could require separate account access or a separate product console. That does not mean it would; no platform has announced such a scenario. But if management and business control must be separated, shared operational administration becomes a natural pressure point.

Reporting is the next seam. Most buyers do not care whether a conversion path was legally handled by an AI services subsidiary or a non-AI ad platform. They care whether spend, impressions, clicks, modeled conversions, revenue, and asset-level diagnostics still reconcile. A forced split could create duplicate reporting exports, changed attribution windows, interrupted historical comparisons, or fields that no longer populate across product lines. None of that is stated in the bill. It is what happens when a platform product that used to behave as one system has to prove it is not one system.

Conversion signals would be harder. PMax, Advantage+, AI Max, and Symphony-style workflows all depend on feedback loops: conversion events, value signals, audience behavior, creative performance, and inventory outcomes. If a separated AI services business could not freely share pipelines with the non-AI ad business, buyers might see weaker learning continuity or new consent and data-transfer prompts. The bill mechanism supports the possibility of a separation problem; it does not specify the platform architecture that would result.

The transition window is where agencies would feel the risk most acutely. Ninety days is short for corporate separation and painfully short for advertising systems that need stable tags, APIs, feeds, catalogs, creative libraries, billing records, and client approvals.[2] A platform might preserve most of the front-end workflow through contractual or technical bridges. It might also freeze migrations, pause feature releases, or limit experiments while legal teams decide what can cross the wall. Buyers would not control that timeline.

This is the practical reason not to wave the bill away just because passage odds are poor. Regulatory pressure and financial pressure change platform behavior before they change law. The same pattern shows up when platforms push more aggressive defaults under market pressure, which is why the adjacent discussion of AI stock selloffs and platform defaults is relevant here. A platform does not need a final statute to start designing products around the categories it expects regulators to use.

The bill points in the opposite direction from current AI ad-platform momentum

The near-term market direction has been integration: more AI inside the same ad account, more automation inside the same campaign object, and more cross-surface optimization inside the same reporting layer. That is the product logic behind PMax, Advantage+, AI Max, and Symphony Agent. The Sanders bill uses the opposite organizing principle. It asks whether AI activity is systemically important enough to be separated from the rest of the business.

That makes the proposal a useful counterweight to the deregulation track. If deregulation lets platforms ship AI ad products faster, structural separation asks where the platform stops and the AI service begins. The contrast matters for buyers who are already watching policy move in the other direction through Trump’s AI deregulation and ad-tech performance. One direction reduces friction for platform integration; the other creates a vocabulary for forced boundaries.

The broader “AI nationalization” label is too blunt for campaign planning, but the neighboring issue is real: government claims over frontier AI capacity can reach the ad-tech supply chain if the same models power advertising automation. That is why the AI nationalization risk discussion is more useful when translated into product boundaries than when left as ideology.

The bill’s data-center and computing-infrastructure categories also matter, though they are not the center of this article.[2] If AI infrastructure is pulled into tax, ownership, or separation rules, the cost and availability of the model layer behind ad products changes. That connects to the separate pressure already visible in AI data center spending and structurally higher ad costs. Buyers may never negotiate compute contracts, but they do inherit the bidding, pricing, and automation defaults that come from them.

Public support is a signal, not a forecast

The politics are not irrelevant, but they should not swallow the product analysis. A Verasight/CNBC finding reported that 69% of U.S. workers supported forcing AI companies to transfer 50% of stock to a public fund.[4] That number says the category has public resonance: large AI companies, public claims, equity transfers, and a fund meant to share gains. It does not prove the Sanders bill will pass, and it does not tell us how Commerce or the FTC would classify any single ad product.

Roll Call’s passage warning still controls the near-term risk: this is highly unlikely to become law in 2026 under the current congressional alignment.[1] The better use of the bill is as a dated classification marker. It shows one way lawmakers may define the boundary between an advertising platform and an AI service: not by the tab name in the UI, but by the model capability being developed, distributed, or sold.

What to track, without restructuring accounts yet

Media buyers do not need to split campaigns, rebuild account structures, or warn clients that PMax and Advantage+ are about to disappear into separate entities. The bill is too unlikely, too new, and too dependent on future threshold-setting to justify operational changes now. The useful monitoring list is more specific.

• Watch whether future FTC or Commerce language defines AI services by model training compute rather than by consumer-facing product category.
• Watch whether ad platforms describe automated campaign products as ordinary ad features or as separate AI services, assistants, agents, or model layers.
• Watch for changes to conversion-signal sharing, API access, reporting exports, account permissions, and support ownership across AI-heavy and non-AI ad products.
• Watch ad-tech vendors outside the largest walled gardens, especially DSPs and commerce-media platforms, because the $200 million AI-related gross-receipts threshold could matter beyond the companies usually treated as frontier AI labs.

The Sanders bill is less a live campaign-management emergency than a preview of how regulators may decide where an ad platform ends and an AI service begins. If that vocabulary survives in agency rulemaking, the products most likely to be tested first are the ones buyers already rely on most: Performance Max, Advantage+, AI Max, Symphony Agent, and the AI optimization layers inside large ad-tech platforms.

References

1. Sovereign wealth fund, tax on AI companies unveiled by Sanders, Roll Call, June 18, 2026
2. Sanders calls for tax on systemically important AI activity payable in equity, Thomson Reuters Tax & Accounting
3. An End to AI Competition? Senator Sanders’ Plan, American Action Forum
4. Verasight/CNBC worker survey on AI company stock transfers, Verasight/CNBC, July 12, 2026