Global digital advertising crossed $700B in 2025. Nearly every dollar of it is now touched by an AI system — bid, placed, optimized, and measured without a human making the core decision.
This did not happen with a press release. It happened quietly, campaign type by campaign type, auction by auction, until the humans in the loop became the exception rather than the rule. Ad tech is not becoming an AI industry. It already is one.
The Three Layers AI Now Controls
The AI takeover in advertising happened in three distinct layers, each reinforcing the others and widening the moat for platforms that got there first.
Layer 1: Bidding and Auction Optimization
Google processes over 8.5 billion searches per day. Meta serves ads across 3.2 billion daily active users. Neither platform lets humans set individual bids anymore — AI models evaluate each impression in under 100 milliseconds, factoring in user intent, historical behavior, advertiser CPA targets, and inventory supply. Smart Bidding on Google and Advantage+ Shopping on Meta now manage the majority of spend on both platforms, and their performance has consistently exceeded manual bidding by 15-30% on conversion rate metrics in controlled tests.
Layer 2: Creative Generation and Testing
The creative bottleneck used to limit ad testing to 5-10 variants per campaign. AI-powered creative tools have removed that ceiling entirely. Google's Responsive Search Ads test up to 43,680 combinations of headlines and descriptions. Meta's Advantage+ Creative automatically adjusts image brightness, aspect ratio, and overlay text based on real-time performance data. Third-party tools from companies like Smartly, AdCreative.ai, and Pencil now generate hundreds of creative variants from a single brief — cutting creative production cost by 60-80% while increasing the volume of tested combinations by 10x or more.
Layer 3: Measurement and Attribution
The deprecation of third-party cookies destroyed the last illusion of deterministic attribution. In its place, AI-powered probabilistic modeling has become the industry standard. Meta's Conversions API, Google's enhanced conversions, and platforms like Northbeam, Triple Whale, and Rockerbox all use ML models to infer attribution when the traditional cookie signal is absent. This is not a workaround — it is a better architecture for a privacy-first world, and the companies with the largest first-party data pools have the most accurate models.
Who Is Winning and Why
The ad tech AI race is not even close to competitive outside of a handful of platforms. The moat is data volume, and the leaders have a compounding advantage that independent vendors cannot replicate.
Search intent signal + YouTube engagement + Maps data. Performance Max campaigns grew to represent 40%+ of Google Ads spend by 2025. AI controls the full funnel across six properties simultaneously.
Meta
Social graph + behavioral data across Facebook, Instagram, WhatsApp, and Threads. Advantage+ campaigns showed 32% lower cost per acquisition than standard campaigns in Meta's own published benchmarks.
Amazon Ads
Purchase intent data nobody else has. Amazon's ad business crossed $55B in 2024 and grew 19% YoY — powered entirely by AI models trained on transaction history, not just browsing behavior.
The Trade Desk
Independent DSP playing the open internet angle. UID2 is their bet on a cookieless identity layer. AI-powered Koa bidding algorithm is the product differentiation. $2.4B revenue in 2025 growing 22%.
Where Startups Can Still Win
I've seen dozens of ad tech pitches. The ones that get my attention are not trying to out-Google Google. They are finding the gaps the platforms create by being too large, too generic, or too focused on their own ecosystem.
- →Retail media infrastructure — Walmart, Target, Kroger, and hundreds of retailers are building ad networks but lack the AI tooling to monetize them at scale. This is a $50B+ market that the big platforms barely touch.
- →Vertical-specific creative AI — Legal, healthcare, financial services, and real estate all have compliance requirements that generic creative AI cannot handle. A startup with domain-specific guardrails and fine-tuned models has a real moat.
- →Cross-platform incrementality measurement — Advertisers running on Google, Meta, Amazon, and TikTok simultaneously have no unified truth. The companies building truly platform-agnostic measurement using causal AI (geo experiments, synthetic control groups) are solving a real, expensive problem.
- →B2B intent data and signal — LinkedIn is the only premium B2B ad platform and it is expensive and limited. Startups that can build AI-powered intent models on top of proprietary B2B behavioral data are building something the platforms cannot easily replicate.
- →Creative performance prediction — Before you spend $100K testing a campaign, an AI model that predicts performance from creative assets alone has enormous value. This is a pure data and ML problem that does not require competing for inventory.
The Cookie Collapse Was a Gift to AI
The conventional wisdom was that removing third-party cookies would hurt ad tech. The reality is the opposite for AI-native platforms and a death sentence for legacy vendors.
Companies that relied on deterministic cookie-based targeting had a brittle, data-leaky system that regulatory pressure was always going to dismantle. AI-powered probabilistic modeling — trained on first-party signals, contextual data, and cohort behavior — is both more privacy-compliant and, counterintuitively, often more accurate at scale.
Google's Privacy Sandbox, Meta's Conversions API, and Amazon's AMC (Amazon Marketing Cloud) all pushed the industry toward server-side, first-party data architectures. The winners of the cookie transition are the same platforms that won before it — because they have the most first-party data to train on. The gap between walled gardens and the open web just got wider.
The ad tech AI story is not about what is coming.
It is about what already happened — while everyone was watching the foundation model wars, advertising became the largest deployed AI application on earth.
Tracking AI investment themes across verticals at Value Add VC. Originally published in the Trace Cohen newsletter.