Who Owns Intelligence? Eddie Drake on AI, Intellectual Property, Data Clouds, and Why Trust Will Decide Enterprise AI

Full analysis

Eddie Drake, Industry Principal at Snowflake, went on Signal & Noise with hosts Rio Longacre and Brett House to argue that the AI battle has moved off model quality and onto proprietary data. His pitch: brands are handing their strategic patterns to martech and ad-tech vendors through standard contract language, and anonymizing the data does nothing to stop it. It's a vendor-shaped argument that also happens to be mostly right.

What's actually being decided: whether ad-tech and martech vendors get to keep the royalty-free "use your data to improve our services" clause that has been boilerplate for a decade, now that "improve our services" means "train models that also serve your competitor." This is Type 1 for the vendors who depend on that data (hard to reverse once buyers wise up), Type 2 for brands (they can renegotiate at renewal). No forcing function beyond the next RFP cycle. No news, no deal, no regulator. A framework fight, not an event.

The Market Analyst. Strip away the Snowflake jersey and the structural read holds: data gravity is moving to the warehouse, and every vendor built on extract-and-host is on the wrong side of it. Drake names Hightouch, cites zero-copy activation, says identity resolution is "up for grabs." That's the CDP business getting hollowed out in real time. For an informed outsider: the value used to sit in the company that copied your data and organized it; now the data stays put and the organizing tool has to come to it. The losers are anyone whose moat was possession. LiveRamp, Experian, the classic identity graphs, the ingest-heavy CDPs. The winners are Snowflake, Databricks, and vendors that already run natively inside the warehouse. This is a slow repricing of who owns the customer record, and it favors the cloud layer.

The Skeptic. Drake reviewed 98 contracts himself. Not audited, not peer-reviewed, and he sells the alternative. The whole thing has to be taken at a discount. And notice where the critique goes soft: he lets the frontier model providers off easy while hammering martech and ad-tech, which is exactly the split that makes Snowflake look like the safe harbor. The Experian MCP anecdote, nearly zero partner uptake on the agent-connection servers they built, comes from an unnamed CPO with no date and conveniently supports the "agentic AI is early" line. For a general reader: the guy warning you that vendors exploit your data sells the vault. The IP-leakage point is real. The salesman delivering it is not disinterested.

The Operator. Try to actually own the "Marketing Context Layer" Tuesday morning. It means codifying tribal knowledge, metric definitions, brand rules, into machine-readable form, jointly run by the CMO and the CIO. Two chairs that agree on nothing and share no budget. That project dies in a governance meeting before a single rule gets written. Meanwhile the renewal is due, the DSP works, and "audit the data-use clause" is item nine on a list that ends at four. What breaks at 90 days is nothing, because nobody has time. The token-routing idea is the exception: routing proofreading and content versioning to a two-generation-old model is a real invoice you can cut this quarter without a committee.

The Customer / End User. The buyer here is the brand CMO and CDO, and Drake is genuinely speaking to their fear: my agency and my martech stack know my playbook, and now that playbook trains a model my competitor rents. That lands. But brands want three sentences of cleaner language at renewal, not an architecture rebuild. The vendor that shows up with tightened AI and data-use terms, model-tier transparency, and a straight answer on what "improve our services" covers wins the RFP. The one that stonewalls gets a due-diligence checklist thrown at it. That's the practical shift: this becomes RFP language, not a re-platforming.

The CFO. The token argument is the only line in this episode with a dollar attached. If 70 to 80% of enterprise AI work runs fine on cheaper, older, or open-source models, then every AI-native ad-tech tool pricing as if every query hits a frontier model has margin the buyer is about to come after. Creative generation, audience modeling, brand safety, all of it gets a "what tier are you actually running" question in procurement. The Marketing Context Layer, by contrast, is pure cost with a payback measured in years and a governance tax up front. A CFO funds the token savings now and files the context layer under "strategic," which is where good ideas go to wait.

The tensions. Two worth naming. First, the Market Analyst says data gravity is a genuine structural shift; the Skeptic says the messenger owns the warehouse and the 98-contract study is a sales asset. Both are true, which is why the read is "right thesis, discount the source." Second, the Operator and the CFO agree the token-routing savings are the only thing that ships this quarter, while the Customer's version, tighter contract language at renewal, is the thing that actually reprices vendor relationships over the next year. The context layer everyone talked about is the least likely to get built.

What it hinges on: does the IP-leakage argument harden into RFP and due-diligence language, or stay a podcast framework? It hinges because brands act on contracts at renewal, not on theses. The council leans that the contract clause is the real lever and the architecture rebuild is the fantasy. What to verify before believing the big version: whether any large advertiser actually strikes the royalty-free data-use clause from a martech or identity contract, not whether they say they're worried about it.

Prediction: By the 2027 upfront and RFP season (Q2 2027), tightened AI and data-use language, capping or removing vendors' royalty-free right to train models on client data, becomes a standard line item in enterprise martech and ad-tech RFPs, and at least one major identity or CDP vendor publicly revises its data-use terms to compete on it.

Confidence: Medium. Buyer inertia and vendor foot-dragging control the timing even when the mechanism is real.

Why: The royalty-free "use your data to improve our services" clause was harmless when improvement meant a better product for you; it becomes a competitive leak the moment that model also serves your rival, and Drake's contract review, discount it as you like, describes language that genuinely sits in most martech and ad-tech agreements. Sophisticated brands renegotiate at renewal, and this is a cheap ask: three sentences, not a re-platform. Vendors compete on it because the first mover turns "we won't train on your data" into a sales advantage against peers who won't say the same, exactly the way privacy terms became a selling point after GDPR. The opposite outcome, everyone keeps the boilerplate, requires buyers to stay uninformed after this argument is already circulating on the podcast and conference circuit, which is the less likely world once a fear this specific is named.

Revisit by 2027-06-30: We're right if AI/data-use clause language shows up as a named requirement in enterprise martech or ad-tech RFPs and at least one major identity or CDP vendor revises its published data-use terms to win on it. We're wrong if the royalty-free training clause remains untouched boilerplate across the major vendors and no buyer moves on it.

The architecture rebuild and the Marketing Context Layer stay slideware. The contract clause is where this actually bites.