The $1.5 Trillion AI Problem Nobody Talks About - Until Now

David Carmel is not a trained data scientist. He is a former corporate lawyer who noticed a $1.5 trillion problem hiding in plain sight: most enterprises cannot actually use AI because their data is either trapped in legacy mainframes, scattered across unstructured cloud storage, or duplicated in ways nobody has audited. He built DataRockit to solve the plumbing problem before the AI strategy problem. His claim: data is just data until it is clean enough for AI.

The Wall Street Journal calls this technical debt $1.52 trillion per year. The US government alone spends $80 billion annually on legacy system maintenance that could be replaced. 75% of the Fortune 500 still run mainframes. 99% of the world has unstructured cloud data problems. And 90% of enterprise data is unstructured. The reason most AI deployments underperform is not the model - it is the fuel going in.

The Victorian Plumbing Problem

David's most useful metaphor: AI adoption is blocked by Victorian era plumbing. The pipes are old, incompatible with modern infrastructure, and nobody wants to tear out the walls to replace them. The result is enterprises doing the splits - one foot in mainframe legacy systems, one foot in cloud, and the two systems do not communicate. Data scientists spend their time functioning as potato peelers: cleaning dirty data by hand instead of extracting insight from it.

DataRockit's core operation: clients upload their data - legacy mainframe exports, unstructured cloud files, or bloated LLM datasets - and receive it back in any format they choose, cleaned and AI-ready, in hours rather than years. The dashboard accepts data from multiple continents simultaneously, can combine 40 mainframes from 300 different locations into a single clean output, and scales from 5 gigabytes to petabytes. Clients also receive a full index showing what was done, with optional recommendations - but DataRockit does not tell clients how to use their data. Their job is to put it in the best position.

Three Layers of AI Data Readiness

David identifies three distinct data problems that each block AI differently:

• Legacy mainframe data: Structured but siloed in systems designed before the internet. Expensive to maintain, incompatible with modern APIs, and unable to communicate with cloud infrastructure. The US government spends $80 billion annually on this problem.
• Unstructured cloud data: 90% of all enterprise data. Documents, emails, PDFs, recordings, images - files that exist in cloud storage but have never been organized, deduplicated, or formatted for machine consumption. DataRockit finds companies with the same data duplicated in 1,500 different locations.
• Over-provisioned LLM data: Enterprises feeding large volumes of unoptimized data to LLMs, using enormous compute to compensate for data quality problems that could be solved upstream. David's framing: you are dropping bunker busters when you need a scalpel.

Outcome-Linked Pricing for Enterprise

David is candid that pricing is a work in progress - and his model reflects a philosophy rather than a fixed structure. The target: seven-figure annual revenue per client. The structure: an upfront fee to begin the engagement, followed by a percentage of the savings the client realizes from eliminating their legacy technical debt and cloud waste, plus a monthly support fee for ongoing data hygiene.

The logic is deliberate alignment. If DataRockit participates in savings, both sides are incentivized to maximize the outcome. The vendor is not paid for the work - they are paid for the result. David also insists on the right to share the client as a case study: the ability to showcase what was done together is part of the deal. For clients that cannot accept outcome-linked pricing, DataRockit will negotiate a fair flat rate - but the outcome model is the preference because it puts both parties' incentives on the same side.

Distribution-First Enterprise GTM

Getting a direct meeting with a Fortune 500 CIO as an unknown startup is nearly impossible. David's solution: find people who already have a seat at the table and give them a reason to introduce DataRockit. These are value-added resellers, managed service providers, and VC firms with enterprise portfolios - people who serve the target buyer and are actively looking for the next differentiating product to bring to their clients.

David is also piloting an unconventional model: working with VC firms not for funding, but as distribution partners. The VC has portfolio companies and LP relationships with the banks and enterprises DataRockit needs to reach. The VC can earn meaningful revenue as a reseller. Neither party needs to close a funding round for the relationship to create value. This approach treats distribution as a co-investment rather than a sales channel - and offers a model for other B2B founders trying to reach enterprise buyers through trusted intermediaries.

Q&A