The Big AI Privacy Problem Costing Enterprises Billions

There is a quiet revolt happening in corporate America, and it is not about valuations or interest rates. It is about who owns your data when you hand it to an AI. Andrei Jikh has spent three recent videos hammering a single thesis: the cloud AI business model has a fundamental trust problem, and the businesses that figure this out fastest will win.

The core concern is not paranoia. Jikh points to Anthropic launching Claude Design while simultaneously partnering with Figma as a concrete example of an AI vendor using client relationships to encroach on a client's own market. The worry is structural: every prompt you send through a third-party AI potentially teaches that system your proprietary workflows, trade secrets, and competitive edges. As Jikh puts it, you are essentially paying to train your future competitor.

The response from technically sophisticated businesses is increasingly on-premise AI deployment. Own the GPUs. Own the model weights. Own the outputs. And critically, keep hyperscalers from ever seeing the data. Jikh also raises a pointed question about pricing: if OpenAI and friends truly believed AI delivered the value they market, they would price on outcomes, not tokens. A flat per-token fee regardless of results is, he argues, a tell. Real confidence looks like revenue share. Token billing looks like SaaS with better PR.

Meanwhile, Alphabet ($GOOGL) is facing its own version of this tension internally. A CNBC report this week revealed that Google Cloud and DeepMind are in a zero-sum fight over TPU compute capacity, with researchers frustrated that cloud customer allocations leave too little headroom for training frontier models. The company that invented the transformer architecture may be too commercially successful at cloud to stay at the frontier of AI research. That is a genuinely weird problem to have.

The other headline out of Alphabet this week is a leadership reshuffling that sent shares down roughly 4%, with Demis Hassabis stepping into a chairman role and away from day-to-day AI operations. Replacing the person who ran DeepMind at a moment when compute strategy is the most important question in tech is the kind of timing that makes investors nervous.

The broader takeaway from this week's YouTube creator consensus: AI infrastructure spending is real and accelerating (Alphabet just issued a 10-tranche bond drawing $115 billion in demand to fund roughly $300 billion in AI capex), but the business model layer above that infrastructure is still being figured out in real time. The companies charging per token are hoping you never do the math on what they are actually learning from you.