Karp Says Frontier AI Labs Are Stealing Enterprise Value And VCs Are Listening

Palantir CEO Alex Karp has dropped a bombshell on the AI industry, accusing frontier AI labs of overselling their models while quietly extracting valuable intellectual property from enterprise customers. In a CNBC interview that sent ripples through Silicon Valley, Karp warned that companies like OpenAI and Anthropic are capturing disproportionate value from the relationship, leaving enterprises with commoditized tools and compromised data. His critique comes at a pivotal moment when venture capitalists are rethinking where the real money in AI will be made.

Karp, whose company built the foundational software for counterterrorism and data analytics, argues that the current AI boom is structurally flawed. Frontier labs—those at the cutting edge of large language models—promise transformative productivity gains but, according to Karp, deliver closed ecosystems that siphon enterprise secrets and train on proprietary data without fair compensation. He called the dynamic 'stealing enterprise value,' a phrase that resonated with investors already wary of AI hype cycles.

The underlying tension is between model builders and application layers. While OpenAI, Google DeepMind, and Anthropic race to build ever-larger models, Palantir focuses on deploying AI within secure, sovereign enterprise environments—what Karp calls the 'application and sovereignty layer.' This layer protects sensitive corporate and government data, ensures compliance, and retains value for the customer. Karp's argument is that without such protections, enterprises become mere feeders for the AI giants' training sets.

VCs are taking notice. Several prominent venture firms have begun shifting deal flow from pure-play AI model companies toward infrastructure and application startups that prioritize data sovereignty. This mirrors a broader trend: as the cost of training frontier models skyrockets—some estimates exceed $1 billion per model—the return on investment for enterprises remains uncertain. Karp's warning aligns with growing concerns that AI labs may follow the playbook of Big Tech, locking customers into ecosystems while extracting data and recurring fees.

Palantir itself has doubled down on this narrative, recently launching AIP (Artificial Intelligence Platform) for sovereign deployments. The platform allows enterprises to run AI on their own infrastructure, using their own data, without sending it to third-party labs. This positions Palantir as a direct competitor to the frontier labs' enterprise offerings, but also as an alternative that promises value retention.

The implications are significant. If Karp is right, the current wave of AI spending—over $200 billion in global corporate investment in 2025—could be misallocated. Enterprises pouring money into frontier model subscriptions may be sacrificing long-term competitive advantage for short-term productivity gains. Meanwhile, VCs are already recalibrating: sovereign AI, application-layer startups, and privacy-focused infrastructure are becoming hot sectors.

Karp himself did not mince words during the interview, stating that 'the labs are taking more than they're giving.' Investors are now asking which companies will benefit from the inevitable shift toward enterprise-centric AI. The answer may determine the winners of the next technological cycle. As for Palantir, its stock saw a modest uptick following the interview, reflecting market belief that Karp's perspective has merit. The coming quarters will reveal whether frontier labs adjust their models or face a backlash from cost-conscious enterprises.