Podcast
Are Companies Giving Away Their Secrets to AI?
The Enterprise AI Show
- CEOs Fear Frontier Models Steal Corporate Alpha
- CEOs feel distrust toward frontier AI labs because they believe model interactions expose proprietary business data.
- Brian Gracely highlights Alex Karp’s claim that companies fear LLM prompts and usage let labs learn their “alpha” and steal competitive advantage. (Time 0:03:36)
- Palantir NVIDIA Partnership Used To Spotlight Sovereignty
- Gracely recounts Palantir and NVIDIA announcing a sovereign AI cloud and Alex Karp using the CNBC platform to warn CEOs.
- The pairing illustrates software and hardware vendors promoting sovereignty as an alternative to public frontier models. (Time 0:05:23)
- VPC Privacy Is Often Illusory
- Private VPC deployments give a false sense of complete isolation because calls still route to centralized frontier models.
- Gracely stresses that VPC hosting doesn’t carve off an independent model; your environment still touches the main model. (Time 0:10:11)
- Factor True Competitive Cost Into AI Decisions
- Enterprises should factor the full competitive cost of sharing data with external LLMs when deciding deployments.
- Gracely relays Karp’s view to value decades of proprietary knowledge and include that loss in true cost calculations. (Time 0:13:44)
- Why Charge Tokens Instead Of Outcomes
- Alex Karp questions why frontier labs charge by tokens rather than by measurable outcomes.
- Gracely explains the argument: if models can deliver outcomes, customers would prefer outcome-based pricing tied to business value. (Time 0:16:06)
- Token Pricing Masks Unsustainable Economics
- Token pricing may hide subsidy dynamics; frontier labs appear to subsidize low prices while still struggling for profitability.
- Gracely notes this contradicts the expectation that token models would yield large ad-style or subscription profits. (Time 0:19:35)
- True Cost Must Include Decades Of Proprietary Value
- True cost includes the value of decades of proprietary processes and data that enterprises have built.
- Gracely warns companies undervalue that IP when they allow frontier models access, underestimating how quickly it can be synthesized. (Time 0:21:20)
- Aggregated Data Lets Labs Evolve Into Competitors Fast
- Frontier labs can rapidly productize capabilities built from aggregated customer interactions, displacing incumbents.
- Gracely likens this to early cloud providers who turned customer patterns into native services that competed with their clients. (Time 0:25:11)