Databricks CEO Breaks Down the AI Hype-Cycle
EP 124 of The Logan Bartlett Show: Untold stories from tech's inner circle
Ali Ghodsi, CEO of Databricks (now a $61B company), has become one of the top minds in AI, evolving from researcher to leading one of the biggest AI-centered private companies. In our conversation, he broke down the AI hype cycle in detail and made bold predictions for the future.
Click here to view the episode transcript | Watch on Youtube | Listen on Spotify or Apple Podcasts
✉️ Episode Memo
→ What people underestimate with AI (reading vs writing)
There’s probably a 10:1 ratio on how often people are using AI to read vs write, and people underestimate how fast AI is at reading data compared to writing outputs. While we often focus on the output AI generates, its reading power offers enormous potential on its own. For example, AI is already being used in finance to analyze SEC filings and in healthcare to structure handwritten medical records. Beyond these, hundreds more untapped use cases across every industry have yet to be productized, and we’re still in the early stages of discovering where AI can automate the process of crunching information at scale.
→ Super AGI is still far off
Ali thinks we’re going to see a slowdown in progress toward AGI. Over the past decade, we’ve had a 100 million times improvement in compute to throw at the problem, but the next few years are unlikely to sustain that pace. Ali would find the idea of cracking super AGI soon (requiring a recursive self-improvement loop) more likely if GPT-5 only took $1 & 1 second to make, or if progress were becoming cheaper and faster. Instead, we’re moving in the opposite direction, as the next models will be more expensive, involve more people, and require greater caution. Thus, Ali expects humans will remain essential for many years to ensure models are verifiable.
A few more bull and bear takes:
→ We’re likely near the top of the hype cycle: Ali predicts a period of disillusionment in AI, where many companies will struggle or fail, and this will fuel a narrative of AI being overpromised. However, game-changing applications for humanity are likely to emerge afterward.
→ Compound AI systems will allow for faster improvement: Multiple components working together make it easier to improve AI output for specific tasks, as it’s easier to consistently debug and upgrade each part than to improve one gigantic, complex model.
→ Many startups will disrupt incumbents: Ali believes many startups (especially those with a data advantage) can disrupt slower-moving incumbents.
→ Most companies shouldn’t build their own chat apps: Many rushed to create internal AI tools (ex: for HR), but a single SaaS provider owning the workflow makes more sense. Ali sees this as the beginning of a growing trend in specialized AI companies.
⭐ Trailer
📱 Follow The Show!
If this was helpful to you, please consider forwarding!