# AI Enterprise - Databricks & Glean | BG2 Guest Interview Source: https://www.youtube.com/watch?v=jA8ZQfq_Hzs Recap page: https://rapidrecap.app/video/jA8ZQfq_Hzs Generated: 2025-12-24T00:04:55.113+00:00 --- ## Quick Overview Databricks CEO Ali Ghodsi argues that the current AI landscape, despite massive capital spending, is experiencing a bubble because LLMs are becoming commoditized, emphasizing that true enterprise value and differentiation will come from proprietary, specialized data and engineering, not generic foundation models. **Key Points:** - Ali Ghodsi believes Large Language Models (LLMs) are becoming a commodity, similar to how one gets gas from any station, which leads to an AI bubble where much of the spending is misplaced. - The differentiation for enterprises will be in the data they possess and the specialized engineering applied to it, rather than relying solely on generic foundation models. - Ghodsi cites the high failure rate (95% of projects fail) and the need for deep, company-specific data understanding (like the Royal Bank of Canada example) to achieve real economic value. - He contrasts the current situation with earlier tech cycles (Internet, Mobile, Cloud) where the core technology itself was the differentiator, whereas now the differentiator is proprietary data and specialized application layers. - The second camp consists of the original creators of the technology (like the Turing Award winners), who are highly focused on fundamental advancements. - The third camp focuses on building agents that can automate complex tasks, which is where the real future value lies, not just in probabilistic next-token prediction. - The core advice for CIOs is to focus budgets on securing proprietary data and building specialized AI products on top of foundation models, rather than spending heavily on commoditized LLMs. ![Screenshot at 00:20: Databricks CEO Ali Ghodsi asserting there is an AI bubble while contrasting the commoditization of LLMs with the need for proprietary data in enterprise applications.](https://ss.rapidrecap.app/screens/jA8ZQfq_Hzs/00-00-20.jpg) **Context:** The video features a panel discussion, titled "Guest Interview: State of Enterprise AI," involving Ali Ghodsi, CEO of Databricks, and Arvind Jain, CEO of Glean, moderated by Apoorv Agrawal of Altimeter Partner. The discussion centers on the current state of Artificial Intelligence adoption in the enterprise, particularly focusing on whether the hype surrounding LLMs constitutes a bubble and where true value creation resides in the AI stack. ## Detailed Analysis The discussion on the state of Enterprise AI immediately dives into the topic of an AI bubble. Ali Ghodsi asserts that there is an AI bubble because LLMs are rapidly becoming commoditized, citing the analogy of comparing gas stations—the product (LLM output) is interchangeable, making price comparison the only differentiator. He contrasts this with earlier tech cycles (Internet, Mobile, Cloud) where the foundational technology itself was the value driver. Ghodsi points out that the high failure rate (95% of AI deployments failing) suggests misallocated spending. He emphasizes that real enterprise value comes from companies leveraging their unique, proprietary data and specialized engineering—the 'secret sauce'—rather than just using off-the-shelf LLMs. Arvind Jain agrees, noting that many companies are focused on building generalized tools or are overly reliant on consumer-facing LLMs like ChatGPT, which may not solve specific enterprise problems. Ghodsi outlines three camps: the first are those trying to solve Superintelligence (like the Turing winners), the second are the engineers building on top of existing tech (like Databricks), and the third are those building agents that automate complex workflows. He argues that the focus should shift from the commoditized LLM layer to the proprietary data layer and application layer to achieve defensible differentiation. He cites the example of financial institutions needing high-precision, deterministic outputs, which generic LLMs struggle with, highlighting the need for domain-specific knowledge and systems of record. ### AI Bubble & Commoditization - Ali Ghodsi claims LLMs are becoming a commodity, leading to an AI bubble; differentiation must come from proprietary data and specialized engineering, not generic models - The high failure rate of AI projects (95%) suggests misplaced capital spending. ### Three Camps in AI Development - Camp 1 focuses on Superintelligence (fundamental breakthroughs); Camp 2 focuses on engineering/products built on existing tech (like Glean); Camp 3 focuses on building automating agents. ### Value Accrual - Value will accrue not to the commoditized LLM layer, but to the application layer that solves specific, high-value enterprise problems (like finance or life sciences) using proprietary data. ### Data as the Differentiator - The key to defensible advantage is unique, proprietary data and the engineering required to extract value from it, which generic LLMs cannot replace. ### The Role of Agents - The future involves agents that can perform complex, sequential tasks, moving beyond simple next-token prediction, exemplified by the financial industry's need for deterministic, auditable outputs. ### Arvind Jain's Perspective (Glean) - Glean's success comes from building deep, specialized knowledge and context within the enterprise, which generic tools lack, and this specialized knowledge is what drives value. ![Screenshot at 00:04: Panelists Ali Ghodsi \(Databricks CEO\), Arvind Jain \(Glean CEO\), and the moderator assembling for the interview on the State of Enterprise AI.](https://ss.rapidrecap.app/screens/jA8ZQfq_Hzs/00-00-04.jpg) ![Screenshot at 00:21: Ali Ghodsi explaining his view that there is an AI bubble, gesturing to emphasize his point.](https://ss.rapidrecap.app/screens/jA8ZQfq_Hzs/00-00-21.jpg) ![Screenshot at 01:18: Apoorv Agrawal discussing the hype cycle, mentioning the 'trough of disillusionment' faced by previous tech waves.](https://ss.rapidrecap.app/screens/jA8ZQfq_Hzs/00-01-18.jpg) ![Screenshot at 02:07: Apoorv Agrawal asking the panel to describe the current reality of AI adoption in the enterprise.](https://ss.rapidrecap.app/screens/jA8ZQfq_Hzs/00-02-07.jpg) ![Screenshot at 04:54: Ali Ghodsi detailing examples where AI value is clear, such as in finance by automating SEC report analysis.](https://ss.rapidrecap.app/screens/jA8ZQfq_Hzs/00-04-54.jpg)