By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
The Twenty Minute VCWhy Neo-Clouds are Great Businesses & The State of Compute with Kush Bavaria, Co-Founder @ Ornn
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
Great conversations run long. Their best ideas don't have to.
uListen is free on iPhone, iPad, Android, Chrome, and Claude.
See howCHAPTERS
Jump to publisher timestampsChapter summaries
Kush’s worldview on AI and why compute markets matter
Kush opens with a striking view of AI’s future impact on humans, while Harry sets the frame for the episode: understanding the cost of intelligence, compute, and GPU pricing. The conversation positions Ornn as a company building price indices and futures for GPU compute.
- •Kush suggests humans may become a “biological vessel” for AI and questions whether dying will remain “real” in the future
- •Harry introduces Kush as co-founder/CEO of Ornn and summarizes Ornn’s focus on compute markets, indices, and futures
- •Kush claims infrastructure investors tend to win because they keep “collecting rent”
- •Harry previews a deep dive into chips, compute costs, and market structure
Early startup story: building a sports betting app at MIT
Kush recounts building a sports betting-style trading app in college after being rejected for an internship. The project worked within MIT but didn’t scale due to weak go-to-market execution.
- •Kush and Wayne wanted to intern at a company but weren’t accepted, motivating them to build their own product
- •They built an app for buying/selling sports bets, including player-stat wagering
- •They got adoption through MIT fraternities but didn’t expand beyond campus
- •Kush attributes the limited outcome to being strong at coding but inexperienced in marketing, legal, and distribution
Learning from Travis Kalanick: ambition, aggression, and pricing mistakes
Kush explains why he studies Travis Kalanick’s approach, mainly to push himself to be more aggressive and ambitious. He gives an example of underpricing early commercial agreements for Ornn’s data.
- •Kush calls Travis Kalanick a major entrepreneurial influence and says he asks “what would Travis do?” in tough situations
- •He says he’s naturally “nice” and uses Travis as a counterbalance to push harder
- •Kush describes selling early data products too cheaply to hedge funds and trading firms
- •He realized later that sophisticated buyers would have paid far more, and that immediate agreement was a signal he underpriced
Running marathons while starting a company
Kush shares a story of running a marathon with little training during Ornn’s earliest days. He uses it to illustrate commitment and the intensity of startup life squeezing out everything else.
- •Kush describes signing up for a marathon before starting the company, then losing training time once the startup began
- •He emphasizes startup time demands and how training can consume hours per day
- •He still ran the marathon despite being unprepared and describes it as painful but completed
- •Harry and Kush compare marathon times and connect endurance to founder mentality
Why infrastructure investors “collect rent”: landlords, tenants, and AI demand risk
Kush lays out a landlord-tenant analogy for GPU infrastructure, arguing infrastructure owners keep earning as long as there are renters. He and Harry discuss how enterprise ROI concerns and demand shocks can still pressure rents and GPU-hour pricing.
- •Kush argues infrastructure investors are like landlords; GPU owners/renters are the tenants
- •He says infrastructure investors win as long as demand exists for renting compute
- •Harry raises the concern that enterprise ROI doubts could reduce demand and ripple through providers to infrastructure owners
- •Kush agrees pricing is ultimately supply-and-demand: weaker demand leads to lower GPU-hour prices
GPU supply constraints and the path of pricing: lead times, debt worries, and forward curves
The discussion turns to current GPU undersupply, procurement lead times, and whether debt-funded buildouts are risky. Kush claims demand for AI products is real, and he argues market expectations on future GPU-hour prices have shifted upward.
- •Kush says supply is very undersupplied and mentions long wait times to buy capacity from major providers
- •He outlines chip procurement lead times and additional installation time for smaller-scale buildouts
- •Harry asks about the debt cycle funding infrastructure; Kush says demand is real because products are useful to users
- •Kush explains earlier markets expected future prices to be lower, but now market behavior implies future GPU-hour prices could be higher; he contrasts past “free call options” with today’s reluctance to offer them
Neo-cloud skepticism vs the bull case: CoreWeave/Nebius, leverage, and tenant credit risk
Harry asks why neo-cloud providers get criticized and why their debt costs can be high. Kush argues the market fears a bubble, but points to rising GPU-hour prices and says he’d be long certain providers—while acknowledging credit risk depends on whether AI-company tenants can pay.
- •Harry asks why people criticize neo-clouds like CoreWeave and Nebius and why debt costs are elevated
- •Kush attributes skepticism to leverage and doubts about long-term chip residual value and AI demand
- •He says a major risk is tenant credit: whether AI companies renting compute can keep paying
- •Kush states he would go long on CoreWeave and Nebius if he could
Bear case for AI buildout and why hedging compute risk matters
Kush presents the bear case: enterprises may not get enough value from AI, leading to overbought compute and potential defaults. He frames Ornn’s mission as helping buyers hedge the uncertainty of future compute prices and demand.
- •Kush says the bear case is AI demand not continuing, especially if enterprises fail to see business impact
- •He claims many large companies haven’t yet implemented AI in a useful way
- •He describes a downside scenario where renters can’t pay, lenders end up with devalued chips, and defaults cascade
- •Kush connects this uncertainty to the need for hedging compute price risk over forward purchase windows
Tooling stack and AI productivity: GrokBot, Instinct, Cursor, Factory vs Devin/Cognition
A detour into day-to-day AI tooling reveals how Kush’s team uses assistants and coding tools. Kush highlights cost differences and why his team prefers certain “harness” products for flexibility and economics.
- •Kush says he uses GrokBot for daily tasks as a replacement for a traditional assistant
- •Harry says he uses Instinct daily; Kush notes his co-founder uses it and considers switching
- •Kush describes using Cursor and Factory for coding, emphasizing bring-your-own-key and pricing advantages
- •He contrasts Factory with Devin/Cognition, saying Cognition was expensive and Factory’s underlying harness felt better for their usage
Chip depreciation, older GPUs, and how inference providers may specialize
Kush argues older GPUs aren’t becoming obsolete as quickly as expected, pointing to sustained rental demand for prior generations. He predicts inference will segment into fast, cheap/slow, and specialized categories, with many providers surviving by verticalizing.
- •Kush says market expectations about rapid depreciation shifted after seeing long rental periods for older chips
- •He claims older chips can remain useful for smaller-model inference even if not used for frontier training
- •He defines “fast inference,” “cheapest but slowest,” and “specialized inference” (e.g., video/image generation)
- •Kush predicts a mix of horizontal base inference plus many specialized providers; he expects verticalization even if consolidation happens
AI investing dynamics: FOMO rounds, overheated categories, and “underheated” infrastructure
Kush critiques ultra-fast fundraising as hype-driven and rooted in fear of missing the next breakout company. He identifies categories he believes are crowded and highlights physical/energy infrastructure as a place he finds more compelling.
- •Kush says rounds done in hours are often fear-based rather than diligence-based
- •He attributes momentum investing to investors feeling they missed prior winners and chasing the next one
- •He says some application categories feel “done,” citing examples of strong incumbents in those segments
- •He argues physical/energy and data-center infrastructure is attractive, noting a lack of startups building key components and tools for infrastructure buildout
New York vs Silicon Valley and founding motivation: loyalty, friends, and long-term intent
Kush explains why he prefers building in New York, describing a culture he sees as more loyal and less transactional. He also shares a personal motivation for founding: working daily with close friends and avoiding corporate life.
- •Kush argues New York is better for him, saying Silicon Valley can feel “coin operated”
- •Harry adds that Silicon Valley can produce groupthink and overly transactional behavior
- •Kush says founding let him work with his best friends from school every day
- •He recounts disliking prior corporate experiences and wanting a different path
Why compute needs a marketplace: fragmentation, fungibility, indices, and trust
Kush details the core thesis behind Ornn: many suppliers and buyers of similar GPU products create a market structure that benefits from standard pricing and tradable contracts. He explains why picking a standardized reference compute product is key, and why trust matters for index/ratings businesses.
- •Kush says compute pricing is supply-and-demand with many suppliers and many buyers seeking similar GPU goods with varying parameters
- •He claims neo-clouds lack strong moats beyond owning infrastructure, motivating them to expand into adjacent offerings
- •He argues the market didn’t exist before because CPU supply was concentrated, while now there are many neo-cloud suppliers and more demanders
- •Kush explains how an index can be built by standardizing a specific reference product, similar to how commodity reference prices work
- •He says trading and index leadership depend heavily on trust and long-standing relationships, analogous to major index and ratings providers
Building Ornn’s exchange: getting data requires going “to the chip level”
Kush explains why Ornn couldn’t remain purely a software/data company: transaction data is too valuable for large providers to share. He describes how Ornn sources compute through partner capital structures and uses a physical market to generate robust index data.
- •Kush says Ornn started with index products because compute needs futures/options and other derivatives
- •He explains that to build indices you need spot-market transaction data, requiring a physical exchange
- •Kush describes an “Airbnb-style” approach where others finance chips while Ornn controls/monetizes them via structures like sale-leasebacks
- •He says major providers won’t freely share transaction data, forcing Ornn to build deeper infrastructure to access reliable, large-scale data
Who trades compute futures and how indices standardize clusters and reduce risk
Kush outlines the natural participants in compute futures: entities exposed to compute price and credit risk. He also argues that indices encourage standardization and fungibility, improving lender confidence and making it easier to replace defaulting customers.
- •Kush says early futures traders will be those exposed to compute price swings and tenant credit risk
- •He describes lender exposure when startups buying compute may not generate enough revenue, creating default risk and forced resale risk
- •He compares compute hedging to farmers selling futures to lock in uncertain future prices
- •Kush argues indices drive standardization of clusters and increase fungibility, enabling easier tenant replacement when a customer backs out
Company-building: focus, fundraising philosophy, choosing investors, and naming Ornn
Kush discusses the temptation to expand into everything, while explaining why he raised capital despite feeling cash-flexible: to assemble the right long-term partners. He explains how he evaluates investors and shares the practical origin of the company name.
- •Kush says retaining focus is hard because there are many potential product directions, including deeper physical infrastructure
- •He says they raised to get the right partners for what he calls a long-term, “family business” approach
- •Kush explains choosing investors based on how much they care about founders as people and how tied their success is to the company’s success
- •He recounts fundraising dynamics across boutique firms and larger firms and why the partner matters more than brand
- •Kush says the name Ornn came from League of Legends and selecting a cheap four-letter domain
Founder mental toll, lifestyle tradeoffs, and building a high-performance culture
Kush describes the emotional volatility and stress of founding, and the personal tradeoffs he’s made to stay focused. He then explains how Ornn hires for resilience and communication, emphasizing athletes and social-fit testing in uncomfortable settings.
- •Kush says founders underestimate the mental stress and daily emotional volatility of startup life
- •He jokes about being “single till” later milestones and discusses distraction vs emotional regulation
- •He says he’s motivated more by working with friends than by money and describes living simply to stay “hungry”
- •Kush argues culture becomes critical as time spent with teammates exceeds time with family and friends
- •He says Ornn often hires people with sports backgrounds for communication and competitiveness, and uses dinners/uncomfortable situations to test team fit and ability to handle uncertainty
Quickfire: board pick, compute as a bottleneck, model-harness conviction, and hopes for humanity
In rapid-fire questions, Kush names who he’d add to the board and what he’d ask them, and reiterates his belief that compute access is a key constraint. He shares a major belief update about coding harnesses, picks a single company he’d invest in, and closes with optimism about AI enabling longevity and broader human flourishing.
- •Kush says he would add Travis Kalanick to his board and would ask how he handles emotional volatility
- •Asked to choose between major AI labs, Kush picks the one he believes has more compute, arguing money can’t accelerate chip fabrication lines
- •Kush says he changed his mind on coding harnesses, now believing they matter for quickly switching models and matching the right model to tasks
- •He says if he could invest all his money in one company, he’d choose Factory based on product quality
- •Kush says he’s excited about AI’s potential to improve longevity and believes expanding compute access and hedging risk can help direct compute toward positive outcomes
Publisher timestamps
- 1:06Meeting Kush and His Early Startup Journey
- 2:56Learning From Travis Kalanick
- 4:45Running Marathons While Building a Startup
- 6:04Why AI Infrastructure Investors Will Win
- 8:44Is There Enough GPU Supply?
- 10:13The Future of GPU Pricing
- 11:27CoreWeave, Neoclouds and the AI Bubble
- 12:44The Bear Case for AI
- 14:44Hedging GPU Costs and Compute Risk
- 17:37Chip Depreciation and the Future of Inference
- 20:47AI Investing, FOMO and Overheated Markets
- 24:11Why New York Beats Silicon Valley
- 26:01Why Kush Became a Founder
- 27:57Why Compute Needs a Marketplace
- 30:03How GPU Price Indices and Futures Work
- 33:08Building a Compute Exchange
- 35:31Who Will Trade Compute Futures?
- 38:32Compute as a Fungible Commodity
- 40:53How Big Could the Compute Market Become?
- 42:12Training vs Inference Demand
- 44:35AI Adoption and the Future of Humanity
- 47:08Credit Ratings for AI Infrastructure
- 49:33Fundraising and Choosing Investors
- 50:09How the Company Got Its Name
- 56:25The Mental Toll of Building a Startup
- 59:29The Founder Lifestyle
- 1:01:07Building a High-Performance Company Culture
- 1:03:58Hiring and Testing for Resilience
- 1:05:52Quickfire Questions
- 1:11:05AI in 2030 and the Future of Humanity