▶Apoorv consistently argues for OpenAI's market dominance using multiple data points, including its 700 million weekly active users, a user base he estimates is more than 10 times larger than all competitors combined, and strong 'smile curve' user retention metrics similar to top-tier social apps.Apr 2026
▶He repeatedly highlights the superhuman performance of the AI company Expo, whose agents achieved the #1 ranking for hackers in the U.S. and then globally on the HackerOne platform, as definitive proof of advanced AI capabilities.Apr 2026
▶Apoorv's investment thesis, articulated through Altimeter's strategy, is consistently focused on making large, concentrated bets on companies believed to be on a 'power law' trajectory, citing OpenAI as the firm's largest investment ever and comparing it to a past successful bet on Snowflake.Apr 2026
▶He presents a coherent theory of an 'inverted' value stack in the AI supercycle, where the application/model layer (like OpenAI) captures significantly more revenue than the infrastructure layer, a direct contrast to the value distribution seen in the previous cloud supercycle.
▶Apoorv posits that ChatGPT has weaker network effects than platforms like Instagram or X because a user's decision to join is not strongly influenced by their friends' presence, a view that contrasts with the common belief that data network effects are a primary driver of AI platform dominance.Apr 2026
▶He asserts that 'speed is the only moat' for OpenAI, emphasizing rapid feature velocity as its key defensibility, which de-emphasizes other widely cited moats such as proprietary data, massive capital investment, and unique talent concentration.
▶Apoorv claims that code generated by AI models is inherently more vulnerable because it is trained on open-source codebases containing existing flaws, presenting a specific security risk that challenges the overwhelmingly positive narrative of AI's impact on software development productivity.Apr 2026
▶He argues that standard AI benchmarks have become saturated and are no longer definitive measures of performance, advocating instead for proprietary, real-world tests like Expo's cyber exploit detection as the true measure of a model's superiority.
Not enough data for timeline
Sign up free to see the full intelligence report
Get started free