Skip to content

By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.

Sourcery with Molly O'SheaSourcery with Molly O'Shea

Dylan Patel on Anthropic's IPO & AI Markets

I sat down with Dylan Patel, Founder & CEO of SemiAnalysis, to discuss AI markets, and how he's building a research and investment empire with the acquisition of Citrini Research and the launch of a $500 million venture fund.

Molly O'SheahostDylan Patelguest
Nov 17, 202650mWatch on YouTube

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 how

CHAPTERS

Jump to publisher timestamps

Chapter summaries

  1. SemiAnalysis at scale: global team, customers, and a data-center tracking edge

    Molly opens by framing SemiAnalysis as a fast-growing “empire,” and Dylan describes the company’s footprint and what they cover. He emphasizes their global headcount, multi-office presence, and how closely they track data centers and serve hyperscalers, semiconductor companies, and AI labs.

    • •Dylan describes a global team across many countries and multiple offices
    • •He says the firm tracks every data center in the world and positions their data-center knowledge as best-in-class
    • •He identifies their biggest customers as hyperscalers, semiconductor companies, and AI labs
    • •Molly sets the stage for a wide-ranging conversation across regulation, models, and company-building
  2. Setting the scene: “Sourcery,” wizards, and the Hogwarts-library office vibe

    The conversation detours into the show’s name and a playful debate about whether it’s about “source code” or magic. Molly and Dylan riff on LA “witchy” culture and the office’s hidden library aesthetic.

    • •Dylan questions whether the show name implies wizard theming or source code
    • •Molly clarifies the name came from VC deal sourcing, not programming
    • •They joke about LA’s crystals, reiki, and energy-healing culture
    • •Molly calls the office setup a “Hogwarts library,” with a secret library space
  3. Space data centers and the near-term reality: permitting bottlenecks on Earth

    Molly asks whether putting data centers in space could relieve pressure from domestic regulation. Dylan says space may matter later in the decade, but it won’t be meaningful in the next few years; meanwhile, the biggest constraint he sees is permitting delays that push capacity timelines out.

    • •Dylan frames space data centers as plausible later but not a near-term capacity fix
    • •He discusses domestic regulatory pressure and why the next few years matter most
    • •He describes permitting as the key bottleneck and notes capacity pushed from 2027 into 2028
    • •He explains multiple permits can be used to slow or stop projects
  4. Who’s really against data centers: awareness gaps, politics, and narrative drivers

    Dylan argues many jurisdictions banning data centers don’t have meaningful data-center buildouts, while places with major projects often haven’t banned them. He cites survey results showing many residents don’t even know there are data centers in their county, and he describes how misinformation and broader anti-AI sentiment can amplify backlash.

    • •Dylan says most bans happen in places without meaningful existing or planned capacity
    • •He cites a poll showing many residents in counties with large data centers didn’t know they existed
    • •He links backlash to a mix of reasonable concerns, misinformation, and alleged astroturfing
    • •Molly suggests data centers become a physical target for broader anti-AI sentiment; Dylan agrees
  5. Community impact vs. complaints: taxes, noise, water—and the “3 In-N-Outs” comparison

    Molly brings up local benefits like public funding; Dylan adds examples where tax revenue led to payouts for teachers. He also explains what can go wrong (noise, inefficient water usage, poor community return) and makes a provocative comparison about water usage, emphasizing embedded supply-chain accounting.

    • •Molly cites a public example of communities benefiting from data-center revenue
    • •Dylan mentions a Louisiana county where teachers received large payouts tied to tax rules
    • •He outlines legitimate concerns: noise near residences, inefficient cooling water use, limited long-term local jobs
    • •Dylan claims Elon Musk’s data centers use less water than “three In-N-Outs,” with an embedded-water caveat
  6. AI winners rotate—but markets smooth it out: model momentum across major labs

    Molly pivots to AI company dynamics and asks how to interpret shifting momentum among leading labs. Dylan says performance and buzz may swing quarter to quarter, but market perception doesn’t whipsaw as much; he gives examples of periods when different labs performed better and notes release delays as a factor.

    • •Molly frames a rotating “top dog” dynamic among major AI players
    • •Dylan argues markets don’t revalue leaders as quickly as online discourse does
    • •He describes quarters where different labs performed better and points to model-release timing
    • •He suggests people over-index on single data points when tracking AI progress
  7. Why Meta’s Muse (and other apps) can beat better models: product UX vs. safety friction

    Molly asks whether attention has moved from model releases to applications and points to Meta’s Muse. Dylan argues some products win despite weaker underlying models because they’re better packaged and less constrained, criticizing OpenAI’s product experience and describing how safety layers can degrade usability.

    • •Molly highlights applications gaining momentum and asks how to evaluate them
    • •Dylan says some apps are better products even with weaker models
    • •He claims OpenAI’s safety mechanism can be overly restrictive and harm product quality
    • •He criticizes OpenAI’s chat app complexity and praises Muse’s UI from his beta experience
  8. How SemiAnalysis spends on AI: shifting from tools to persistent agents via a better harness

    Dylan explains how his organization’s AI spending has changed over time and what they’ve standardized on. He describes a major shift toward persistent agents and says the purchasing decision is influenced by product “harness” quality rather than only the underlying model provider.

    • •Dylan describes SemiAnalysis’ AI spend as fluctuating throughout the year
    • •He says a large share recently shifted to persistent agents
    • •He says much of that agent usage flows through Perplexity Computer as a better interface/harness
    • •He contrasts stickiness in products vs. developer tools that can be swapped more easily
  9. Is there real ROI in AI startups? Growth, margins, and who captures value

    Molly asks about ROI given the level of spend required to scale. Dylan says most startups fail, some succeed dramatically, and he points to examples of companies growing quickly while also questioning valuation multiples and whether some businesses are thin pass-throughs with volatile margins.

    • •Dylan says failure rates are high but some startups will be extremely successful
    • •He cites companies he views as growing quickly and having real traction
    • •He questions whether some valuations (including inference-related businesses) are justified
    • •He notes margin volatility and worries some companies may be pass-through businesses
  10. Inside Dylan’s $500M solo GP fund: why infrastructure and hardware, not apps

    Molly transitions into Dylan’s venture fund and asks why expand beyond research. Dylan shares high-level details: a $500M solo GP fund focused on seed and Series A, centered on infrastructure and hardware-related areas, and he explains he avoids pure app investing because he doesn’t believe he can add value there.

    • •Dylan says he raised a $500 million solo GP fund focused on seed and Series A
    • •He describes the focus as infrastructure/hardware across areas like silicon, optics, networking, and manufacturing
    • •He says he won’t use his research platform to promote portfolio companies
    • •He argues he’s not suited to app investing because he doesn’t understand scaling consumer apps
  11. Sourcing, secrecy, and “most pitches are boring and wrong”

    Molly probes for details like check size and sourcing strategy; Dylan declines specifics to preserve flexibility. He describes relying on his team and network for deal flow while expressing skepticism that most pitches are compelling or accurate.

    • •Molly asks about check sizes; Dylan refuses to share specifics
    • •He says disclosing too much reduces his optionality
    • •He describes sourcing through his team and network rather than a single channel
    • •He bluntly claims most pitches are “boring and wrong”
  12. Acquiring Citrini: keeping independence while combining strengths (including AI skepticism)

    Molly asks about the Citrini acquisition and notes occasional worldview conflicts. Dylan says Citrini is allowed to remain bearish on AI and keep its voice; he frames the deal as a way to help operationalize and strengthen a distinct research approach known for unusual, high-effort fieldwork.

    • •Dylan says Citrini’s team has its own research style and can remain AI-bearish
    • •He emphasizes Citrini remains a separate company with editorial control staying with its leadership
    • •He contrasts SemiAnalysis’ supply-chain and technical tracking with Citrini’s experiential investigations
    • •He gives examples of Citrini doing unconventional research to answer real-world questions
  13. Why Gartner is a racket: monetizing attention without selling opinions

    Dylan describes how information services and media are being rewritten and criticizes legacy research models. He argues credibility requires not taking pay-for-opinion arrangements, even when criticizing major customers, and explains how he tries to separate consulting, investing, and public research integrity.

    • •Dylan argues legacy information firms can be low-quality while information value keeps rising
    • •He calls Gartner a “racket” and describes rankings he views as inconsistent with reality
    • •He says SemiAnalysis avoids ads and pay-for-opinion monetization to protect credibility
    • •He explains the importance of staying objective even when it upsets customers or conflicts with investments
  14. Hard lessons running SemiAnalysis: tough calls, compliance, and managing a global org

    Molly asks about the hardest moments in building the business. Dylan says running a company is difficult and involves lost sleep, tradeoffs, and painful decisions like firings and disputes; he also stresses the need for compliance controls given consulting and investing activities.

    • •Dylan says leadership involves constant tough decisions and personal tradeoffs
    • •He contrasts operating experience with investors who haven’t faced certain management realities
    • •He reiterates compliance matters and that consulting/investing/research must be firewalled appropriately
    • •He restates the company’s scale across countries and offices as operational complexity
  15. Knowing just enough Chinese—and closing with what’s next: chips, GDPX, and “seeing better in the dark”

    The conversation winds down with a language and culture tangent, then Molly asks what Dylan is excited about over the next year. Dylan highlights upcoming chip testing work and a benchmark initiative, then ends on a playful exchange about snack “chips,” color vision, and why he claims to see better in the dark.

    • •Dylan jokes about knowing enough Chinese to make people laugh and hints at learned phrases
    • •He says SemiAnalysis is launching more chip testing and model evaluation work, including GDPX
    • •He expresses excitement about progress unlocking advances in technology and hardware innovation
    • •They close with banter about Takis, color perception, and Dylan’s claim of better night vision

Publisher timestamps

  1. 0:00Dylan Patel, Founder & CEO at SemiAnalysis
  2. 0:37Inside the Hogwarts library
  3. 2:17Are data centers in space the answer?Read this chapter
  4. 6:42Who’s really against data centers?Read this chapter
  5. 8:47Elon's data centers vs. 3 In-N-OutsRead this chapter
  6. 12:04Who's winning: Anthropic, OpenAI or Google?Read this chapter
  7. 13:46Why Meta's Muse beats OpenAI on productRead this chapter
  8. 15:55Where SemiAnalysis spends $15M on AI
  9. 20:38Is there real ROI in AI startups?Read this chapter
  10. 22:30Inside Dylan's $500M fund
  11. 25:22Why Dylan won't invest in apps
  12. 28:43Most pitches are boring and wrong
  13. 30:30Buying Citrini, the AI bearsRead this chapter
  14. 33:20Sending an analyst to the Strait of HormuzRead this chapter from 33:21
  15. 34:03Why Gartner is a racketRead this chapter
  16. 41:07The hardest lessons of running SemiAnalysis
  17. 43:29Knowing just enough Chinese
  18. 46:32Why hardware is about to explode
  19. 48:03Why Dylan sees better in the darkRead this chapter

Get more out of YouTube videos.

Read along on your phone, search it beside the video, or ask your favorite AI app what was said.