Dylan Patel's SemiAnalysis Empire: $500M Fund, Citrini & NVIDIA?
Interview, Fireside Chat
Market & Supply Chain Dynamics
- Dylan Patel's firm (SemiAnalysis) operates 15 countries with 115 employees across offices in New York, Chicago, SF, Oregon, Singapore, and Korea.
- The firm tracks every data center globally and claims superior knowledge of data center infrastructure compared to competitors.
- A poll conducted with David Shore found that 49% of residents in counties hosting 100MW+ data centers are unaware a data center exists in their area.
- Only 40% of residents in counties expecting a 100MW+ data center in the next three years are aware of the upcoming construction.
- Hyperscalers, semiconductor companies, and AI labs constitute SemiAnalysis's largest customer base.
- The firm has delayed the deployment of multiple gigawatts of data center capacity from 2027 to 2028 due to permitting delays rather than labor or equipment shortages.
- Jurisdictions banning data centers often lack existing meaningful capacity, yet bans are driven by public perception and misinformation rather than local impact.
The Nvidia/Jensen Huang "Scandal" & Geopolitics
- The incident involves allegations regarding Jensen Huang's personal relationship with a woman who is a well-connected Alibaba executive and asset to the CCP.
- While Huang personally slept with a married woman (morally wrong), the primary business concern is the potential export of restricted GPUs to China via Southeast Asian hubs.
- Intelligence suggests hundreds of thousands of high-end GPUs were purchased and deployed in Southeast Asia primarily for Alibaba, likely violating US export restrictions if they reached China.
- The US Intelligence Community (IC) is described as "super hawkish," with leaks likely originating from the IC regarding these supply chain breaches.
- Economic interests initially favored relaxing export restrictions (e.g., allowing H200 chips), but the scandal has likely terminated any future approval for high-end chips like Blackwell to China.
- The "worst-case scenario" for Nvidia involves Jensen Huang being removed as CEO, which Patel argues would cripple the company due to a lack of clear succession among his 50 direct reports.
- AMD is positioned to gain market share by selling to companies renting GPUs to Chinese entities, as US restrictions specifically target Nvidia's direct sales.
- The US government is expected to increase scrutiny, potentially forcing Congress to question Huang and impose stricter diffusion rules on AI chip rentals.
- A proposed "diffusion rule" requiring specific security practices for companies renting chips to China was previously blocked due to complexity but may be revisited.
- The left wing is characterized as more hawkish on AI chip diffusion than the right wing, creating potential friction with a prospective Trump administration.
- Patel notes that while Huang may face endless Congressional questioning, the PR impact is mitigated because Nvidia focuses on B2B AI sales rather than consumer gaming.
Data Center Regulations & Public Perception
- Texas Governor Greg Abbott's administration is noted for being constructive regarding data centers, focusing on preventing electricity price spikes and ensuring water/air permit compliance.
- Major regulatory hurdles in Texas involve water permits, wastewater permits, and stormwater permits, which the Governor can use to halt construction.
- Union labor mandates present a potential friction point, as data center construction jobs (e.g., electricians, plumbers) currently earn high wages ($200k+) and operate outside traditional union structures.
- The narrative against data centers is driven by misinformation, including a New York Times book that allegedly overestimated water usage by 1,000x.
- Elon Musk's data centers use less water than three "In-N-Out" burger restaurants when accounting for the embedded water usage of the entire supply chain.
- Data centers in Louisiana have reportedly paid off teacher salaries in specific counties due to property tax revenue, though this is often contingent on improper tax break structures.
- Public opposition is often based on nebulous fears of noise and water usage rather than actual local impact, as most residents are unaware of nearby facilities.
- Space-based data centers are not a viable solution for capacity needs within the next 3–5 years, with meaningful gigawatt-level capacity unlikely before 2030.
AI Market, Products, & Applications
- The market valuation for AI companies remains stable despite fluctuations in quarterly performance (e.g., Anthropic crushing Q2, OpenAI crushing Q3).
- Meta's "Muse" and Perplexity's "Instinct" are viewed as superior products to OpenAI and Anthropic despite potentially using weaker underlying models.
- OpenAI's "Guardian" safety mechanisms are criticized for being overzealous, resulting in a subpar user experience compared to competitors that allow more freedom.
- SemiAnalysis has shifted 45% of its $15M annual AI spend toward persistent agents (e.g., Perplexity, Google) rather than traditional chat interfaces.
- Enterprise contracts with discount tiers (e.g., 70% spend commitment) create higher stickiness for providers like Anthropic, whereas developer tools like CodeLlama are easily switchable.
- The AI application cycle is shifting from raw model releases to product quality, with "better products" winning over "better models" in user retention.
SemiAnalysis Business Strategy & Venture Fund
- Patel founded a $500M early-stage venture fund (Seed/Series A) focusing on hardware infrastructure, robotics, silicon, optics, and manufacturing.
- The firm acquired Citrini, a research firm known for its bearish AI stance, to broaden its research capabilities while maintaining Citrini's independent editorial voice.
- SemiAnalysis operates on an "uncompromised objectivity" model, refusing paid partnerships or paid opinions to maintain credibility with its customer base.
- The firm frequently criticizes its own largest customers (e.g., calling out NVIDIA's anti-competitive practices, OpenAI's product flaws) to preserve trust.
- The venture fund is legally and operationally firewalled from the research division to prevent conflicts of interest, ensuring investment decisions do not influence research output.
- Patel's investment thesis prioritizes hardware and infrastructure expertise over applications, citing a lack of deep knowledge in scaling consumer apps as a limiting factor.
- The firm employs 115 people with backgrounds from ASML, Applied Materials, TSMC, Intel, Nvidia, Google, Microsoft, and Amazon.
- Future research plans include new benchmarks (GDPX) covering cyber, chip design, material science, bio, and quant, as well as testing of Huawei and Amazon chips.
Personal & Operational Details
- Patel lives in Los Angeles, jokingly attributing his residence to the local culture of "witchery" and crystals, distinguishing it from the "source code" of his business.
- He has never consumed drugs or alcohol and identifies as "completely holy."
- Patel speaks English, Gujarati, some Hindi, and Spanish; his Chinese is limited to humorous phrases and proverbs.
- He is partially colorblind (X-linked), affecting his ability to distinguish blue and purple, though he has better night vision due to rod cell sensitivity.
- Running the business involves significant personal sacrifice, including missed sleep, health issues, and difficult decisions regarding firing employees or negotiating with customers.
- The firm's mission combines four goals: making money, knowing everything, influencing key decisions, and spending time with loved ones.