Dillon Patal

Founder and CEO of SemiAnalysis

Mentions

Appeared in

Discussed in

Key positions and views

The primary bottleneck for scaling AI is not chip supply but physical infrastructure, including power generation capacity, grid stability, and the slow pace of data center construction.

China is a formidable and underestimated competitor in the AI race, demonstrating leadership in open-source models and effectively using state subsidies and sanctions evasion to advance its technological ambitions.

Effective hardware-software co-design is the most critical and rarest competency in the AI hardware space, explaining NVIDIA's continued dominance and the struggles of competitors like AMD and Microsoft's internal teams.

The AI market is undergoing significant shifts, with challengers like Anthropic overtaking incumbents like OpenAI in key metrics (API revenue) and specialized cloud providers outperforming hyperscalers in performance.

Major technology companies are facing severe, underappreciated execution risks in their AI infrastructure build-outs, leading to significant delays and revenue impacts.

Podcast consensus on Patal

Points of consensus

▶Patal presents as established fact that the growth of AI is severely constrained by physical infrastructure, including multi-year backlogs for power generation equipment like gas turbines and slow data center construction speeds.Feb 2026

▶He asserts that China has successfully used state subsidies to achieve or approach global market dominance in key technology and energy sectors, specifically citing solar panels and electric vehicles.Feb 2026

▶Patal's analysis concludes that the first generation of AI hardware startups (e.g., Cerebras, Groq) failed due to a fundamental architectural miscalculation, betting on large on-chip memory which became obsolete as models grew.Feb 2026

▶He identifies a pattern of significant execution challenges in AI hardware and infrastructure projects at major tech companies, citing Amazon's Trainium chip delays, Microsoft's slow construction, and Meta's power grid issues.Feb 2026

Points of debate

▶Patal highlights the highly contested leadership in the AI model market, noting a recent period where Chinese labs dominated open-source models and claiming that Anthropic has surpassed OpenAI in API revenue.Feb 2026

▶He describes an ongoing struggle for AI cloud dominance where specialized 'neoclouds' offer superior performance but are also becoming less accessible to startups as they prioritize large, long-term contracts.Feb 2026

▶The effectiveness of US sanctions against China's tech sector is presented as a point of contention, with Patal claiming that companies like Huawei are actively and successfully circumventing them using shell companies.Feb 2026

▶Patal points to strategic uncertainty and competition in the cloud market, referencing rumors from The Information that CoreWeave attempted to acquire Fireworks AI to move up the software stack.

Key themes

▶The Physical Constraints of AI GrowthFeb 2026

Patal consistently argues that the expansion of AI is bottlenecked not just by semiconductor availability, but more critically by physical world limitations. This includes multi-year backlogs for power generation equipment, the slow pace of traditional data center construction, and local power grid capacity issues.

Investors focused solely on chip manufacturers may be overlooking significant risks and opportunities in the broader infrastructure and energy sectors, which Patal identifies as the true long-term gatekeepers of AI scaling.

▶US-China Geopolitical Tech CompetitionFeb 2026

A central theme is the strategic rivalry between the US and China in critical technologies. Patal details China's use of state subsidies to dominate industries like solar and EVs, its temporary leadership in open-source AI models, and its active circumvention of US sanctions on semiconductors.

Patal's analysis suggests that US policy measures like sanctions are porous and that China's state-directed industrial strategy remains a potent force, challenging the long-term technological supremacy of American firms.

▶The AI Hardware Arms Race and Architectural BetsFeb 2026

Patal provides a critical analysis of the AI hardware landscape, detailing the failure of first-wave startups due to flawed architectural assumptions. He also critiques the internal hardware efforts of giants like Microsoft and AMD, attributing their struggles to a lack of deep hardware-software co-design expertise, a domain where NVIDIA still excels.

This theme indicates that simply manufacturing silicon is insufficient for success in AI; deep integration with software and a correct architectural vision for future model scaling are the key differentiators.

▶Shifting Dynamics in the AI Cloud and Model MarketFeb 2026

Patal portrays a highly fluid market where leadership is not guaranteed. He claims Anthropic has overtaken OpenAI in API revenue, specialized 'neoclouds' are outperforming hyperscalers on key metrics, and even giants like Amazon Web Services can miss revenue targets due to hardware integration failures.

The AI market is far from mature, and incumbent advantages are not absolute. This creates opportunities for specialized providers and suggests that market share can shift rapidly based on technical execution and strategic focus.

Source episodes

Sentiment over time

Not enough data for timeline

Changes over time

Pre-2024

Patal notes that the first wave of AI hardware startups (Cerebras, Groq, etc.) failed because their architectural bet on large on-chip memory was rendered obsolete by the rapid growth in AI model size.

Recent Past

He identifies a 6-9 month period, prior to OpenAI's latest release, where Chinese labs dominated the best-performing open-source AI models, following the era when Meta's Llama 3.1 was considered the leader.

Present

Patal describes a market where major tech firms are hitting physical walls in their AI build-out, citing Meta's grid issues in Ohio, Amazon's Trainium rack integration problems, and Microsoft's slow construction pace.

Present

He claims a significant market shift has occurred, with Anthropic surpassing OpenAI in API revenue, largely driven by its lower-cost, non-reasoning models.

Present

Patal details OpenAI's current strategy to counter infrastructure bottlenecks by diversifying partners globally (Oracle, CoreWeave, G42) and to regain a competitive edge by releasing open-source models with highly optimized custom kernels.

Suggested prompts

How do Patal's claims about physical infrastructure constraints (power, construction) challenge the narrative of AI growth being limited only by chip supply? &nearr;Based on Patal's analysis, what are the key vulnerabilities in the US strategy to contain China's semiconductor ambitions? &nearr;What are the primary factors, according to Patal, that differentiate successful AI hardware strategies from unsuccessful ones, particularly regarding hardware-software co-design? &nearr;How does Patal characterize the evolving competitive landscape between incumbent hyperscalers (AWS, Azure) and specialized 'neoclouds' (CoreWeave) in the AI era? &nearr;

Key concepts

AI Hardware & Semiconductors 6 ep AI Model Competition 6 ep AI Infrastructure & Data Centers 5 ep US-China Tech Competition 5 ep Energy & Power Grid 5 ep Corporate Strategy (AI) 5 ep Cloud Computing Market 4 ep Business & Revenue (AI) 3 ep

Notable quotes

“Anthropic has surpassed OpenAI in API revenue.”

DILLON PATAL · No Priors Ep. 127 | With SemiAnalysis Founder and CEO Dylan Patel

“Microsoft is not building the 'Stargate' supercomputer for OpenAI because its traditional data center construction methods are too slow.”

DILLON PATAL · No Priors Ep. 127 | With SemiAnalysis Founder and CEO Dylan Patel

“Huawei is circumventing US sanctions by acquiring wafers from TSMC in Taiwan through the use of shell companies.”

DILLON PATAL · No Priors Ep. 127 | With SemiAnalysis Founder and CEO Dylan Patel

“There is currently a multi-year backlog, potentially as long as four years, for General Electric's gas turbines used for power generation.”

DILLON PATAL · No Priors Ep. 127 | With SemiAnalysis Founder and CEO Dylan Patel

Report last updated: Apr 21, 2026

Get started free

Back to Entities Intelligence Report