Dylan Patel: Tech companies prioritize long-term capex for future infrastructure, Anthropic’s scaling challenges contrast with OpenAI’s aggressive strategy, and GPU depreciation cycles may exceed five years | Dwarkesh

Key takeaways

• Tech companies’ capital expenditure is often allocated for long-term projects rather than immediate compute capacity.
• A significant portion of Google’s capex is dedicated to future infrastructure projects like turbine deposits and data center construction.
• Anthropic needs to significantly scale its inference capacity to meet its revenue growth projections.
• Anthropic’s conservative approach to compute acquisition contrasts with OpenAI’s aggressive strategy, impacting their market positions.
• AI labs are entering long-term deals at higher prices, indicating a shift in market dynamics.
• The depreciation cycle of GPUs may be longer than previously assumed, affecting financial models.
• GPU pricing is influenced by performance improvements and real-world utility.
• The release of new chips will likely decrease the value of existing GPUs due to performance advancements.
• The adoption potential for GPT-5.4 could exceed $100 billion, but competition and adoption lag are factors.
• Dario’s conservative approach to compute investment seems inconsistent given the potential revenue from advanced AI models.
• AI labs are paying premiums for compute resources, reflecting increased demand and competitive pressure.
• The semiconductor market is experiencing shifts due to technological advancements and strategic investments.
• Understanding capex allocations can provide insights into future tech infrastructure developments.
• The competitive landscape between AI companies is shaped by their compute acquisition strategies.
• GPU depreciation assumptions are crucial for tech investment and financial planning.

Guest intro

Dylan Patel is the Founder and CEO of SemiAnalysis, a leading semiconductor and AI research and consulting firm with offices across the US, Japan, Taiwan, and Singapore. He began consulting on semiconductor architecture in 2017 before going full-time in 2020, and subsequently launched SemiAnalysis’s Substack newsletter, which has grown to approximately 50,000 subscribers and become the second-largest tech newsletter in the world. His deep expertise in the semiconductor supply chain, from chip design to fab operations to AI infrastructure economics, has made him one of the most cited analysts advising hyperscalers, AI labs, and semiconductor manufacturers on industry bottlenecks and strategy.

Capital expenditure strategies in big tech

• The capital expenditure (capex) from big tech companies is not all for immediate compute capacity; much of it is for future projects.

  — Dylan Patel

• Google’s capex includes significant investments in turbine deposits and data center construction for future years.

• When you look at hey Google’s got a $180,000,000,000 actually a big chunk of that is spent on turbine deposits for ’28 ’29 a chunk of that is spent on data center construction for ’27.

  — Dylan Patel

• Understanding the timeline of tech companies’ investments is crucial for forecasting future compute capacity.
• Tech companies strategically plan their capex to support long-term infrastructure rather than immediate needs.
• These investments indicate a focus on sustainable growth and future-proofing their operations.
• The allocation of capex for future projects reflects the strategic priorities of major tech companies.
• Insight into capex strategies helps stakeholders anticipate future industry trends and infrastructure developments.

Scaling challenges for AI companies

• Anthropic needs to significantly scale its inference capacity to meet projected revenue growth.

• Anthropic needs to get to well above five gigawatts by the end of this year and it’s gonna be really tough for them to get there but it’s possible.

  — Dylan Patel

• The scaling challenges faced by AI companies highlight the competitive pressures in the industry.
• Achieving the necessary compute capacity is critical for AI companies to meet their growth targets.
• The ability to scale effectively can impact an AI company’s financial stability and market positioning.
• Strategic planning and investment in infrastructure are essential for overcoming scaling challenges.
• The competitive landscape in AI is influenced by each company’s ability to scale its compute resources.
• Scaling challenges are a key factor in determining the success of AI companies in the market.

Conservative vs. aggressive compute acquisition strategies

• Anthropic’s conservative approach to acquiring compute contrasts with OpenAI’s aggressive strategy.

• Anthropic was a lot more conservative… we’ll sign contracts but we’ll be principled and we’ll purposely undershoot what we think we can possibly do and be conservative because we don’t wanna potentially go bankrupt.

  — Dylan Patel

• These differing strategies impact the financial stability and market positioning of AI companies.
• A conservative approach may reduce risk but could limit growth opportunities.
• An aggressive strategy could lead to rapid growth but also increase financial risk.
• The choice of strategy reflects each company’s risk tolerance and market objectives.
• Understanding these strategies is crucial for analyzing the competitive dynamics in the AI industry.
• The strategic differences between AI companies influence their long-term success and market share.

Market dynamics in AI compute pricing

• AI labs are signing long-term deals at significantly higher prices, indicating a shift in market dynamics.

• I’ve seen deals where certain AI labs have signed at as high as $2.40 for two to three years for H100s.

  — Dylan Patel

• The increased demand for AI compute resources is driving up prices and affecting market dynamics.
• Long-term deals reflect the strategic importance of securing compute resources in a competitive market.
• The pricing dynamics in the AI compute market are influenced by supply and demand factors.
• Understanding these dynamics is crucial for stakeholders navigating the AI compute landscape.
• The shift in pricing indicates a growing recognition of the value of AI compute resources.
• Market dynamics are shaped by the strategic decisions of AI companies and their investment in infrastructure.

GPU depreciation and financial implications

• The depreciation cycle of GPUs may be longer than previously thought, potentially exceeding five years.

• Michael Burry was saying it’s you know three years or or less right it’s like sort of his argument… but in fact you’re pointing at like maybe the depreciation cycle is even longer than five years.

  — Dylan Patel

• This insight challenges existing assumptions about GPU depreciation, affecting financial models.
• Longer depreciation cycles could impact the profitability of cloud computing and tech investments.
• Understanding GPU depreciation is crucial for financial planning and investment strategies.
• The implications of GPU depreciation extend to cost management and resource allocation.
• Financial models need to account for the potential extension of GPU depreciation cycles.
• The depreciation cycle is a critical factor in the economic analysis of tech investments.

Factors influencing GPU pricing

• The pricing of GPUs is influenced by performance improvements and real-world utility.

• The price of a gpu would continue to fall… what is the value I can derive out of this chip today.

  — Dylan Patel

• As new chips are released, the value of existing GPUs will decrease significantly.

• The hopper is only worth 70¢ an hour… the price of a gpu would continue to fall.

  — Dylan Patel

• Understanding the dynamics of GPU pricing is crucial for stakeholders in the tech industry.
• Performance advancements and market demand play a significant role in determining GPU prices.
• The release cycles of new GPU technologies impact the valuation of existing resources.
• Strategic planning in tech investments requires consideration of GPU pricing trends.

Future market potential of AI models

• The adoption potential for GPT-5.4 could exceed $100 billion, but there will be an adoption lag and competition.

• The value of an h one hundred is now predicated on the value that g p d five point four can get out of it instead of the value that g p d four can get out of it.

  — Dylan Patel

• The competitive landscape and technological advancements influence the market potential of AI models.
• Understanding these factors is crucial for stakeholders assessing the future of AI technologies.
• The adoption lag and competition are key considerations in evaluating the market potential of AI models.
• Strategic planning and investment decisions are influenced by the projected market potential of AI technologies.
• The future success of AI models depends on their ability to navigate competitive pressures and adoption challenges.
• The market potential of AI models is a critical factor in the economic analysis of AI investments.

Strategic inconsistencies in AI investment

• Dario’s conservative approach to compute investment seems inconsistent given the potential revenue from advanced AI models.

• The point i was trying to make is that given what dario seems to be saying… it just does not make sense why he keeps making these statements about being more conservative on computer.

  — Dylan Patel

• This insight highlights a critical inconsistency in strategic decision-making within a major AI company.
• The inconsistency could impact the company’s future growth and market positioning.
• Understanding these strategic inconsistencies is crucial for stakeholders assessing the company’s potential.
• The revenue potential of advanced AI models suggests a need for more aggressive investment strategies.
• Strategic alignment is essential for maximizing the growth potential of AI companies.
• The inconsistency in investment strategies reflects broader challenges in the AI industry.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.