Savlar4 sources healthy · latest episode 20 Oct 2026
Speaker

Dylan Patel

Guest on Dwarkesh Podcast. 257 claims across 1 episode, Aug 2026.

257claims, each timestamped
4 of 12signals on the brief
0verified changes of mind
87open predictions

Positions over time

Every pair of statements that was compared. Only a verified change counts as a change of mind; rewording, added nuance and different questions do not.

No two statements by this speaker have been compared yet.

Predictions

As stated. A window appears only where the speaker gave one.

ExpectsBySaidStatus
Global compute capital expenditure is expected to exceed $2 trillion by 2028. by 31 Dec 202825 Aug 2026pending
AI labs' annual spending will rise from tens of billions of dollars to hundreds of billions, with contracts and forecasts pointing to spending trillions of dollars per year toward the end of the decade. by 31 Dec 202925 Aug 2026pending
Anthropic began turning a profit in Q2, while OpenAI could potentially begin turning a profit at some point in Q3. no date given25 Aug 2026pending
Based on existing signed agreements, Anthropic and OpenAI may take as much as 40% to 50% of compute next year. 1 Jan 2027 – 31 Dec 202725 Aug 2026pending
The concentration of compute among major AI labs is not slowing or stopping and appears to be accelerating. no date given25 Aug 2026pending
The identity of the companies building compute for AI labs will change. no date given25 Aug 2026pending
The speaker expects SpaceX to lease a substantial amount of the compute it is building to Anthropic and OpenAI, likely because those labs can pay the highest marginal price. no date given25 Aug 2026pending
Anthropic and OpenAI together are expected to receive half of the world’s newly added compute by the end of next year. by 31 Dec 202725 Aug 2026pending
If the trend continues, Anthropic and OpenAI together will control most of the world’s usable FLOPs by late 2028. by 31 Dec 202825 Aug 2026pending
Anthropic and OpenAI will account for 45% of compute next year and half of the world’s incremental new compute by December 2027. 1 Jan 2027 – 31 Dec 202725 Aug 2026pending
Roughly 100 ASML tools will be supplied by 2030, which the speaker says remains the right estimate given the current pace of supply-chain expansion. by 31 Dec 203025 Aug 2026pending
The speaker does not expect the proposed supply-chain production expansion to happen in 2026, 2027, or 2028, attributing the delay to the world being capital-constrained. 1 Jan 2026 – 31 Dec 202825 Aug 2026pending
The speaker does not expect the AI labs to spend the very large amount of capital proposed in the preceding discussion. no date given25 Aug 2026pending
The AI labs are expected to generate hundreds of billions of dollars in revenue next year. 1 Jan 2027 – 31 Dec 202725 Aug 2026pending
Total capital expenditure next year is approximately $2 trillion. 1 Jan 2027 – 31 Dec 202725 Aug 2026pending
The wafer-fabrication-equipment supply chain is expected to generate roughly $200 billion. no date given25 Aug 2026pending
The data-center-market supply chain is expected to generate more than the roughly $200 billion attributed to the wafer-fabrication-equipment supply chain. no date given25 Aug 2026pending
The accelerator supply chain is expected to generate more than the roughly $200 billion attributed to the wafer-fabrication-equipment supply chain. no date given25 Aug 2026pending
The energy supply chain is expected to contribute a substantial, unspecified amount to the overall spending or market activity being discussed. no date given25 Aug 2026pending
Combined capital expenditure across the discussed supply chains is expected to be well above $2 trillion. no date given25 Aug 2026pending
AI labs may account for 70–80% of incremental compute by 2028. by 31 Dec 202825 Aug 2026pending
The speaker is uncertain about what will happen to markets if the labs take 70–80% of incremental compute by 2028. by 31 Dec 202825 Aug 2026pending
If the labs reach 100 gigawatts of combined compute in 2028, their scale will be highly disruptive to the market. by 31 Dec 202825 Aug 2026pending
Reaching 100 gigawatts of lab compute in 2028 requires labs to outbid other buyers, potentially raising compute prices from $10–15 million to $25 million or $40 million per megawatt. by 31 Dec 202825 Aug 2026pending
AI labs will probably continue acquiring a larger share of available compute. no date given25 Aug 2026pending
The companies would need to pay roughly $25–50 million per megawatt to capture 70% of the world's compute and reach 100 gigawatts by 2028. by 31 Dec 202825 Aug 2026pending
The companies may be unable to reach 100 gigawatts of compute if safety constraints matter. no date given25 Aug 2026pending
The speaker believes the described regulatory and safety constraints will prevent the companies from reaching 100 gigawatts of compute. no date given25 Aug 2026pending
The leading AI companies can generate at least $100 million per megawatt and pay $50 million per megawatt, at a price other buyers cannot match. no date given25 Aug 2026pending
Unspecified forces could potentially slow the projected compute expansion. no date given25 Aug 2026pending
The speaker expects most compute transactions to continue occurring at under $20 million per megawatt, including at the end of the year following the episode. by 31 Dec 202725 Aug 2026pending
Anthropic's or OpenAI's blended company revenue could exceed $70–80 million per megawatt, though reaching that level is challenging. by 31 Dec 202725 Aug 2026pending
If the projected high revenue per megawatt is achieved, Anthropic may consider incremental compute valuable enough to potentially pay $40 million per megawatt for SpaceX compute. no date given25 Aug 2026pending
The supply chain will rebalance over time, with costs increasing as it does. no date given25 Aug 2026pending
AI-related providers will probably raise prices because they have the ability to do so. no date given25 Aug 2026pending
In a takeoff scenario, Anthropic could keep its best model six months ahead of externally available models because of safety and regulatory constraints and competitive advantage. no date given25 Aug 2026pending
The speaker predicts that this dynamic will limit revenue-per-megawatt growth to well below the rate seen in the first half of the year. no date given25 Aug 2026pending
The speaker believes AI labs will devote a progressively smaller share of their compute to inference over time, while noting that this view is non-consensus. no date given25 Aug 2026pending
Most compute will probably be used for training forward passes rather than necessarily for revenue-generating inference. no date given25 Aug 2026pending
The speaker expects Anthropic and OpenAI, including their boards, to prioritize building AGI because he considers it much more profitable. no date given25 Aug 2026pending
AI labs will devote a growing share of compute to training, despite the increasing profit each additional unit could generate if allocated to inference. no date given25 Aug 2026pending
The company selling tokens may allocate compute to both internal and external use if internal access to fast or best-model AI generates more value than external sales. no date given25 Aug 2026pending
The speaker thinks global AI compute growth can continue accelerating, but considers forecasts beyond four years difficult because the future regime and economic growth are uncertain. no date given25 Aug 2026pending
China’s compute deployment might begin accelerating in 2028. 1 Jan 2028 – 31 Dec 202825 Aug 2026pending
China will have approximately 30 gigawatts or less of AI compute in 2028. 1 Jan 2028 – 31 Dec 202825 Aug 2026pending
China’s deployable compute capacity is expected to increase substantially in 2028. 1 Jan 2028 – 31 Dec 202825 Aug 2026pending
The speaker expects fabs associated with SMIC and CXMT to ramp in 2028, bringing Chinese domestic production to many millions of units annually. 1 Jan 2028 – 31 Dec 202825 Aug 2026pending
China could add approximately 5–10 gigawatts of compute capacity from domestically produced chips in 2028. 1 Jan 2028 – 31 Dec 202825 Aug 2026pending
Chinese chips will be inferior to the chips NVIDIA, Google, and OpenAI are expected to have in 2028. by 31 Dec 202825 Aug 2026pending
The speaker expects China to obtain increasing quantities of foreign chips for domestic use, or at minimum to narrow the gap in access created by U.S. limits on NVIDIA sales to China. no date given25 Aug 2026pending
China could add 50 gigawatts of compute capacity in 2029, including some capacity purchased from foreign suppliers; the speaker considers this entirely plausible. 1 Jan 2029 – 31 Dec 202925 Aug 2026pending
China will not slow its AI efforts and will instead accelerate them. no date given25 Aug 2026pending
If AI takeoff takes longer than implied, China could catch up substantially in semiconductors, eventually leading to greater compute capacity. no date given25 Aug 2026pending
Anthropic is expected to have nearly five gigawatts or more of compute by the end of the year. by 31 Dec 202625 Aug 2026pending
The speaker expects automated coding, automated research, and continual learning to make the division between research and model training less distinct, or to increase the share of compute used for training. no date given25 Aug 2026pending
The speaker estimates that accounting for future growth could raise the required capital expenditure from $5 trillion to roughly $7–10 trillion. no date given25 Aug 2026pending
Building 100 gigawatts would require capital expenditure substantially greater than $5 trillion once the broader infrastructure is included. no date given25 Aug 2026pending
Dylan Patel estimates that combined capital expenditure across the relevant AI-compute, data-center, energy, and supply-chain items could reach $3–4 trillion by 2028. by 31 Dec 202825 Aug 2026pending
By 2028, hyperscalers and their supply chains will each be raising hundreds of billions of dollars in debt. by 31 Dec 202825 Aug 2026pending
Investors may redirect money away from housing-related assets and government debt toward debt issued by hyperscalers, data centers, or Anthropic. no date given25 Aug 2026pending
The speaker's modeling estimates approximately $11 trillion in total capital expenditure from 2024 through 2029. 1 Jan 2024 – 31 Dec 202925 Aug 2026pending
The speaker estimates that more than $5 trillion in credit will need to be issued to fund an investment buildup exceeding $11 trillion, even after cash flows are used as much as possible. no date given25 Aug 2026pending
AI revenue will not continue rising indefinitely. no date given25 Aug 2026pending
Some AI labs will face specific incentives when constraints are reached as AI revenue increases. no date given25 Aug 2026pending
The speaker estimates that through 2029, approximately $11 trillion in capital expenditure will be funded by about $6 trillion in cash and $5 trillion in debt. by 31 Dec 202925 Aug 2026pending
One possible outcome is that AI labs increase revenue per megawatt while maintaining large inference allocations, capturing profits as other companies pay to reduce costs. no date given25 Aug 2026pending
The speaker's current model projects that, through 2029, infrastructure investment will total $5 trillion in credit-funded investment and $6 trillion in cash-funded investment. by 31 Dec 202925 Aug 2026pending
Revenue per megawatt will continue to grow, presented as the obvious answer to the compute-demand shortfall. no date given25 Aug 2026pending
Interest rates could rise if the world economy is growing substantially. no date given25 Aug 2026pending
Meta could pay 8% for financing because it expects enormous returns from the compute it builds. no date given25 Aug 2026pending
The discussed interest-rate and economic effects will happen before the singularity, according to the speaker. no date given25 Aug 2026pending
If AI demand is driving a drastic economic transformation, valuations across the market should fall to roughly two or three times earnings, and the stock market should crash. no date given25 Aug 2026pending
Meta may monetize its accumulated infrastructure and compute at very high revenue per watt, either through tokens generated by its own lab or by selling capacity to Anthropic and OpenAI. no date given25 Aug 2026pending
Governments may respond to AI by regulating it, raising interest rates, prohibiting data centers, and stopping semiconductor-fab construction. no date given25 Aug 2026pending
Anthropic is expected to expand its infrastructure over the next couple of years, including contracting its own data centers and building more of its own infrastructure. 25 Aug 2026 – 25 Aug 202825 Aug 2026pending
The speaker hopes that a slow AI takeoff remains possible, potentially because economic and regulatory constraints—including government restrictions on releasing models and using them internally—may arise soon. In a cluster of 15 →no date given25 Aug 2026pending
The U.S. government will oppose Anthropic using Mythos 4 internally and pressure the company to slow down for regulatory reasons. no date given25 Aug 2026pending
Elected officials will continue to dislike AI, according to the speaker. no date given25 Aug 2026pending
Real-world constraints will limit the progress, development, and deployment of AI, although AI will eventually be adopted. In a cluster of 15 →no date given25 Aug 2026pending
Society could tear itself apart before AI is eventually adopted. no date given25 Aug 2026pending
If recursive self-improvement is possible and AI labs are the most effective users of compute and generate the most value from it, compute will become centralized. no date given25 Aug 2026pending
The discussed AI-related factors will point toward centralization. no date given25 Aug 2026pending
A privately owned economy may grow more slowly than an AI economy because AI could make private ownership less efficient. no date given25 Aug 2026pending
The speaker expects AI to produce an extreme concentration of resources, leaving society dependent on one company getting everything right, unless AI progress slows or governments impose very heavy regulation. In a cluster of 4 →no date given25 Aug 2026pending
Governments and people may slow AI progress, hopefully creating a more balanced distribution of power. In a cluster of 4 →no date given25 Aug 2026pending
Some actor will capture an increasing share of resources at each stage on the path toward AGI, ASI, and RSI. no date given25 Aug 2026pending
The rest of the economy may capture more profit from Anthropic's activity than Anthropic does. no date given25 Aug 2026pending

Appearances

Claims per episode