Dylan Patel：Anthropic 与 OpenAI 到 2028 年将控制全球大部分算力

再次和我的双胞胎兄弟 Dylan Patel 聊天非常有趣。

我们讨论了未来几年的实验室经济——从推理向训练的转变随着 RSI 的临近；以及 Anthropic 和 OpenAI 如何在未来几年内控制全球大部分可用 FLOPs（因为他们能更好地将计算能力货币化，从而出价超过所有人）。

然后我们讨论了，到本世纪末，我们将看到的总 AI 资本支出超过 10 万亿美元，这是否会引发主权债务危机：超大云供应商的债务推高利率，使那些没有 AI 投入的国家破产，并导致非 AI 股市崩盘。

我们无法解决的一个问题是，是否存在任何力量可以对抗这个行业中所有朝向集中化推进的力量——训练中的规模经济、计算资源的稀缺性，以及最终的持续学习和 RSI。

在 YouTube 上观看：https://youtu.be/aV26V1UvkJw；在 Apple Podcasts 上收听：https://podcasts.apple.com/us/podcast/dylan-patel-anthropic-openai-will-have-most-of/id1516093381?i=1000785793715 或在 Spotify 上收听：https://open.spotify.com/episode/1chA0sqLyHUL684tUEE3ek?si=Dgplc0zpTz-H6O642qY5mA。

Grok Bot：https://x.ai/bot 在我寻找新编辑的过程中非常有帮助。我创建了一个招聘机器人，并描述了我寻找的编辑类型。然后该机器人生成了一些子代理，梳理了我的电子邮件和 X 私信，阅读了我喜欢的各种纪录片的片尾字幕，并找出谁为我最喜欢的一些 YouTuber 编辑内容。它综合了所有结果，然后向我提供了符合我标准的候选人名单。自己试试 Grok Bot：https://x.ai/bot x.ai/bot：http://x.ai/bot

Antithesis：https://antithesis.com/dwarkesh 让你可以在软件测试工具包中添加时间旅行功能。由于 Antithesis 平台完全确定性，它内部发生的一切都是完全可复现的。所以如果你的软件崩溃，你可以回到确切的时刻，冻结时间并进行调查。或者你可以通过扰动系统来测试不同假设：关闭一个节点或禁用一个功能，查看结果，然后重置轨迹再尝试其他操作。了解更多：https://antithesis.com/dwarkesh antithesis.com/dwarkesh：http://antithesis.com/dwarkesh

Jane Street：https://janestreet.com/dwarkesh 目前正在招聘两个不同的机器学习实习岗位，一个主要侧重于研究，另一个侧重于工程。在这两种情况下，实习生都需要为实际工作做出贡献，而不是完成人为设计的练习：一个常见项目是将前沿的大型语言模型论文应用于金融市场，其中会遇到大量复杂的挑战。重要的是，申请者不需要有任何金融背景。2027 年的申请现已在 janestreet.com/dwarkesh 开放：https://janestreet.com/dwarkesh

(00:00:00 ) – 很快，两个实验室将掌控世界大部分的计算能力

(00:07:01 ) – 60 亿美元的晶圆厂资本支出可以产生超过 1 万亿美元的终端收入

(00:13:08 ) – 如果实验室出价高于所有人，计算价格将会上涨

(00:18:22 ) – 哪一层将获得大部分剩余价值？

(00:25:40) – 什么可能会减慢进度？

(00:29:43 ) – 实验室正在将计算能力从推理转向研发

(00:33:27 ) – 中国获得的新计算量不足 10%，但其实验室所需较少

(00:48:48 ) – 人工智能会引发主权债务危机吗？

(01:07:52) – 世界未来的劳动力会属于少数几家公司吗？

好了，我回来了，这次带来了 SemiAnalysis 的创始人 Dylan Patel：https://semianalysis.com/。我们版本的家庭感恩节晚餐是一档常规年度播客。但实际上我们并没有血缘关系。

它将摧毁神话。基本上，世界经济的走向越来越多地取决于实验室经济的走向、计算市场的走向等等。我想弄清楚几年内疯狂的未来最终会发展到哪里。但让我们先从今天的情况开始。带我了解一下当前实验室的计算能力和收入情况，并可能预测一两年后的情况。

当我们回顾去年时，即使在年底，美国大部分 GDP 增长只是由 AI 基础设施推动：https://fortune.com/2025/10/07/data-centers-gdp-growth-zero-first-half-2025-jason-furman-harvard-economist/。展望今年，大约三分之一的新上线计算能力是为实验室准备的，比如 OpenAI 和 Anthropic。虽然这些计算能力可能由其他公司建造后租给他们，但最终客户是他们本身。

随着我们走向未来，计算需求的数字正在膨胀：https://epoch.ai/data-insights/hyperscaler-capex-trend。今年我们在资本支出上略超过一万亿美元：https://www.investopedia.com/terms/c/capitalexpenditure.asp。到2028年，这一数字将超过2万亿美元。实验室也在占据这一支出的越来越大比例。因此，最终你会看到一个非常有趣的情况：实验室的支出从每年几十亿美元增长到每年数百亿美元，并预测到十年末甚至每年达到数万亿美元。这至少是他们开始与合作伙伴签署的一些合同。

这需要对他们的经济模式进行大规模重塑。直到现在，他们大多还是亏损的公司。Anthropic在第二季度开始盈利。有人认为，OpenAI 可能在第三季度某个时间点开始盈利，特别是随着Codex：https://openai.com/index/introducing-the-codex-app/ 和5.6：https://openai.com/index/previewing-gpt-5-6-sol/ 的大幅增长。但是如果回顾一年前，他们所有的钱都是来自风险投资的亏损。如果回到今年年初，也还是风险投资亏损。他们现在已经转亏为盈，实际上开始盈利。

这并不意味着他们不在引入新资本。新资本仍在注入，以进一步加速增长。但最终，他们越来越多的业务是依靠自身收入而非资本注入来资助。在过去一年半里，他们的利润率确实飙升。计算的基础成本通常在每兆瓦1000万到1500万美元左右：https://www.nrc.gov/docs/ML1209/ML120960701.pdf。

目前发生的最有趣的方面是：之前，如果他们提供一个模型——GPT-4（https://openai.com/index/gpt-4-research/），运行在Nvidia（https://en.wikipedia.org/wiki/Nvidia）Hopper（https://en.wikipedia.org/wiki/Hopper_(microarchitecture)）GPU（https://en.wikipedia.org/wiki/Graphics_processing_unit）上——这会给OpenAI带来负毛利（https://www.investopedia.com/terms/g/grossmargin.asp）。但现在，当OpenAI提供GPT-5.6或Anthropic提供Opus 5（https://www.anthropic.com/news/claude-opus-5）或Fable 5（https://www.anthropic.com/claude/fable）时，他们的收入已经远远超过每兆瓦增量10-15百万美元的水平。在Anthropic的情况下，收入甚至达到了每兆瓦5,000万美元。这使他们现在能够做到：“嘿，如果我在推理（https://cloud.google.com/discover/what-is-ai-inference）能力上花10美元，我实际上可以生成50美元的收入，然后我可以将所有这些利润用于训练。”

我非常感兴趣的是想理解你如何看待实验室的计算集中化，或者流向世界与流向实验室的计算比例。如果你说现在三分之一的边际计算资源流向实验室，那么在什么时候世界上超过一半的新增计算资源会流向实验室？实验室在何时基本上拥有世界上大多数的计算资源？

今年年初，OpenAI的计算能力为2吉瓦，而Anthropic不到2吉瓦。今年年底，他们都超过了5吉瓦。所以他们整体计算能力提升了3-4倍。当你看今年新增的增量计算能力时，那大约占今年新增计算能力的30%。

展望明年，鉴于已经签署和落实的内容，你会看到更为显著的情况。明年，Anthropic和OpenAI将占用多达40%到50%的计算资源。这种集中化似乎没有放缓或停止的迹象。实际上，看起来只会加速。

谁在为他们建造计算能力会发生变化。比如明年，一个新的大型参与者是 SpaceX（https://en.wikipedia.org/wiki/SpaceX），它正在建造大量计算能力。他们将积极地把相当一部分租给 Anthropic（https://www.wsj.com/livecoverage/stock-market-today-dow-sp-500-nasdaq-05-20-2026/card/anthropic-rents-1-25-billion-of-spacex-data-center-capacity-each-month-uNS0TiOKvaa8wo4GhxII）和 OpenAI，很可能是后者，因为他们有支付最高价格的边际能力。此外，OpenAI 和 Anthropic 也开始建设自己的计算能力——OpenAI 使用自己的芯片（https://www.wsj.com/tech/ai/openai-broadcom-develop-custom-chip-for-ai-inference-beafd74a），Anthropic 使用他们从谷歌购买的 TPU（https://www.anthropic.com/news/google-broadcom-partnership-compute）并通过 Fluidstack 部署（https://www.anthropic.com/news/anthropic-invests-50-billion-in-american-ai-infrastructure）。

所以你会问，“嘿，世界上新增计算能力的一半什么时候会只流向 OpenAI 和 Anthropic？”实际上，到明年年底时，增量计算能力的一半已经流向 Anthropic 和 OpenAI。

因为计算能力增长非常快，增量计算能力基本上将占据大部分计算能力。所以很快——你可以说也许在一年半到两年内——世界上大部分计算资源将被两个实验室拥有，或者至少是满足两个实验室的需求。

有这么一个趋势：也许世界计算能力按千兆瓦每年翻一番，但前沿实验室的计算能力每年却会翻三倍。如果保持当前趋势，今年年初的 2，到年末接近 6。仅按三倍计算，到 2027 年年底是 18，2028 年年底是 54。你会想，“好吧，到那时，考虑到世界计算量，他们根本不可能继续每年翻三倍”？你怎么看未来几年世界计算能力的情况？

所以最终你这里有一个巨大的阶梯。如果 Anthropic 和 OpenAI 明年占据 45% 的计算能力，假设到 2027 年 12 月，他们已经占据了全球新增计算能力的一半。但那半数新增计算能力实际上比之前所有的计算能力性能更高。所以你还有另一个乘数。如果这个趋势持续，到2028年底——我看不出有什么能阻止它——他们就几乎单独控制了全球大部分可用的浮点运算能力（flops）：https://en.wikipedia.org/wiki/Floating_point_operations_per_second。

让我困惑的是，既然我们进入了一个计算价值大幅增加的世界，你为什么认为我们在2028年只新增 80 吉瓦。

顺便说一下，那是上限。就像是，“我真是太看涨了”。

好，让我们理一下思路。几个月前我采访你时：https://www.dwarkesh.com/p/dylan-patel 你说，为了制造一吉瓦的 Vera Rubins：https://nvidianews.nvidia.com/news/nvidia-vera-rubin-platform，你需要 55,000 片 N3：https://en.wikipedia.org/wiki/3_nm_process 晶圆：https://en.wikipedia.org/wiki/Wafer_%28electronics%29，6,000 片 N5：https://en.wikipedia.org/wiki/5_nm_process 晶圆，以及 170,000 片 DRAM：https://en.wikipedia.org/wiki/Dynamic_random-access_memory 晶圆。我知道这些数字可能已经有所变化。

我要捣你一下，但你说“晶圆”的发音真他妈印度式。Vafers。

顺便说一句，当我们刚搬到美国时，我的 v/w 发音问题很严重，而且我还是个素食主义者。

在北达科他州，我上小学时，我会像——

总之，那是针对一吉瓦。 我让一个大语言模型运行你的晶圆厂设备模型，计算每年生产一吉瓦计算能力的工具成本。模型说是 30 到 40 亿美元。现在假设你再加上无尘室：https://en.wikipedia.org/wiki/Cleanroom、厂房外壳以及晶圆厂的其他所有设施：https://en.wikipedia.org/wiki/Semiconductor_fabrication_plant。所以 60 亿美元的晶圆厂资本支出每年可以生产一吉瓦。一吉瓦现在带来 1,000 亿美元的收入。

但同样，这60亿美元的资本支出每年可以产生一吉瓦的产能，而这一吉瓦每年能产生1000亿美元的收入。所以在五年的时间里，第一吉瓦已经产生了五年的利润，第二吉瓦的晶圆厂产能产生了四年的利润，依此类推。晶圆厂层面的60亿美元资本支出将生成超过一万亿美元的最终人工智能收入。

是的。途中还有很多运营支出。还有许多其他资本支出，比如数据中心、电力。

而且你必须向 OpenAI 支付研发费用。

安装。这里有很多不同的人需要资金。

扣掉一半给所有这些中间人。这仍然意味着晶圆厂资本支出和最终收入之间存在100倍的差距。实际上更多，但我们只是非常保守地估计。因此……这就是资本主义。你有这样一个巨大的差距，可以把1美元变成100美元。他们不会想着办法制造更多的镜子：https://www.zeiss.com/semiconductor-manufacturing-technology/smt-magazine/so-does-euv-lithography-work.html？

会的。只是这些镜子制作需要一些时间。

但紧急情况非常严重，Anthropic和OpenAI就像说：“我们现在可以赚一万亿美元，但是我们被ASML：https://en.wikipedia.org/wiki/ASML_Holding 机器里使用的镜子瓶颈了。”如果我们花1000亿美元，我们怎么能制造更多的镜子呢？这就是我们很快将要面临的情况。我们恐怕无法解决这个供应限制？这似乎很难想象。

你已经看到有人在这里做有趣的套利，他们购买涡轮：https://en.wikipedia.org/wiki/Gas_turbine 然后试图转售，因为涡轮的价值更高，因为它是你的数据中心的瓶颈。我认为，如果有人有4亿美元，并且能够说服ASML卖给他们一台EUV：https://en.wikipedia.org/wiki/Extreme_ultraviolet_lithography 工具，他们完全应该去买一台，等待，然后再以超过十亿美元的价格卖出。

但归根结底，是的，资本主义会导致这些事情扩张。但它就像一根鞭子。鞭子的信号传达到尾端需要很长时间。供应链不会立即做出反应。事实上，你去跟蔡司公司的一些人谈谈：https://en.wikipedia.org/wiki/Carl_Zeiss_SMT，他们会说，“是的，是的，是的，我们需要在十年内制造100台EUV工具。”我们在今年早些时候的一期节目中，他们甚至都没认为自己需要制造那么多，甚至连一年制造100台EUV工具所需的镜子都不够。现在他们说，“好吧，我们需要做到。”但实际上，由于所有经济因素的影响，这个数量应该更多。需要很长时间才能整合。

假设堆栈中的每一家公司都进行了私募股权融资。有人进来，他对超级通用人工智能（AGI）https://en.wikipedia.org/wiki/Artificial_general_intelligence 非常了解，然后说，“我们要最大化生产。”你认为制造更多东西的物理限制是什么？我之所以问，是因为我们很快就要进入一个世界，在那里实验室的收入，或者仅仅是人工智能的现金流——因为显然加速器也有巨大的现金流——会非常巨大，以至于你可以仅靠这些现金流来资助所有生产的极端扩张。

我总体上同意。显然存在一些物理限制。按照目前供应链的扩张方式，100台仍然大致是正确的数字。

2030年的100台ASML工具。但如果你说，“蔡司，这里有100亿美元，请他妈的扩产”，那情况就会改变。你必须对供应链中的每家公司都这样做。

但你认为这会在明年发生吗？

我认为今年不会发生。我认为明年不会发生。我认为后年也不会发生，因为世界受资本限制。

但在一个世界中，比如说顶级实验室明年即便合计产生1万亿美元收入，他们也无法拿出其中的100亿美元——

我不认为他们会这么做，但是……

或者至少数千亿美元？他们似乎意识到世界的发展方向。我觉得他们完全可以制造……

问题是，实验室明年的收入将达到数千亿美元。但归根结底，明年的资本支出大约是 2 万亿美元。所以你会看到这种巨大的不匹配。晶圆制造设备供应链的规模大约是 2000 亿美元。数据中心市场供应链的规模会更大。加速器供应链会更大。能源供应链也会有一定的规模。把这些加起来，总额肯定远超过 2 万亿美元的资本支出。因此，实验室的现金流还没有达到能够资助这些项目的水平。

显然，他们永远也不会达到这个水平，因为你希望将资本支出保持在比回报更高的水平。

我真正想理解的关键问题是：如果当前趋势持续，到 2028 年底，每个实验室将超过 50 吉瓦的配置。它们加起来将有 100 吉瓦。这些吉瓦数量，正如你所说，到 2028 年将带来比现在多得多的吞吐量或性能，因为硬件变得更好了。不仅每瓦浮点运算能力增加了，而且硬件在处理 AI 工作负载方面也更出色。

好的，所以到 2028 年底，实验室总共将有 100 吉瓦。全球计算能力是多少？

Had a lot of fun chatting again with my twin brother Dylan Patel.

We went through lab economics over the next few years - the shift from inference to training as RSI draws near; and how Anthropic and OpenAI are on track to control most of the world’s usable FLOPs within the next few years (because they can monetize compute better and thus outbid everyone).

And then we discuss whether the >$10T of total AI capex we’ll see by the end of the decade will cause a sovereign debt crisis, where hyperscaler debt raises interest rates, drives non-AI exposed countries into bankruptcy, and crashes non-AI equities.

One question we weren’t able to resolve is whether there’s anything that can counter all the forces barrelling towards centralization in this industry - the economies of scale in training, the scarcity of compute, and eventually continual learning and RSI.

Watch on YouTube：https://youtu.be/aV26V1UvkJw ; listen on Apple Podcasts：https://podcasts.apple.com/us/podcast/dylan-patel-anthropic-openai-will-have-most-of/id1516093381?i=1000785793715 or Spotify：https://open.spotify.com/episode/1chA0sqLyHUL684tUEE3ek?si=Dgplc0zpTz-H6O642qY5mA .

Grok Bot ：https://x.ai/bot has been quite helpful with my search for a new editor. I created a recruiter bot and described the type of editor I was looking for. That bot then spun up a handful of subagents that combed through my emails and X DMs, read the end credits of various documentaries I like, and figured out who edits for some of my favorite YouTubers. It took all of those results, and then delivered me a shortlist of candidates that matched my criteria. Try Grok Bot for yourself at ：https://x.ai/bot x.ai/bot ：http://x.ai/bot

Antithesis ：https://antithesis.com/dwarkesh lets you add time travel to your software testing toolkit. Since the Antithesis platform is fully deterministic, everything that happens inside of it is perfectly reproducible. So if your software crashes, you can rewind to the exact right moment, freeze time, and investigate. Or you can test different hypotheses by perturbing the system: kill a node or disable a feature, see what happens, then reset the trajectory and try something else. Learn more at ：https://antithesis.com/dwarkesh antithesis.com/dwarkesh ：http://antithesis.com/dwarkesh

Jane Street ：https://janestreet.com/dwarkesh is hiring for two separate ML internships right now, one focused primarily on research and one focused on engineering. In both cases, interns are expected to contribute to real work, not contrived exercises: one common project is adapting a frontier LLM paper to financial markets, which tend to come with a ton of different gnarly challenges. Importantly, you don’t need any finance background to apply. 2027 applications are open now at janestreet.com/dwarkesh ：https://janestreet.com/dwarkesh

(00:00:00 ) – Two labs will soon control most of the world’s compute

( 00:07:01 ) – $6 billion in fab capex enables $1t+ of end revenue

( 00:13:08 ) – Compute prices will rise if the labs outbid everyone

( 00:18:22 ) – Which layer will capture most of the surplus?

( 00:25:40 ) – What could slow down progress?

( 00:29:43 ) – Labs are shifting compute from inference to R&D

( 00:33:27 ) – China gets less than 10% of new compute, but its labs need less

( 00:48:48 ) – Will AI cause a sovereign debt crisis?

( 01:07:52) – Will the world’s future workforce belong to a few companies?

Okay, I’m back with Dylan Patel, founder of SemiAnalysis ：https://semianalysis.com/ . Our version of a family Thanksgiving dinner is a regular yearly podcast. But we’re not actually related.

It will destroy the myth. Basically where the world economy is headed is more and more becoming a function of where lab economics are headed, where the compute market is headed, et cetera. I want to understand where the crazy future ends up within a few years. But let’s start with where we are today. Walk me through lab compute and lab revenue right now, and maybe project out a year or two.

When we go back to last year, even at the end of the year, most of GDP growth in America was just AI infrastructure ：https://fortune.com/2025/10/07/data-centers-gdp-growth-zero-first-half-2025-jason-furman-harvard-economist/ . As we look towards this year, about a third of the compute coming online is for the labs, for OpenAI and Anthropic. It may be built by others and then rented to them, but at the end customer, it’s them.

As we go forward into the future, the numbers for compute are ballooning ：https://epoch.ai/data-insights/hyperscaler-capex-trend . We’re at a little bit over a trillion dollars of CapEx ：https://www.investopedia.com/terms/c/capitalexpenditure.asp this year. As we go out into ’28, it’s going to be more than $2 trillion. The labs are also taking an increasing percentage of this. So ultimately, you’ve got a very interesting situation where the labs are going from companies that spend tens of billions of dollars a year to hundreds of billions of dollars a year, to forecasting to spend trillions of dollars a year even towards the end of the decade. This is at least some of the contracts they’ve begun signing with their partners.

This requires a big reshaping of what happens with their economics. Up until now, they have been companies that mostly lost money. Anthropic started turning a profit in Q2. It’s believed at some point in Q3, OpenAI could start turning a profit even, with the bigger rise of Codex ：https://openai.com/index/introducing-the-codex-app/ and 5.6 ：https://openai.com/index/previewing-gpt-5-6-sol/ and all this. But if we go back a year ago, all the money they had was venture-funded losses. If we go back to even the beginning of this year, it was venture-funded losses. They’ve now turned the corner and are actually starting to profit.

That doesn’t mean they’re not taking in new capital. The new capital is still coming in to accelerate the growth further. But ultimately, more and more of their business is being funded off of their own revenue rather than capital injections into them. Over the last year and a half, their margins have really skyrocketed. The base cost of compute tends to be around $10 or $13 or $15 million per megawatt ：https://www.nrc.gov/docs/ML1209/ML120960701.pdf .

The most interesting aspect about what’s happening now is this: Before, if they served a model — GPT-4 ：https://openai.com/index/gpt-4-research/ being served on Nvidia ：https://en.wikipedia.org/wiki/Nvidia Hopper ：https://en.wikipedia.org/wiki/Hopper_(microarchitecture) GPUs ：https://en.wikipedia.org/wiki/Graphics_processing_unit — it was generating negative gross margin ：https://www.investopedia.com/terms/g/grossmargin.asp for OpenAI. But now, when OpenAI serves GPT-5.6 or Anthropic serves Opus 5 ：https://www.anthropic.com/news/claude-opus-5 or Fable 5 ：https://www.anthropic.com/claude/fable , their revenue generation has passed well beyond the incremental $10-15 million per megawatt. In the case of Anthropic, the revenue has gone as high as $50 million per megawatt. What that now enables them to do is: “Hey, if I spend 10 bucks on inference ：https://cloud.google.com/discover/what-is-ai-inference capacity, I actually generate 50 bucks of revenue, and then I can turn around and incrementally spend all of that profit on training.”

One thing I’m very interested in understanding is how you see the centralization of compute happening at the labs, or the relative ratio of compute that goes to the world versus the labs. If you say right now a third of marginal compute is going to the labs, by when is over half of the incremental compute in the world going to the labs? By what point do the labs have basically a vast majority of the world’s compute?

At the beginning of this year, OpenAI started at 2 gigawatts and Anthropic at less than 2. End of this year, they’re both above 5. So they’ve 3-4x’d compute as a whole. When you look at the incremental compute added, that’s about 30% of the compute added this year.

As we step forward to next year, given what’s already been signed and penned and inked, you’ve got something even more dramatic. Anthropic and OpenAI are taking as much as 40% to 50% of compute next year. This centralization doesn’t look like it’s slowing down or stopping. In fact, it looks like it’s only accelerating.

Who’s building that compute for them will change. Next year, a big new entrant is, for example, SpaceX ：https://en.wikipedia.org/wiki/SpaceX , which is building a ton of compute. They’re actively going to lease quite a bit of it to Anthropic ：https://www.wsj.com/livecoverage/stock-market-today-dow-sp-500-nasdaq-05-20-2026/card/anthropic-rents-1-25-billion-of-spacex-data-center-capacity-each-month-uNS0TiOKvaa8wo4GhxII and OpenAI, most likely, because they’re the ones who have the marginal capability to pay the highest price. In addition, OpenAI and Anthropic are also starting to build their own compute — OpenAI with their own chips ：https://www.wsj.com/tech/ai/openai-broadcom-develop-custom-chip-for-ai-inference-beafd74a , Anthropic with TPUs that they’re purchasing from Google ：https://www.anthropic.com/news/google-broadcom-partnership-compute and deploying with Fluidstack ：https://www.anthropic.com/news/anthropic-invests-50-billion-in-american-ai-infrastructure .

So you ask, “Hey, when does half of the world’s incremental new compute go to just OpenAI and Anthropic?” It’s really by the end of next year when half of the incremental compute is already going to Anthropic and OpenAI.

Because compute is growing so fast, incremental compute is going to be basically most of compute. So it’s very soon — you’re saying maybe within a year and a half or two years — that most of the world’s compute is owned by two labs, or at least is serving the demand from two labs.

There’s this trend where maybe world compute in gigawatts doubles every year, but the compute at the frontier labs triples every single year. If you keep the current trend going, it goes from 2 at the beginning of this year to close to 6 at the end of this year. Just multiplying out by 3. It’s 18 by the end of 2027, 54 by the end of 2028. Are you like, “Okay, at that point, they simply can’t continue tripling given the amount of world compute”? How do you see the world compute situation over the next few years?

So ultimately you’ve got a huge ladder here. If Anthropic and OpenAI take on 45% of compute next year, you’ve got them in, let’s say, December ’27 having taken on half of the world’s incremental new compute. But that half of the world’s new incremental compute is actually at a higher performance than everything else before it. So you’ve got another multiplier on that. By the time you’re towards the end of 2028 — if this trend continues, and I see nothing that’s stopping it — you’ve got them just controlling most of the usable flops ：https://en.wikipedia.org/wiki/Floating_point_operations_per_second in the world on their own.

The thing I’m confused about is why you think we only add 80 gigawatts in 2028 if we enter a world in which the value of compute increases so much.

That’s the upper bound, by the way. That’s the like, “I’m so fucking bullish.”

Okay, let’s do some chain of thought here. When I interviewed you a few months ago ：https://www.dwarkesh.com/p/dylan-patel , you said that in order to make a gigawatt of, I think, Vera Rubins ：https://nvidianews.nvidia.com/news/nvidia-vera-rubin-platform , you need 55,000 N3 ：https://en.wikipedia.org/wiki/3_nm_process wafers ：https://en.wikipedia.org/wiki/Wafer_%28electronics%29 , 6K N5 ：https://en.wikipedia.org/wiki/5_nm_process wafers, and 170K DRAM ：https://en.wikipedia.org/wiki/Dynamic_random-access_memory wafers. I know if those numbers might have changed.

I’m going to troll you, but the way you said wafers was so fucking Indian. Vafers.

By the way, when we first moved to the US, I had the v/w thing pretty bad, and I was a vegetarian.

In North Dakota, I was in elementary school, and I’d be like—

Anyways, so that’s for one gigawatt. I had an LLM run your wafer fab equipment model and figure out how much the tooling costs to produce a gigawatt of compute basically every single year. It said $3-4 billion. Now suppose you add in cleanrooms ：https://en.wikipedia.org/wiki/Cleanroom and shell and everything else at the fab ：https://en.wikipedia.org/wiki/Semiconductor_fabrication_plant . So $6 billion of fab CapEx produces a gigawatt every single year. A gigawatt produces right now $100 billion of revenue.

But also that $6 billion in CapEx is producing a gigawatt every single year, and that gigawatt is producing $100 billion every single year. So over the course of five years, the first gigawatt has generated five years of profits, the second gigawatt the fab has produced has generated four years of profits, and so on. $6 billion of CapEx at the fab level will have generated over a trillion dollars of end AI revenue.

Yeah. There’s a lot of OpEx along the way. There’s a lot of other CapEx, like the data center, the power.

And you had to pay OpenAI for the R&D.

Installation. There’s a lot of different people who need money here.

Take away half of it for all these middlemen. That still means there’s a 100x discrepancy between fab CapEx and end revenue generated. More than that, actually, but we’re just being very conservative. As a result… This is capitalism. You have this huge discrepancy where you can turn $1 into $100. They’re not going to figure out a way to make more mirrors ：https://www.zeiss.com/semiconductor-manufacturing-technology/smt-magazine/so-does-euv-lithography-work.html ?

They are. It’s just that these mirrors take some time to make.

But the emergency is so big where Anthropic and OpenAI are like, “We could make a trillion dollars right now, but we’re just bottlenecked on the mirrors that go into the ASML ：https://en.wikipedia.org/wiki/ASML_Holding machines.” How can we make more mirrors if we spend $100 billion on this? That’s the situation we’re going to be in pretty soon. We’re not going to be able to solve that supply constraint? That just seems quite hard to imagine.

You’ve seen people do funny arbitrages here where they buy turbines ：https://en.wikipedia.org/wiki/Gas_turbine and then try and resell them, because the value of a turbine is way more since it’s the thing bottlenecking your data center. I think if anyone had $400 million and the ability to convince ASML to sell them an EUV ：https://en.wikipedia.org/wiki/Extreme_ultraviolet_lithography tool, they should totally just go buy one, wait, and then sell it for north of a billion dollars.

But ultimately, yes, capitalism will cause these things to expand. But it’s a whip. It takes a long time for the whip signal to get to the tail end of that. The supply chain doesn’t react immediately. In fact, you go talk to someone at Carl Zeiss ：https://en.wikipedia.org/wiki/Carl_Zeiss_SMT , they’re like, “Yeah, yeah, yeah, we need to make 100 EUV tools by the end of the decade.” When we had our episode earlier this year, they didn’t even think they needed to make that many, enough mirrors to make 100 EUV tools a year. Now they’re like, “Okay, we need to do that.” But in reality, because of all the economics of what’s going on, it should be even more. It takes so long to pill.

Suppose that every single company in the stack got private equitied. Somebody came in who was super AGI ：https://en.wikipedia.org/wiki/Artificial_general_intelligence -pilled and was like, “We’re going to maximize production.” What do you think the physical constraints on making more things would be? The reason I ask is we’re pretty soon going to be in a world where the lab revenue, or just AI cash flows — because obviously the accelerators also have these huge cash flows — will be so big that you can just fund extreme expansion of all this production from cash flows themselves.

I do agree generally. There’s obviously some physical constraints. The way the supply chain is expanding currently, 100 is roughly still the right number.

100 ASML tools for 2030. But if you said, “Carl Zeiss, here’s $10 billion. Please fucking just expand production,” that would change things. You would have to do this with every company in the supply chain.

But you don’t think that’s gonna happen next year?

I don’t think it’ll happen this year. I don’t think it’ll happen next year. I don’t think it’ll happen the year after, because the world is capital constrained.

But in a world where, say, the top labs are generating, even combined, a trillion dollars in revenue next year, they’re not able to take $10B of that—

I don’t think they’re going to do that, but…

Or hundreds of billions at least? It just seems like they realize where the world is headed. I feel like they could just make…

The thing is, the labs are going to generate hundreds of billions of revenue next year. But ultimately, CapEx next year is like $2 trillion. So you’ve got this big mismatch. The wafer fabrication equipment supply chain will do something on the order of $200 billion. The data center market supply chain will do even more. The accelerator supply chain will do even more. The energy supply chain will do a number. You sum all this up, it’s going to be well north of $2 trillion of CapEx. So the labs have not yet gotten to the point where their cash flows can fund this stuff.

Obviously they will never get to that point, because you want to keep your CapEx higher than your returns.

The key question I really want to understand is: if the current trend continues, it’d be north of 50 gigawatts per lab by the end of 2028. So between them they’d have 100 gigawatts. Those gigawatts, as you’re saying, drive many-fold more throughput or performance by 2028 than they do now, because the hardware’s gotten better. Not only have flops per watt increased, but also the hardware gets better at working with AI workloads.

Okay, so 100 gigawatts for the labs by the end of 2028. How much is world compute?