---
title: "AGI: what it means · The AI Compute Race · DRYBURGH"
source: https://dryburgh.com/vol1/glossary/agi
description: "AGI: Artificial general intelligence. Its definitions differ, and the shortest forecasts of its arrival use the loosest ones. Chapter 3."
---

# AGI

What does “AGI” mean?

AGIArtificial general intelligence. Its definitions differ, and the shortest forecasts of its arrival use the loosest ones. *Chapter 3.*

From *The AI Compute Race*. [Read the book](https://dryburgh.com/vol1#buy)

## Where the book uses it

- [Chapter 3 — The Missing Researcher](https://dryburgh.com/vol1/notes/chapter-3)The dated promise of an AI that does its own research, what the laboratories have published, and what the public record can and cannot yet check.

## Where the record uses it

- Person [Ghodsi, Ali](https://dryburgh.com/vol1/people#ghodsi-ali)
- Person [Kokotajlo, Daniel](https://dryburgh.com/vol1/people#kokotajlo-daniel)

## Sources that use it

- Dwarkesh Patel with Gavin Leech, *The Scaling Era: An Oral History of AI, 2019–2025* (Stripe Press, 2025) (curated interviews from the Dwarkesh Podcast, 2019–2025, the book’s own note on its text being “The interview excerpts have been edited for clarity and flow”, so each quotation here is the editors’ text, and where the recording differs the chapter cites the podcast’s own row; Amodei’s unpublished 2017 OpenAI memo “The Big Blob of Compute”; Zuckerberg, 2024, as the editors print him: “I think it’s worth investing $100 billion plus in building the infrastructure”, where the podcast’s transcript and the recording have the $10Bs or $100B+ and “and assuming that if it keeps going, you’re going to get some really amazing things… I don’t think anyone in the industry can really tell you for sure that it will continue scaling at that rate” (chapter 5, the ellipsis for “that make amazing products”); Zuckerberg, 2024, energy the coming bottleneck: “No one has built a 1 GW data center yet”, the recording’s a single gigawatt, and of new power plants and transmission lines, “You’re talking many years of lead time”; Dylan Patel, 2024, in the editors’ text and found on no published recording or transcript of the podcast: “In 1990, ExxonMobil was not laying fiber. In 2025, Microsoft is building data centers”; “prior bubbles didn’t have the most profitable companies that humanity has ever created investing in them. They were debt financed instead. AI is not debt financed yet”, the editors’ wording of the recording’s they were debt financed. This is not debt financed yet; Dylan Patel, 2024, in the editors’ text and found on no published recording or transcript of the podcast: “If human capital is infinite, which is what AGI is, then theoretically the returns of AGI are infinite. If I’m Mark Zuckerberg or Satya Nadella, I now have potentially infinite returns — if I get there first” (chapter 6, the book’s closed dash spaced); Aschenbrenner, 2024: a 10 GW cluster “a couple hundred billion”, the recording’s maybe before it dropped by the editors, the payoff test “you have to get to, like, $100 billion a year in revenue”, the podcast transcript’s form of the recording’s eventually you got to get to like 100 billion a year; Aschenbrenner, 2024: “Let’s estimate the GPT-4 cluster at around $500 million”, and of the $100 million figure, “That’s just the rental price. If you’re building the biggest cluster, you have to build and pay for the whole cluster”, the second sentence the podcast transcript’s form of the recording’s you got to build the whole cluster, you got to pay for the whole cluster; Aschenbrenner, 2024, on buying an aluminum smelter’s gigawatt: “a lot of these power contracts are locked in long term. Also, people don’t like it”, the recording’s people don’t like things like this; Aschenbrenner, 2024: “Would you do the Manhattan Project in the UAE?”, “Once the cluster is there, it’s much easier for them to exfiltrate the weights”; Shulman: a percentage point of search share “a billion-dollar matter”, $50–70 trillion of annual wages as the prize, and “the chance of advanced AI happening is relatively concentrated in the next 10 years [between 2024 and 2034] compared to the rest of the century, because we can’t keep going with our current rapid redirection of resources into AI. That’s a one-time thing”; Amodei: datacenter networks “cost as much as aircraft carriers”, a forecast on the recording (we’re anyway heading to a world where), which the editors set as already; Legg’s 50 percent by 2028; Sholto Douglas, March 2024, as the editors print him: “That is exactly the problem. But the key issue is that your base task solve rate is 90 percent. If it was 99 percent, it wouldn’t be a problem”; of the memo, “an unpublished memo he wrote at OpenAI in 2017—before scaling laws, before the Transformer, and before LLMs”, the editors’ order, where the recording they draw on has Amodei saying of the transformer paper It might have been just before. It might have been just after; chapter 5, “Inputs”, the same exchange as the book edits it: “You can put the clusters in the US and in allied democracies. Once you put them in authoritarian dictatorships, you create this irreversible security risk”; “Another thing: They can just seize the compute”; “Suppose we put 25 percent of the compute capacity in these Middle Eastern dictatorships. Say they seize that. Now the US has a 3-to-1 compute ratio.”). [In the Sources](https://dryburgh.com/vol1/sources/book#scaling-era-patel)
- Karen Hao, *Empire of AI: Dreams and Nightmares in Sam Altman’s OpenAI* (Penguin Press, 2025), ch. 5 “Scale of Ambition”, ch. 12 “Plundered Earth” and the Epilogue (the Microsoft–OpenAI supercomputer “phases”, from an internal OpenAI document: Phase 1 West Des Moines, 10,000 V100s, GPT-3; Phase 2 Iowa, 18,000→~25,000 A100s, GPT-4; Phase 3 Arizona, hundreds of thousands of H100s; Phase 4 Wisconsin, ~$10bn, B100s; Phase 5 ~$100bn; ch. 5 on the post-ChatGPT funding of Chinese models; the book takes only the environmental filing’s 169 liters a second for the Cerrillos site, not Hao’s population comparison, which she later retracted; the quotation “In AI, whoever has the biggest computer gets the most benefits” is read in the full text of the book, not on the publisher’s page the URL names; in the Epilogue, Uruguay is told as the campaigner Daniel Pena’s account in Hao’s words: the environmental ministry “quietly approved the project, revealing the decision only after the thirty-day public contestation window was over, he says”; ch. 5, on Chinese funding after ChatGPT: companies and investors began funding the development of gargantuan models “with gusto”, having “seen enough evidence that they could recoup their investments through commercial applications”; ch. 5, on Microsoft’s investment: “$1 billion investment came with $1 billion expectations; OpenAI was on the clock to deliver something that would justify the expense”; ch. 12, an OpenAI employee on Phase 5: “we’re running out of land and power”; ch. 5, Hao’s own sentence, reporting without quotation marks a Chinese AI researcher’s analysis of China before ChatGPT: no team “would get $1 billion, let alone ten times more, to develop a massively expensive technology without an articulated vision of exactly what it would look like and what it would be good for”; ch. 5, her conclusion: “Everything OpenAI did” … “was the opposite of inevitable” (everything lower-case mid-sentence in the source); ch. 6, “Ascension”, on a memo Altman sent to the company in late 2019, quoted from her copy of it, her note reading “Copy of the memo.”: OpenAI needed to “be number one” by the end of 2020 in four categories, which she lists as technical results, compute, money to acquire more compute, and preparation, and “We still need many more 10x leaps to get to AGI”; the Prologue, “A Run for the Throne”: the empires “seize and extract precious resources to feed their vision of artificial intelligence: the work of artists and writers; the data of countless individuals posting about their experiences and observations online; the land, energy, and water required to house and run massive data centers and supercomputers” and “exploit the labor of people globally to clean, tabulate, and prepare that data”; ch. 12, “Plundered Earth”, on Cerrillos: “In late 2019, as communities across the country began to hold referendums in response to the movement to reform local and national politics, MOSACAT piggybacked on the referendum in Cerrillos to add a question about whether residents agreed for Google to build a facility that would consume so much of the community’s water”; “In December 2019, MOSACAT won; the referendum to build the data center was rejected with a slim majority of the vote.”; ch. 7, “Science in Captivity”, on China after GPT-3, in a sentence whose note cites Ding and Xiao: “In China, GPT-3 similarly piqued intensified interest in large-scale models”, the Chinese giants treating the direction “as a novel addition to their research repertoire, not a new singular path of AI development warranting the suspension of their other projects”; the sentence is Brockman’s, in Hao’s own words, “Brockman told me in 2019,” in the chapter on scale; Brockman’s title at the time of her reporting, from her MIT Technology Review profile of OpenAI of 17 February 2020, “The messy, secretive reality behind OpenAI’s bid to save the world”, whose caption reads “Greg Brockman, co-founder and CTO”; the same profile’s caption gives “Dario Amodei, research director” and its text “the policy team, especially Jack Clark, the director”, their titles at the time of her 2019 reporting; ch. 12, the Phase 1 code names: “Microsoft code-named the Phase 1 supercomputer Odyssey; OpenAI called it Owl”; ch. 5, on the machine: “Microsoft was about to deliver a new supercomputer to OpenAI as part of its investment, with ten thousand Nvidia V100s”; ch. 5 notes: “Author interview with Dario Amodei, August 2019” and “Author interview with Jack Clark, August 2019”; Notes, “Author interview with Brockman, August 2019”; ch. 5, the run: “Amodei wanted to use all of those chips, all at once”, “Before then, models were already considered large-scale if trained on a few dozen chips”; ch. 5: “Where Amodei saw a larger language model as a necessary prerequisite for AI safety research, Altman saw its potential for fulfilling OpenAI’s promise to Microsoft”; ch. 6, Kaplan: “a close friend from grad school and former roommate who worked part time at OpenAI and had led the discovery of scaling laws”; ch. 12, Phase 5: Altman was “casually floating the idea of the $100 billion supercomputer for Phase 5. After the blowout success of ChatGPT, Altman tempered his expectations”; ch. 5: “Not even in Silicon Valley did other companies and investors move until after ChatGPT to funnel unqualified sums into scaling. That included Google and DeepMind”; ch. 5, the date of the Chinese researcher’s analysis: “As ChatGPT swept the world by storm in early 2023”) [In the Sources](https://dryburgh.com/vol1/sources/book#hao-empire-ai)
- Nick Srnicek, *Silicon Empires: The Fight for the Future of AI* (Polity, October 2025; the US edition January 2026, the e-book December 2025; the three dates read on the UK and US pages of the publisher’s distributor, Wiley, on 30 September 2026, where the row gave January 2026 from the US page alone from 18 September 2026) (read directly in chapters 1 to 4 and then whole on 20 September 2026, with the author’s March 2026 book talks and his July 2025 and July 2026 podcast interviews; Srnicek dates the reprieve himself: “In May 2025, the global platforms and semiconductor exporters appear to have won at least a brief reprieve, as the Trump administration decided not to go ahead with the AI diffusion plan”, adding that “Plans remained afoot” to craft a rule meeting both goals and that the tensions were left intact; defence tech investment “has gone from around $2 billion in 2013 to nearly $35 billion in 2023”, the book’s note 135 citing PitchBook; chapter 1’s conclusion, on a bubble deflating: “The risks of a collapse in financial enthusiasm for AI are likely to be borne by pools of capital that are able to weather poor investment choices, whereas the dot-com boom involved many more retail investors unable to bear the shocks of a sudden drop in equity values”, the funding for AI companies “generally not coming from the stock market”; on the hardware’s shelf-life, in the same passage: “four years after the launch of Nvidia’s A100 chip, it remained the dominantly deployed chip in AI research, while the seven-year-old V100 chip remained the third most used chip”, the book’s note 87 citing Air Street Capital’s State of AI Report 2024, p. 94; in chapter 4, ‘China’s Diffusion Strategy’, the Chinese strategy “is necessarily shaped by constraints” in a way the US approach is not; many of China’s public statements about AI strategy “have focused on creating world-leading capacities in innovation”; China’s approach “is not without companies working on a frontier strategy – DeepSeek is also a major example of this”; Zhou Hongyi of Qihoo 360, through the book’s note 46 citing Mary Hui’s A/Symmetric newsletter of 27 April 2024: “it will be as meaningful as us catching up with foreign countries in super AGI.”; on a definition of AGI: “an artificial intelligence that did not require adaptations to the model to make it highly useful to different industries” (ch. 2, Strategies of Capture, its conclusion); chapter 3, “An Interregnum”, on the twelve months of 2024: “Over the course of a single year” … “the military use of AI had been normalized”; on the diffusion rule, Oracle’s words through the book’s note 111 citing Ken Glueck’s Oracle post of 19 December 2024: “history’s worst government technology idea” (chapter 4, Restricting Rivals, where Oracle also calls it “dystopian” and NVIDIA “misguided”); on the commodity future: “may become an indistinguishable commodity, especially if some form of AI winter comes about and the scarcity of compute becomes a glut” (chapter 2, its conclusion); on OpenAI in January 2024, which “quietly rescinded their ban on using AI for ‘military and warfare’ purposes”, the chapter’s note 1 citing The Intercept of 12 January 2024; “but instead via investors like venture capital, Big Tech, and sovereign wealth funds”; ch. 4, “Hemispherical Stacks”: “In what has been called ‘omnialignment’ or ‘polyalignment’, countries are able to partner with multiple great powers at the same time”, a term he reports in his note to Schindler et al.; the Conclusion: “AI can be beneficial to all, but will it be so equally?” and “this book will not pretend to offer answers”; chapter 1, “Control without Ownership”: “The result of all these activities by Big Tech – investments, partnerships, and acquihires – is the construction of a network of dependencies and influence across the AI start-up scene. Beneath the surface level of competition, therefore, lies a deeper system of centralization – a form of ‘control without ownership’”, the phrase credited in his note 72 to Cecilia Rikap; on the literature on general-purpose technologies: “their value is rarely captured by the inventors themselves to any significant extent. Instead, the imitators and the implementers more often benefit from new enabling technologies”; on AI: “value is currently draining down the AI stack towards the more foundational layers of hardware and infrastructure, with existing tech giants such as Microsoft, Google, and Nvidia likely to capture most of the gains”, the cloud strategy capturing the value “by taking a portion of each use across all its varied applications”; on the Manhattan Project model the believers reach for: “the Manhattan Project model aims at nationalistic control over a leading technology,” meaning “much more government control and oversight, with research and development happening in secret” and development “supported regardless of monetary return on investment”; chapter 4: “as Jeffrey Ding has argued, this story of GPTs leading to power transitions in the international system is underdeveloped”; the chapter “arguing that the United States has an innovation focus, while China increasingly has a diffusion focus”, its sections “America’s Innovation Strategy” and “China’s Diffusion Strategy”; “the American Stargate project is oriented towards building out the largest compute clusters to be made available to one single company in the pursuit of one grand model”, against the “Eastern Data and Western Computing” project, “aimed at decentralized clusters that provide collective compute for many organizations”; after Meta’s November 2024 opening, “A few days later, Anthropic announced that they too would allow their models to be used by the military and that they were partnering with the defence firm Palantir”; “As the year ended, OpenAI announced their own partnership with the defence start-up Anduril”, the chapter’s note 4 citing MIT Technology Review of 4 December 2024; on the turn to the military, chapter 3: “But this doesn’t explain the rapidity of the shift nor the fact that all the leading American AI research labs moved in the same direction. Something more is going on.”; on CUDA: “Big Tech companies including Microsoft, Google, and Meta”; chapter 1’s bubble passage, its first reason: “it seems unlikely that the bubble ending will lead to a significant collapse in equity markets or the broader economy. This is first and foremost because, unlike that earlier period, the major tech companies involved in investing billions of dollars are all immensely profitable with or without generative AI”; its third, on the infrastructure: “Moreover, the investments into the infrastructure of AI today appear to be investments that can be deployed for some time and for alternative purposes – like the laying down of terrestrial and submarine fibre cables in the earlier dot-com bubble”, followed by Pichai’s “long useful lives” through the book’s note 85). [In the Sources](https://dryburgh.com/vol1/sources/book#srnicek-silicon-empires)
- Microsoft, “The next chapter of the Microsoft–OpenAI partnership,” blog post, 28 October 2025 (Microsoft “continues to have exclusive IP rights and Azure API exclusivity until Artificial General Intelligence (AGI)”; “Once AGI is declared by OpenAI, that declaration will now be verified by an independent expert panel”; IP rights for models and products “extended through 2032 and now include models post-AGI”; research IP rights “until either the expert panel verifies AGI or through 2030, whichever is first”; “The revenue share agreement remains until the expert panel verifies AGI, though payments will be made over a longer period of time”; Microsoft’s stake in OpenAI Group PBC valued at approximately $135 billion, roughly 27 percent on an as-converted diluted basis). [In the Sources](https://dryburgh.com/vol1/sources/company#msft-openai-oct25)
- Mustafa Suleyman, “Towards Humanist Superintelligence,” Microsoft AI, 6 November 2025 (an essay under the byline of Microsoft AI’s chief executive announcing the MAI Superintelligence Team — a position, not evidence: “In doing this we reject narratives about a race to AGI, and instead see it as part of a wider and deeply human endeavour to improve our lives and future prospects. We also reject binaries of boom and doom”; its energy sentence, “Electricity consumption is estimated to rise 34% through 2050, driven in no small part by the rise in datacentre demand”, is relayed with no source named on the page) (the same essay on its own buying: “We have an exciting roadmap of compute at MAI, with our next-generation GB200 cluster now operational”). [In the Sources](https://dryburgh.com/vol1/sources/company#suleyman-hsi-nov25)

And 46 more: [search the Sources](https://dryburgh.com/vol1/sources).

## Related terms

- [Adoption clock](https://dryburgh.com/vol1/glossary/adoption-clock)
- [Agent](https://dryburgh.com/vol1/glossary/agent)
- [Alignment](https://dryburgh.com/vol1/glossary/alignment)
- [ARC-AGI-3](https://dryburgh.com/vol1/glossary/arc-agi-3)
- [Astra](https://dryburgh.com/vol1/glossary/astra)
- [Automated AI researcher](https://dryburgh.com/vol1/glossary/automated-ai-researcher)
- [Benchmark](https://dryburgh.com/vol1/glossary/benchmark)
- [Brake, the](https://dryburgh.com/vol1/glossary/brake-the)
- [Capability clock](https://dryburgh.com/vol1/glossary/capability-clock)
- [Chain of thought](https://dryburgh.com/vol1/glossary/chain-of-thought)
- [Checkpoint](https://dryburgh.com/vol1/glossary/checkpoint)
- [Critical](https://dryburgh.com/vol1/glossary/critical)

From the glossary of The AI Compute Race.

- [Campuses](https://dryburgh.com/vol1/ledger)
- [Tests](https://dryburgh.com/vol1/tests)
- [Scorecard](https://dryburgh.com/vol1/scorecard)
- [Data](https://dryburgh.com/vol1/data/releases)
- [Chapter by chapter](https://dryburgh.com/vol1/companion)
- [The record at a glance](https://dryburgh.com/vol1/observatory)
- [The machine](https://dryburgh.com/vol1/machine)
- [One campus, four futures](https://dryburgh.com/vol1/hyperion)
- [Glossary](https://dryburgh.com/vol1/glossary)
- [People](https://dryburgh.com/vol1/people)
- [Sources](https://dryburgh.com/vol1/sources)
- [Notes](https://dryburgh.com/vol1/notes)
- [Where the book was wrong](https://dryburgh.com/vol1/corrections)
- [Where its sources were wrong](https://dryburgh.com/corrections?book=vol1)
- [The Paper](https://dryburgh.com/vol1/paper)
- [Consent](https://dryburgh.com/vol1/consent)
- [Who builds and pays](https://dryburgh.com/vol1/parties)

*The AI Compute Race: Five Races, One Machine* is a book by Lee S. Dryburgh, the first volume of THE COMPUTE ECONOMY, published by Dryburgh, an imprint of Contestra OÜ, Tallinn, Estonia; its record closes on 8 November 2026.

[About the book](https://dryburgh.com/vol1)[Tell me when it is out](https://dryburgh.com/vol1#notify)