Alphabet chief executive Sundar Pichai said Google Cloud is giving up some near-term sales so Google can keep enough compute for frontier AI, in a Big Technology Podcast interview with Alex Kantrowitz published on 20 October 2026. “Our cloud business is at any given moment in time forgoing short-term revenues because we are making compute trade-offs to support frontier AI development,” Pichai said, according to the transcript Big Technology published.
He also said Gemini 4 Argon reached almost 25 trillion tokens in its first 72 hours, “about three times what Gemini 3.0 had in the first 72 hours,” and that Google is “capacity constrained as we are serving it.”
Business Pill · GPUS
A short explainer on why AI runs on GPUs. It teaches the general idea only and says nothing about any company in this story.
The key insight: As we read it, Pichai described compute as the constraint Google allocates first to its own frontier models and then to its businesses, including the Cloud customers who would otherwise buy it.
Asked how Google balances Cloud’s demand to sell compute against researchers’ need for it, Pichai said: “Ground zero for us is making sure we are investing in the compute that’s needed to be at the AI frontier.” He described the order of allocation: “The way I think about compute allocation is making sure Google DeepMind, both for research and frontier model development, is cutting edge in terms of compute access. After that, we plan for all our businesses.”
He said frontier compute is “non-negotiable, and everything else works on top,” and that lead times are long: “The compute we have today is a function of decisions I made in 2024.”
Kantrowitz relayed Ben Horowitz’s account that Andy Jassy had said compute was sold out until 2028. Pichai answered: “If today a customer is coming out and asking us for compute in ’27, that is incredibly difficult to do.” He added that Google had “leased some short-term compute to accommodate some customers” last year.
Gemini 4 Argon tokens in the first 72 hours after launch, as Sundar Pichai told the Big Technology Podcast: almost 25 trillion, about three times Gemini 3.0’s first 72 hours. The Gemini 3.0 circle is our arithmetic from his ‘about three times’; Pichai did not give that figure.
What Was Posted on X
New: I spoke with @sundarpichai about Google’s commitment to the AI frontier, how close we are to RSI, balancing DeepMind’s compute needs vs. Google Cloud profits, and much, much more.
Full Big Technology Podcast episode here –>
Chapters:
0:00 Introduction 1:14 Googleโs Frontier AI Strategy 3:47 Do Frontier Models Really Matter 6:35 The Future of AI Agents 7:39 Recursive Self-Improvement 12:58 World Models and the Path to AGI 14:56 Googleโs Compute Dilemma and AI Infrastructure 19:36 Is AI Overbuilt 23:35 The Future of the AI Interface 31:42 Googleโs Answer to Muse and Dots 35:03 Apple, Space Data Centers, and AI Coding 41:42 Optimism and AIโs Future
Alex Kantrowitz, host of the Big Technology Podcast, posted the episode on X on 20 October 2026. He wrote that he spoke with Sundar Pichai about Google’s commitment to the AI frontier, how close we are to RSI and balancing DeepMind’s compute needs against Google Cloud profits, among other topics. The quotes in this piece come from Big Technology’s published transcript.
Google VP Bill Jia says AI demand will still go 10x, 100x, 1,000x, and whoever skips the investment now cannot serve it when rivals can
โYes, I think there are mainly two points, and actually we touched on both of them just now. The first point is that AIโs demand is still enormous, beyond compare. In the future it will still go 10x, 100x, 1,000x.”
“If AI demand in the future is going to be very large, and you know itโs going to grow rapidly, then doing the investment now is what should be done, right?”
“Because if you and I donโt invest, others will invest. Then later, when the demand comes up, you yourself have no way to accommodate the demand while everyone else can accommodate it. Thatโs a big problem, right So this has to be done first. This is definitely correct.โ _______ Link and breakdown of the Google VP interview: https://firesidealpha.substack.com/p/googles-bill-jia-vp-of-engineering
Fireside Alpha posted quotes from an interview with Google VP Bill Jia, in which Jia says AI demand “will still go 10x, 100x, 1,000x” and argues that “if you and I don’t invest, others will invest”, leaving a company unable to serve the demand when rivals can. The figures are Jia’s projection, not Google data. Why it matters: it is a Google VP’s version of the build-ahead case behind the Cloud trade-off in this piece.
Google CEO Sundar Pichai:
“RSI is a critical contributor as we develop Argon, but to be very clear, we are far from the end state of RSI”
The right way to describe it today: you have RSI loops alongside AI researchers, with agents helping researchers in an automated way to build these models
Haider (@haider1) posted on 25 October 2026 a remark attributed to Google CEO Sundar Pichai: that RSI is a critical contributor as Google develops Argon, but that Google is far from the end state of RSI.
Is AI Overbuilt?
Pichai rejected the comparison with the fibre bubble on his own usage: “this is not like fiber. It’s not dark. We are using it. We are capacity constrained on a daily basis.”
He did not rule out an overbuild across the economy. “If your question is, can there be an overbuild? Yes, because you can have some constrained components,” he said, listing land and “2,000 components or more” and adding: “There could be external shocks to the economy. There are macro risks.” Asked about Google’s own plan, he said: “am I confident that the compute we are planning for, we will have uses for it given enough time? Absolutely, yes.”
Muse, Agents, Anthropic and Apple
On Meta’s Muse. Asked where Google’s equivalent of Muse or Dots is, Pichai said: “Yes, we are going to bring personal agents to our consumers. It’ll work in the Gemini app. It’ll work in Google Search. It’ll work in Gmail.” He said it will be “powered by Argon.”
On Anthropic. Pichai said: “We’ve always tried to identify and partner with companies which would be at the cutting edge of technology. We’ve done that with SpaceX. We’ve done that with Anthropic.”
On Apple. On licensing Gemini for Siri: “There’s nothing specific to comment, but it’s a partnership. It’s working well.”
Update, 21 October: The Information Reports Gemini 4 Carbon Is in the Works
The Information reported in a post on X on 20 October that “Google is developing a model codenamed Gemini 4 Carbon that appears to improve on last week’s Gemini 4 Argon release, as competition with Anthropic and OpenAI intensifies.” The full article is behind The Information’s paywall and we could not read it, so this box goes no further than that post.
Other posts on X claim Carbon has already entered early access with select partners. Those are relays that do not cite a named source, and The Information’s post does not mention early access, so we treat that claim as unconfirmed.
Update (25 Oct): what Pichai said about recursive self-improvement. In the same interview, Pichai linked frontier models to recursive self-improvement. “The frontier models will be better at developing the next set of models in an era of recursive self-improvement,” he said, according to Big Technology’s transcript. He added: “Internally, we have definitely seen a steep point of inflection by deploying Argon to our developers.”
On 25 October, Haider (@haider1) posted a clip from the episode and quoted Pichai as saying RSI “is a critical contributor as we develop Argon” but that “we are far from the end state of RSI.” The part of the transcript we could read does not include that passage, so that wording comes from Haider’s post.
Update (25 Oct): another lab leader on RSI. Anthropic co-founder Jack Clark wrote in his Import AI newsletter on 12 October: “I now think it’s more likely we’ll see full RSI become possible in 2027, whereas before I thought it was more likely in 2028.”
Clark tied the change to TasteVal, a benchmark from P-Zero Research, which reports that the experimental research taste of frontier models “has doubled every 3.0 months since December 2025.” The top model on that benchmark is Anthropic’s Opus 5.5, which The Decoder noted on 25 October adds a layer of bias to the forecast. For Clark’s earlier economic case, see our piece on his $44T US GDP scenario.
Update (3 Nov): Pichai on Brockman’s “no interface” AI. A passage from this same interview drew new coverage on 3 November, when outlets including The Hans India reported that Pichai does not fully agree with OpenAI president Greg Brockman on the future of AI interfaces. The remark is not new: it is about 23:40 into the full episode published on 20 October, and Big Technology posted it as a separate clip, “Sundar Pichai: Will AI Kill the Interface?”, on 23 October.
Kantrowitz put to him what Brockman had told him in an interview Kantrowitz published on YouTube on 1 July 2026. At about 2:00 in that episode Brockman said that “the answer is you want almost no interface, you want no product.” Pichai agreed that computing will adapt to people rather than the other way round, and that an AI with context that is “persistently running somewhere” makes sense to him. Then he added: “That doesn’t mean it won’t have an interface, however, to me.” He went on: “But there are going to be moments you will need an interface,” and gave watching a video the AI made for you as an example. His test was that “you are not doing the hard work adapting to whatever the interface is.”
There is money behind the screen. Alphabet’s third-quarter 2026 results release (SEC 8-K, Exhibit 99.1) puts Google Search & other at $63.35 billion of $127.58 billion total revenue. By our arithmetic that is 49.7% of revenue, down from 55.3% a year earlier ($56.57 billion of $102.35 billion), as Google Cloud doubled to $30.71 billion while Search grew 12%. Google advertising as a whole was $82.01 billion, 64.3% of revenue, by our arithmetic. Later in the clip Pichai described Google’s planned persistent agent as “a headless service” that you can reach “in search,” in Gemini and in Workspace.
Quotes come from YouTube’s automatic captions of the Big Technology videos, checked against the clip. Times are approximate. Asked in the clip whether an agent would take a cut when it suggests a product, Pichai said agents will need to fulfil “commercial needs” and that “there’s going to be some mechanism by which it gets fleshed out.”
Big Technology’s 23 October clip of the interface exchange with Sundar Pichai.
Update (9 Nov): the coding numbers from the same interview. About 38 minutes into the episode, Kantrowitz asked what share of Google’s code is now written by AI, recalling the 25% figure Pichai had given earlier. Pichai said the number is “like meaningless in the sense that every part of code we now touch is influenced by AI deeply,” and that Google now tracks other metrics. “In the four weeks since we have launched Argon, we have seen a 30% increase in CLs and, more importantly, almost a similar increase, you know, one-third decrease in like rollbacks,” Pichai said, according to our own transcript of a 58-second clip posted on X on 24 October; Mint reported the same figures. CLs, or changelists, are Google’s internal term for code submissions; a rollback reverses a change that was already submitted. Pichai added that people are “writing code more productively and better code with it.” The figures are Pichai’s. The remarks resurfaced on X on 9 November.
Our Muse piece reported 10.2 million US downloads in Muse’s first 40 days, according to TechCrunch, citing Sensor Tower, the product Kantrowitz asked Pichai about. And our piece on Anthropic’s draft prospectus reported that Anthropic is on the hook for $111 billion of Google cloud computing contracts, the Wall Street Journal reported from a draft; the S-1 Anthropic filed on 26 October puts the Google commitment at at least about $114.5 billion. As we read it, that sets Pichai’s description of Anthropic as a partner beside a large committed buyer of Google Cloud compute.
How Alphabet Makes Money, Visualized
Pichai’s trade-off sits inside Alphabet’s income statement. Our video walks through how Alphabet makes money, from Search and YouTube to Google Cloud.
How Alphabet (Google) Makes Money: 2025 Income Statement, Visualized (The Business Engineer on YouTube)
The Structural Read
As we read it, the interview names an order of priority: Google DeepMind first for research and frontier model development, then the businesses, which plan against demand signals and committed contracts.
As we read it, Pichai’s answer on overbuild has two halves: Google is capacity constrained today, and an economy-wide overbuild remains possible if components bottleneck or external shocks arrive.
As we read it, the token figure is the demand side of the same answer: Pichai said Argon reached almost 25 trillion tokens in 72 hours and that Google is capacity constrained serving it.
Sundar Pichai, Big Technology Podcast, 20 October 2026
“Our cloud business is at any given moment in time forgoing short-term revenues because we are making compute trade-offs to support frontier AI development.”
Three Implications
ALLOCATION Pichai said compute for Google DeepMind’s research and frontier model development comes first, and that Google plans for its businesses after that.
LEAD TIMES Pichai said the compute Google has today is a function of decisions he made in 2024.
DEMAND Pichai said a customer asking Google for compute in 2027 today would find that incredibly difficult to accommodate.
The framework puts it this way: “Training builds the brain. Inference is the brain working.”
As we read it, Pichai’s trade-off fits that split on his own description: compute held for frontier model development builds the next brain, while Cloud customers buy compute to run theirs. The framework was published in March 2026 and does not discuss this interview.
What Is Not Established
Pichai gave no figure for how much revenue Google Cloud has forgone, and no split of compute between DeepMind and Cloud customers. The token count is his figure and we have not checked it independently. The interview was recorded at Google’s Mountain View headquarters, Big Technology wrote; the quotes here are from its lightly edited transcript.
Business Engineer Framework
The State of the Inference Economy
A Business Engineer map of how AI moved from the training era to the inference era, and who pays for it.
Sundar Pichai told the Big Technology Podcast that Google Cloud is forgoing short-term revenues to support frontier AI development, that Google is capacity constrained serving Gemini 4 Argon, and that AI infrastructure is “not like fiber.” As we read it, the interview puts a name on a trade-off every hyperscaler faces: compute sold to customers today versus compute held back for its own models.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. This piece is built on the transcript of the Big Technology Podcast interview with Sundar Pichai as published by Big Technology, Alex Kantrowitz’s post on X, and our own earlier coverage, which cites its sources. Nothing here is legal, financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
