Fri, 20 Dec 2024 22:00:00 +0000
During the upcoming holiday period referenced in this episode (late December 2024), three or four members of the All-In podcast "besties" group will physically meet and ski together at the same location.
“3 or 4 besties will be skiing together.”
No independently verifiable record of a specific All-In hosts group ski trip during this holiday period was found.
Aaron Levie will be invited back onto the All-In podcast within approximately two weeks of this recording (early January 2025) to discuss the hypothetical impact of allocating 5% of Box’s treasury to Bitcoin.
“Just buy put 5% of the Treasury 30 million in Bitcoin. And then we'll invite you back in two weeks. We'll see what happens okay.”
No specific, verifiable record confirming Aaron Levie's return to the podcast within exactly two weeks of this recording was found.
David Sacks will return as a host on the All-In podcast in future episodes, and thereafter the show will use a rotating "fifth seat" format featuring various friends of the podcast and newsmakers as recurring guests.
“And Sachs will be back. Sachs will be back and we will be rotating the fifth seat amongst, uh, you know, friends of the pod and newsmakers.”
David Sacks did return to occasional All-In appearances, and the podcast has continued rotating a fifth-seat format with various guests and newsmakers through 2025-2026.
From this point forward (post-December 2024), the combination of large language models and social media will enable the U.S. public to rapidly analyze long bills and communicate preferences to representatives, resulting over time in a noticeably more active and responsive form of U.S. governance, where major legislation can be stopped or advanced based on rapid, internet-coordinated public feedback.
“The bigger issue is going forward, you will have the ability to... then to put it in a digestible format that normal people can consume. Then all you'll have to do is just connect the dots and tell your congressman or congresswoman that you like or dislike this thing, and what you're going to see is a much more active form of government.”
AI tools for bill analysis and public engagement did proliferate through 2025-2026, and some legislation did face heightened public scrutiny via social media, but a clearly measurable shift toward more 'active' governance overall is difficult to confirm.
By the year 2030, there will be approximately 100,000 operational eVTOL "flying cars" in China being used to transport people (i.e., active passenger-carrying air taxis), on top of an already large drone-delivery industry.
“The drone delivery business in China is already $30 billion a year. And they're also launching a pretty significant fleet of what we would call kind of evtols or flying cars. The expectation is that by 2030, there'll be 100,000 flying cars moving people around in China.”
The 2030 timeframe for 100,000 operational Chinese eVTOL flying cars has not yet arrived, so this cannot be evaluated.
xAI’s GPU cluster will scale from roughly 100,000 GPUs to about 1,000,000 GPUs within approximately one year from this December 20, 2024 episode (i.e., by late 2025).
“the fact that they were able to get 100,000 to work, as you know, in one contiguous system and are now rapidly scaling up to basically a million over the next year.”
xAI's GPU cluster scaled from roughly 100,000 toward the 500,000-plus range within about a year, on a trajectory toward the stated 1 million GPU goal.
As AI model markets mature, the per-token price for AI inference will converge to roughly the underlying compute cost plus only about a 10–20% margin, leaving little additional economic rent for model providers on raw token usage.
“what we should expect is that the price of a token in AI land, you know, basically will be whatever the price of running the computers are. And maybe with like a, you know, plus ten, 20% margin.”
AI inference token pricing has fallen dramatically toward compute cost through 2025-2026 amid intense competition, broadly consistent with the prediction, though provider margins vary and haven't uniformly converged to just 10-20%.
Over time, both open-source and closed-source AI model providers will be forced by competition and open-source alternatives to drive their per‑token API pricing effectively to (near) zero above compute cost for large enterprise customers.
“I think what Aaron is saying here, let me let me maybe try to frame it. I think what he's saying is they'll be open source models, they'll be closed source models. But the price that Aaron or me or anybody else pays, these model makers will effectively go to zero.”
Token pricing has fallen sharply due to open-source competition and efficiency gains, though large enterprise API pricing has not reached literally near-zero margins above compute cost for all providers.
Global annual spending on software and software-related activities, currently around $5 trillion, will shrink by roughly an order of magnitude over time, to about $500 billion per year, as AI drastically lowers the cost of producing and delivering software.
“You know how much the world spends on software and software related things. Every year it's about $5 trillion... I, I'm pretty sure that the market here shrinks by an order of magnitude. And instead of fighting over 5 trillion, I think we'll be fighting over 500 billion.”
Global software spending has not been confirmed to have shrunk by an order of magnitude to roughly $500 billion; this remains a longer-term speculative thesis not yet borne out by 2026 data.
Within about seven years from this episode (by around 2031–2032), it will be standard practice for organizations to use AI systems that can automatically design and generate production software, including security, permissions, and regulatory-compliance logic (e.g., for a bank), based on natural-language instructions and analysis of existing software.
“there's no reason that in seven years that is not the standard, is your that I don't have the ability to say, go look at all the software that's out there in the world today. So that help me build a tool that meets compliance standards, that meets all of my security standards.”
The roughly seven-year horizon (to around 2031-2032) for standard AI-driven compliant software generation has not yet elapsed.
It will take roughly another decade from now (through the mid‑2030s) of AI and software tooling evolution before fully AI-generated software systems become sustainably deployable at high quality in heavily regulated and mission‑critical environments.
“We've got another decade of evolution here to make these things sustainable and have a high quality.”
The roughly decade-long horizon (to the mid-2030s) for AI-generated software becoming sustainable in mission-critical regulated environments has not yet elapsed.
Google will eventually introduce an ad-supported version of its Gemini AI assistant, with an option for users to pay to remove ads (a freemium ad/no‑ad business model for Gemini).
“I think you'll end up having a Gemini that has ads eventually, and you could pay and have no ads, or you could not pay and have no”
No confirmed launch of an ad-supported freemium tier for Google's Gemini assistant was found as of mid-2026.
Within three years of this episode (by late 2027), OpenAI will no longer be a top‑two AI model provider by market position; instead it will rank roughly 3rd–5th, with Google Gemini, Meta, and xAI leading it in the market.
“I'm saying it right now. I think we've hit peak OpenAI in the market. I think they're going to be the number 3 or 4 player. I think Gemini, meta and Xai are going to lead them. If we're sitting here in three years, I think OpenAI's number three, 4 or 5, not 1 or 2.”
OpenAI remained the clear leading AI lab by usage and revenue through 2025-2026 (with ChatGPT retaining the largest consumer AI user base and revenue run rate around $20B+), not slipping to third through fifth place as predicted.