Palihapitiya asserts energy control will enable equity participation in AI companies, elevating suppliers from low-margin to high-leverage positions.
“What's a different example? Brad mentioned this before, but energy will be the gating item beyond a shadow of a doubt.”
Palihapitiya argues AI infrastructure OpEx over twenty years equals CapEx, making total spending far larger than reported trillion-dollar figures.
“That's just the CapEx. Then you have to think of what the OpEx is. The OpEx is probably the same multiplied by the useful life, which is probably twenty years.”
Palihapitiya explains OpenAI's Korea deals secure forward HBM memory capacity that Altman can allocate for leverage and equity participation.
“He's buying forward capacity on HBM because now he can allocate that share. So when I saw that deal and I saw that equity, it reminded me of this Rothschild quote.”
Palihapitiya says Google stock doubled on Gemini success, justifying $50 billion spend for trillion in market cap.
“And so if you wanna make Gemini even more incredible, just get pour another billion users into it. And if that costs you 50,000,000,000, it's okay because you'll make a trillion of market cap.”
Palihapitiya says distribution matters most in AI markets, favoring Google and Meta despite Meta being behind.
“And, ultimately, what we are learning, especially as all these markets converge, is that distribution still matters a ton, which favors Google. It favors Meta, although Meta's quite behind.”
Palihapitiya argues physical dollars preserve privacy through fungibility that digital transactions lack.
“The great thing about the US dollar is when you get a dollar and you put it in your pocket, the physical dollar bill, it is completely fungible.”
Palihapitiya predicts AI will create three geopolitical tiers with only China and America exporting intelligence.
“In bucket number one, you're an exporter of that intelligence. And I think right now, steady state, it's gonna be China and America.”
Palihapitiya argues tech companies must deliver measurable benefits to tens of millions of Americans.
“We need to self organize better, and we need to be more on the forward foot. We need to start doing things that are practically measurable by tens of millions of American citizens.”
Palihapitiya frames China competition as existential and says winning requires keeping it focused on AI and economics.
“The war with China is existential because if we can keep it on the battlefield of AI and intellectual and economic prowess, we have a very good chance of winning.”
Palihapitiya cites Carnegie building 2,500 libraries as a model for tech companies to provide tangible benefits.
“Andrew Carnegie built 2,500 libraries. The idea was as he built the railroads, you're going to scale GDP. You're going to scale education and knowledge.”
Palihapitiya predicts a silicon renaissance where small teams building decode silicon will make fortunes over 10-20 years.
“If you look at the Trail of Breadcrumbs, my belief, there's gonna be a renaissance in silicon. Young, small teams building decode silicon can make a fortune over the next ten to twenty years.”
Palihapitiya predicts marrying next-gen systems to open source will cut AI costs by 90 percent.
“If you take these next generation systems and you marry them to open source, you're gonna cut the cost of AI by 90%.”
Palihapitiya explains forking open source AI from Moonshot becomes worthless due to version drift.
“Because there's so much investment by Moonshot, that's the company that makes Kimikay two in that example, that you'll have so much drift in one version, your fork will be worthless.”
Palihapitiya says Kimi K 2.5 is a trillion parameter mixture of experts model.
“So this is why it's so important because it was incredibly profound. It's a trillion parameter mixture of expert model.”
Palihapitiya argues open source AI gives sovereignty to companies and especially countries through transparency.
“But open source is important because it's transparent. It gives everybody their own sovereignty, whether you're a company and, frankly, really more importantly, whether you're a country.”
Palihapitiya predicts enterprises will face a critical decision between cloud convenience and on-prem data security due to AI.
“When they realize that that is a problem, the enterprise will have to decide, do I just give up and keep running all of this stuff in the cloud in a shared experience, or do I bear the incremental cost of running this stuff in a more coordinated manner that I control on prem?”
Palihapitiya uses GEICO actuaries as an example of why companies cannot use open LLM instances with proprietary data.
“Do I, if I'm GEICO, want to have all of my actuaries using all of our proprietary private and confidential data on risk pricing in an open instance of an LLM?”
Palihapitiya says his company now sets token budgets for developers due to AI costs escalating rapidly.
“We're getting to a place where we have to basically now say, what is the token budget that we're willing to give our best devs?”
Palihapitiya says his costs are increasing three times every three months while revenues are not.
“But the problem is that my costs are going up three x every three months. My revenues are not. Token use is very addicting.”
Palihapitiya observes automated AI agent-based hedge funds are emerging to replace traditional firms like Citadel and Millennium.
“If you go on x, I've gone down the automated trading rabbit hole. So I don't know if you guys know, but there are these incredible, young, vibrant projects that are basically replacing a citadel, replacing a millennium, and they're building these automated agent based hedge funds that are transacting across all kinds of markets all the time.”
Palihapitiya questions whether vulnerabilities can be fixed in months when capitalism drives forward regardless
“What do you think they can actually accomplish in two months? Do you actually think that if there's these vulnerabilities, it's all gonna get fixed?”
Palihapitiya argues GPT-2's 2019 doomsday rollout by Dario was theater that ended as a nothing burger
“In February 2019, when Dario was still at OpenAI, they did the same thing with GPT two. That was a 1,500,000,000 parameter model, which sounds like a total fart in the wind in 2026.”
Palihapitiya says Jane Street did a billion dollar investment and $6 billion compute deal, showing massive compute constraints.
“I don't know if you guys saw recently, but Jane Street did billion dollar investment in, essentially, a neos neoscaler and then also did a $6,000,000,000 compute deal with them.”
Palihapitiya says his team finds Codex better than Anthropic for complex coding tasks, Claude better for daily work.
“As it relates to complex long horizon coding tasks, what I can tell you from my team at eighty ninety is codex is better generally than anthropic.”
Palihapitiya claims AI company forecast misses are entirely due to power supply constraints, not demand issues.
“The reason that these folks may miss a number or a forecast have nothing to do with demand. It is entirely 100% due to the supply of the power necessary to generate the output token.”
Palihapitiya reveals Microsoft's Three Mile Island energy deal costs more than double the prevailing spot rate.
“Do you know what their forward purchase agreement was? It was for more than two x the prevailing spot rate for energy. More than two x.”
Palihapitiya notes less than half of announced energy infrastructure projects are actually being constructed.
“People have announced all these projects, Jason, but less than half of it is actually being built.”
Palihapitiya suggests investing in companies receiving the trillion dollars annually spent by hyperscalers, claiming they are underpriced.
“Like, a trillion dollars a year going out of the hyperscalers. Where is it going? Just follow those dollars and buy those companies because those companies are already underpriced.”
Palihapitiya predicts hyperscalers will become heavily leveraged industrial businesses within five years due to infrastructure spending.
“They'll have all kinds of different vehicles and term loans and revolvers and all of this stuff. And so they're gonna look like this big bulky industrial business in five years.”
Palihapitiya reveals the foundational AI paper is titled Attention is All You Need.
“You know what the name of the critical paper in AI is? Like, when you go back to, like, the Magna Carta of AI, do you know what it's called?”
Palihapitiya cites Verizon CEO's forecast that 30% of white collar jobs will disappear by 2030.
“He put out this very public forecast. You know, very smart guy, well regarded in business. And I think he said something like 30% of all white collar jobs will be gone by 2030.”
Palihapitiya reveals Andre Karpathy spent about a quarter of his time hand-labeling Tesla video data in 2016-17.
“I don't know if he said this publicly or not, but where he spent a portion of his time, I wanna say a quarter of his time labeling data.”
Palihapitiya asserts AI has not delivered measurable productivity gains yet despite token consumption in enterprises.
“But underneath that is not AI because we know this. It hasn't done anything measurable yet at the end consumption of these tokens.”
Palihapitiya predicts silicon and software innovations will drop AI training costs from $10B to $10M runs.
“At the scale of what they're trying to do, those kinds of innovations are going to make the cost of model training so much cheaper that it's like, why would we stick to the $10,000,000,000 training runs when we can have the $10,000,000 training runs?”
Palihapitiya cites Rogo eval showing top frontier models now indistinguishable within three-tenths of a percentage point.
“At the top of the leaderboard, Opus four seven, GPT five five, Sonnet four six appear almost indistinguishable, separated by less than, in this case, you know, three tenths of a percentage point overall.”
Palihapitiya reports Fortune 20 company spent $200M on AI tokens seeking $1B savings but got minimal results.
“CEO asked for a billion in AI generated OpEx savings at the beginning of the year. So we're six months in. The team has spent $200,000,000 on tokens and with minimal results.”
Palihapitiya argues frontier AI labs use game theory to absorb capital and shape regulation by creating less capable oversight bodies.
“And how do you create an oversight body that is less capable and intellectually aware as you are about the actual details?”
Palihapitiya predicts hyperscalers will pitch governments on becoming AI gatekeepers using KYC and audit trails to control access.
“And the very simple opportunity is to convince governments all around the world, not just America, that they should be the gatekeeper.”
Palihapitiya argues neoscalers cannot match hyperscalers' decades and trillions invested in infrastructure, making their data centers obsolete.
“It's an impossibility, Jason. So you can take all of those data centers off the map.”
Palihapitiya says Silicon Valley has decayed from humble misfits building important things in 2001 to less serious endeavors.
“We built important things. Then we veered away from that, and we started building less important things.”
Palihapitiya characterizes Dario's worldview as distrusting all actors except those who reason like him under his designed rules.
“The list of trustworthy ones is conspicuously short, and it has a suspicious tendency to resolve towards people who reason the way I do operating under the rules I help design.”
Palihapitiya contrasts AI expertise with mere knowledge using transistor design as concrete example.
“Now though, what you can say is design me a transistor, and you have a spec that you can now mail to somebody and they can actually tape that out for you and you can get back in the mail an actual functional thing.”
Palihapitiya accuses frontier labs of timing apocalyptic AI rhetoric to coincide with fundraising cycles.
“You would see people talking about the insanely apocalyptic characteristics of AI leading up to a model release, which usually was leading up to a fundraising.”
Palihapitiya says Meta profoundly fumbled opportunity to be third leg of AI stool alongside closed and open source.
“I mean, they have completely fumbled it. I don't know the organization well enough, nor do I understand the political dynamics to understand why they failed as miserably as they have. But they've profoundly failed.”
Palihapitiya describes AI as putting all expert expertise in everyone's pocket for any use.
“I can create this thing, this magical thing that will give you entirely all the expertise of all the great experts and thinkers in any domain that have come before you.”
Palihapitiya defines real equality as AI making everyone as capable as the smartest, richest, most connected.
“What's real equality? It's when you are as smart as the smartest human. You are as capable as the richest human. You're as ingenious as the most connected human.”
“And the good news is we were able to hold the line in some of these key places, but just barely.”
“They're six months behind on the model, and they're twenty four months behind on the silicon, yet they're only a few months behind in total. So what game are we playing? This is insane.”
“If I had to make a guess, I think what's gonna happen is that since 2021, about 40% of all data centers get contested. Right?”
Palihapitiya predicts Anthropic will face difficulties because they lost trust by disrupting their customers.
“The difficulty that Anthropic is going to face is that I think what Saks said is true, that they have lost this fundamental trust about being able to stay within their sandbox.”
Palihapitiya reports their harness with open source models achieved 16.4x cost savings versus Anthropic alone.
“But if you wrap the open source model with our software factory, it was 16.4x cheaper.”
Palihapitiya claims California's management is worse than Illinois when compared to Florida's performance.
“And then just ask your favorite AI, compare California on the spectrum of Florida versus Illinois. And what you'll find is California is even worse than Illinois in most of those comparisons.”
Palihapitiya claims frontier labs alternated apocalyptic AI rhetoric with promotional narratives timed to their fundraising cycles.
“You would see people talking about the insanely apocalyptic characteristics of AI leading up to a model release, which usually was leading up to a fundraising.”
Palihapitiya frames AI as the culmination of all prior technological infrastructure from internet to Google.
“Or you stand on the shoulders of all these people that have come before you. Think of what has had to happen for AI to happen.”
Palihapitiya defines real equality as AI making everyone as capable as the smartest, richest, most connected person.
“What's real equality? It's when you are as smart as the smartest human. You are as capable as the richest human. You're as ingenious as the most connected human. That's possible today.”
Palihapitiya says countries will build sovereign AI stacks with open source models even if only 95% as good.
“And if the models are 99% as good or 95% as good, there's going to be a claim that some countries make, which is it's just good enough.”
Palihapitiya concludes actual AI ROI from publicly available data is between zero and two percent of EPS growth.
“And so the answer, far as all publicly available data, was that the actual ROI was somewhere between zero and two percent.”
Palihapitiya reveals AI's actual contribution to S&P 493 EPS growth is only 9%, not 50% as initially calculated.
“I asked a different question, which is then what was the EPS growth of the S and P four ninety three? And the answer was 9%.”
Palihapitiya predicts CEOs will discover unexpected OpEx misses from runaway AI token spending they don't know about.
“for us. The CEOs and the CFOs, in my opinion, probably have no idea how much token maxing is going on inside of their organizations.”
Palihapitiya accuses Anthropic of shipping tokenmax with Opus 5, optimized for ARR instead of performance
“They shipped tokenmax. Like softmax but optimized for ARR… https://t.co/smD0eXM637”
Palihapitiya notes Infosys cut revenue forecast as enterprises shift budgets from outsourced hours to controllable AI
“Infosys cut its revenue forecast this week. The reason it gave is that enterprises are moving their budgets out of outsourced hours and into AI they can control. The outsourcing model was built on billing by the hour, which only grows when the work stays slow and manual. So”
Palihapitiya endorses Zuck's essay on whether superintelligence will be centralized or empower everyone
“Zuck’s essay and vision is a really good one. “The defining question of our age isn’t whether superintelligence will exist, but who will have access to it. Will it be centralized and restricted to a few institutions, or will it be a tool that empowers everyone?” “I don’t”
Palihapitiya agrees OpenAI and Anthropic cannibalizing customers is natural as models commoditize
“@itsolelehmann I agree and it’s the natural resting place as the models themselves get commoditized.”
“As of today, let’s assume 99% of all code written on the internet is still human written and 1% is agentic. That means 99% of code is sloppy and error ridden. Humans make errors that agents don’t. As agentic code replaces human code, both the error rates and security https://t.co/mZ2Aa0d1ZS”
Palihapitiya previews upcoming AI efficiency gains that will cut token consumption by 50 to 75 percent for identical tasks.
“What I would tell you is that there are some incredible efficiencies that I think are about to be demonstrated, which effectively cut token consumption by about 50 to 75% for the same task.”
Palihapitiya predicts AI-generated code will eliminate security vulnerabilities by 2028-2030 as models avoid human coding errors.
“These security holes won't exist because the errors that humans make won't be made by these models.”
Palihapitiya highlights DeepSeek completing tasks at 105x lower cost than alternatives
“Why pay $1 when I can pay $105. Said no CFO, CEO or Board Director ever. https://t.co/QoKPFFfORB”
Palihapitiya claims Anthropic's funding mechanics and FUD drove Dario's position on open source more than belief
“In many ways, Anthropic’s funding mechanics - claims made, FUD, post money valuations have dictated Dario’s position more than anything else. Even if he really believed this. https://t.co/tbBmvnCO0x”
Palihapitiya claims modern harness plus open model crushes token consumption while maintaining performance
“A modern harness + open model will crush your token consumption but keep your performance. https://t.co/pcHU8rZY35”
Palihapitiya places Tao and Venice in distributed cloud layer when asked about AI stack positioning
“@heider891 @grok It would sit in the cloud part. But in their case, a distributed cloud.”
Palihapitiya suggests slop cannon behavior develops naturally in teams rather than by explicit decision
“@Vvikramai I don’t think teams make these decisions explicitly. It just sort of develops naturally. There is a ton of slop cannon behavior that hasn’t gotten sorted out yet.”
Palihapitiya praises nbis as the most technical team, citing Arkady favorably
“@ddSuktqYfJ43941 @thetimmorgan They are the most technical of all of them. I like Arkady and his team very much.”
Palihapitiya identifies LPS as fastest path to cash-on-cash returns in AI investing as of August 2026
“Here is my AI investing guide. Sitting here August 2026, my current best thoughts are as follows: 1. LPS (Land Power Shell) is still the most obvious and fastest path to cash on cash returns. Lots of value can be assembled and traded quickly at this layer. And as data centers https://t.co/QSZgT0Gv0S”
Palihapitiya expects hyperscalers to monetize services like KYC around models not tokens, holding 6GW through 2029
“@seagoat73 Hyperscalers want 1000 models to exist and thrive. They will make money from various services around models (like KYC) not just serving tokens energized land will always be valuable.”
Palihapitiya dismisses latency concerns for harness overhead, noting it is not measured in days over long tasks
“@janstevens Why does that matter? The latency, to the extent there is any, isn’t in months…and unlikely even days over long horizon tasks.”
Palihapitiya highlights DeepSeek completing tasks at 105x lower cost than alternatives
“Why pay $1 when I can pay $105. Said no CFO, CEO or Board Director ever. https://t.co/QoKPFFfORB”
Palihapitiya identifies LPS as fastest path to cash-on-cash returns in AI investing as of August 2026
“Here is my AI investing guide. Sitting here August 2026, my current best thoughts are as follows: 1. LPS (Land Power Shell) is still the most obvious and fastest path to cash on cash returns. Lots of value can be assembled and traded quickly at this layer. And as data centers https://t.co/QSZgT0Gv0S”
Palihapitiya confirms Claude wrapped in Software Factory is 1.4x cheaper and 1.5x faster, open models 16.4x cheaper
“This is accurate. That said, we are currently working through a rigorous experiment testing closed and open models with and without Software Factory. Early data is super fascinating and backs up the experiment below. https://t.co/vqWV7ISaeX”
Palihapitiya calls AI models a commodity because many providers achieve similar results simultaneously
“Guess this commodity. Why do I say it’s a commodity? Because the chart shows many providers accomplishing very similar things all at the same time. https://t.co/r0c1bTylIP”
Palihapitiya says opposite of Roose, model capability gains fully priced in and cost going to zero
“@kevinroose Opposite. It’s all priced in. The cost is going to zero. All powers will be unlocked.”
Palihapitiya argues you cannot build a monopoly on intelligence by looking at humanity
“Turns out you can’t build a monopoly on intelligence. Which we would have known simply by looking at humanity.”
Palihapitiya says opposite of Roose, model capability gains fully priced in and cost going to zero
“@kevinroose Opposite. It’s all priced in. The cost is going to zero. All powers will be unlocked.”
Palihapitiya argues AI singularity loop where AGI designs smarter AI repeatedly is now visible in results
“The AI Singularity The argument goes like this: 1. Humans build an AGI. 2. The AGI becomes good at AI research. 3. It designs a smarter AI. 4. That smarter AI designs an even smarter AI. 5. The cycle repeats faster and faster. Looking at the results and capabilities from the”
Palihapitiya notes identical prompts produce vastly different token consumption, questions per-token pricing
“Imagine if you priced by the token…hey! Wait a minute!! https://t.co/JPxLEL0yod”
Palihapitiya predicts bottoms-up GTM will be worst strategic decision of past decade as AI stamps out tools
“In a world of agents + harness + application, bottoms up will turn out to be the worst strategic GTM decision of the past decade. Over the next few years, AI will stamp out clone after clone of various bottoms up tools, meanwhile this same tool sprawl will be viewed as part of”
Palihapitiya predicts bottoms-up GTM will be worst strategic decision of past decade as AI stamps out tools
“In a world of agents + harness + application, bottoms up will turn out to be the worst strategic GTM decision of the past decade. Over the next few years, AI will stamp out clone after clone of various bottoms up tools, meanwhile this same tool sprawl will be viewed as part of”
Palihapitiya says ByteDance training 10 trillion parameter model with zero distillation proves value destruction
“And if reports are accurate, with zero distillation to jumpstart it. Thus proving many things that are more value destructive than value accreting… https://t.co/4FQqd6xZXB”
Palihapitiya contrasts technology people who understand hardware versus shorts operating on non-technical priors
“The market is reducing itself to: Technology people who knows how this stuff works Vs Everyone else who has no idea and so gets short because it violates their non-technology priors https://t.co/H4phQiYKDH”
Palihapitiya contrasts technology people who understand hardware versus shorts operating on non-technical priors
“The market is reducing itself to: Technology people who knows how this stuff works Vs Everyone else who has no idea and so gets short because it violates their non-technology priors https://t.co/H4phQiYKDH”
Palihapitiya notes Perimeter changing cancer surgery with intraoperative AI model decisions
“It took longer than expected but Perimeter is changing the nature of cancer surgery… https://t.co/z3OTXv3ElL”
Palihapitiya says harness should give contextual access without model dependency to preserve memory
“@dragosroua If the harness isn’t dependent on a specific model and can give access to it contextually, then yes.”
Palihapitiya says third-party harness plus model interchangeability changes enterprise risk profile
“This is a really good development. As model costs converge, the next likely step will be the use of independent 3rd party harnesses/control planes (8090 Software Factory, Palantir, Nous’ Hermes etc). A 3rd party harness + the ability to use any model quasi-interchangeably is https://t.co/GtB9aXKN3d”
Palihapitiya reports carriers excluded AI damages from 82% of general liability policies per state regulators
“The insurance industry has priced the AI risk. Carriers went to state regulators this year asking to exclude AI-related damages from ordinary general liability policies. Regulators approved more than 80% of those requests, and the standard forms behind roughly 82% of”
Palihapitiya says major lab zero data retention insufficient for large enterprises
“@draecomino It doesn’t work in all cases as has been shown. It’s “reasonable best efforts” at best. For big companies this will be a non starter as they learn more.”
Palihapitiya predicts all paths lead to AI on-premises or in VPC
“@Srikantht All worlds lead to AI on prem or VPC imo”
Palihapitiya predicts that as AI model costs equalize, independent control plane providers will emerge as the next competitive layer.
“As model costs converge, the next likely step will be the use of independent 3rd party harnesses/control planes (8090 Software Factory, Palantir, Nous’ Hermes etc),”
Palihapitiya says third-party harness plus model interchangeability changes enterprise risk profile
“This is a really good development. As model costs converge, the next likely step will be the use of independent 3rd party harnesses/control planes (8090 Software Factory, Palantir, Nous’ Hermes etc). A 3rd party harness + the ability to use any model quasi-interchangeably is https://t.co/GtB9aXKN3d”
Palihapitiya reports carriers excluded AI damages from 82% of general liability policies per state regulators
“The insurance industry has priced the AI risk. Carriers went to state regulators this year asking to exclude AI-related damages from ordinary general liability policies. Regulators approved more than 80% of those requests, and the standard forms behind roughly 82% of”
Palihapitiya says major lab zero data retention insufficient for large enterprises
“@draecomino It doesn’t work in all cases as has been shown. It’s “reasonable best efforts” at best. For big companies this will be a non starter as they learn more.”
Palihapitiya argues four recent control planes govern model selection and cost but miss original intent
“On Control Planes… Four companies shipped an enterprise AI control plane in the last two weeks. They all do one thing well which is govern which model answers a request and what it costs. That’s a real problem and I’m glad it’s getting solved. But what none of them capture is”
Palihapitiya notes Zuck never mentions advertising because great engineers want space, search or AI company
“Zuck never mentions it because he knows the great engineers want to work at a space/search/ai company… Not an advertising company. https://t.co/ELf6QSgIFk”
Palihapitiya predicts leaders will commoditize features then drive token consumption as Databricks does
“Step 1. Everyone experiments. Step 2. The leaders start commoditizing features and make them much better for themselves. Step 3. Turn the crank on costs and make each capability high margin. As goes Databricks, others will follow. As a result, token consumption will explode https://t.co/8uOYCqwCLi”
Palihapitiya questions where token volume doubling every 11 weeks goes if not to high-value tasks
“Yet, high value tasks don’t seem to be increasing…so where are all these tokens going?? https://t.co/pvlby1K0io”
Palihapitiya says Global 2000 earnings growth has zero to do with AI despite token consumption
“@JTLonsdale Not saying there aren’t exceptions but I’m talking about the existing, largest companies. Those that comprise the Global 2000. They are the source of most of these tokens and revenues to the labs. Their earnings growth has literally zero to do with AI or anything remotely close…”
Palihapitiya says Global 2000 earnings growth has zero to do with AI despite token consumption
“@JTLonsdale Not saying there aren’t exceptions but I’m talking about the existing, largest companies. Those that comprise the Global 2000. They are the source of most of these tokens and revenues to the labs. Their earnings growth has literally zero to do with AI or anything remotely close…”
Palihapitiya questions where AI tokens are being consumed given that high-value tasks don't appear to be increasing.
“Yet, high value tasks don’t seem to be increasing…so where are all these tokens going??”
Palihapitiya says AI productivity boom not yet visible, still first inning with shape undetermined
“It’s not in there yet. All that has happened thus far is the breathless burning of tokens and prematurely declared winners and losers. The graph below explains better than any other that we are still in the first inning and the final shape of AI is still very much TBD. https://t.co/nzRWcbgAqi”
Palihapitiya recommends post on AI silicon architectures as excellent resource
“If you are interested in AI silicon read this. It’s 💯 https://t.co/rMe4l2xacX”
Palihapitiya predicts today's chips become mediocre when agents write kernels and new architectures
“@jacobpeake wow. this is super super good. thank you for writing this. one thought: what happens when agents start writing the kernels and new model architectures keep appearing? I think a lot of today’s chips become meh. they were designed by humans for humans. brute-force approaches”
Palihapitiya predicts today's chips become mediocre when agents write kernels and new architectures
“@jacobpeake wow. this is super super good. thank you for writing this. one thought: what happens when agents start writing the kernels and new model architectures keep appearing? I think a lot of today’s chips become meh. they were designed by humans for humans. brute-force approaches”
Palihapitiya says if Druckenmiller isn't embarrassed to be AI meat proxy, neither should you be
“If Stan Druckenmiller isn’t embarrassed to be a meat proxy, you shouldn’t be either. It’s like after matches was invented, still celebrating the arduous time to rub two sticks together. Dumb. https://t.co/UB319Irqcd”
Palihapitiya warns Claude's CLAUDE.md-only approach will always cause split brain problems
“@tobi unless they change it it will always lead to split brain...”
Palihapitiya calls AI disclosure requirements virtue signaling, compares to not disclaiming reading sources
“@Jason This is a dumb new form of virtue signaling. Do you disclose every article you’ve ever read that gives you an opinion when you spout off on X? No. This is how AI also works. He might as well disclaim the entire internet and every written word. If he puts his name behind it,”
Palihapitiya compares AI to typewriter, says it accelerates differentiation between good and bad writers
“@Icanonlyraz Moving from pen and paper to a typewriter only accelerated the differentiation between good and bad writers not diminish it. Helps get you to your desired destination faster. Slop is still slop. Over time, there will be fewer repetitive AI flourishes that stand out. And your”
Palihapitiya compares AI to typewriter, says it accelerates differentiation between good and bad writers
“@Icanonlyraz Moving from pen and paper to a typewriter only accelerated the differentiation between good and bad writers not diminish it. Helps get you to your desired destination faster. Slop is still slop. Over time, there will be fewer repetitive AI flourishes that stand out. And your”
“@Cernovich I don’t think he outsourced his thinking. He outsourced his penmanship. I’d look through the prose and focus on the message. He’s a 🐐 for a reason and this isn’t an IG reel.”
Palihapitiya notes Anthropic's Fable model only 11% of revenue, sees long J-curve to positive ROI
“Interesting development. The latest models are also the most expensive. One can see a long J curve building, then, to get to positive ROI for these models if folks don’t want to pay for the latest/greatest. On top of that, if most enterprises are realizing that previous https://t.co/ANKAC2zGm1”
Palihapitiya argues greatest AI margins will eventually flow to those closest to end users
“We know from the desktop and Internet eras that the economics started similarly and then flipped - the greatest margins were ultimately from those closest to the end user. The question is not if this will happen again in AI but when. https://t.co/O8wFcJ6Dj1”
Palihapitiya says using AI proxies is now standard practice, comparing resistance to celebrating primitive fire-starting methods after matches were invented.
“If Stan Druckenmiller isn’t embarrassed to be a meat proxy, you shouldn’t be either. It’s like after matches was invented, still celebrating the arduous time to rub two sticks together,”
Palihapitiya argues greatest AI margins will eventually flow to those closest to end users
“We know from the desktop and Internet eras that the economics started similarly and then flipped - the greatest margins were ultimately from those closest to the end user. The question is not if this will happen again in AI but when. https://t.co/O8wFcJ6Dj1”
Palihapitiya suggests AI changes IT transformation program vendor selection calculus
“Every large IT transformation program starts with a seemingly defensible choice. You pick a firm nobody can criticize you for picking, to help you transform. That made sense when the key question was whether anyone could build the thing at all. But AI changes this. The”
Palihapitiya warns liberal cities banning AI in schools will create economic segregation trapping kids in poverty.
“In all of these highly liberal bastions, they are embracing this far left wing ideology, and I think it's going to come with these kinds of policies.”
Palihapitiya estimates Anthropic and OpenAI IPOs will generate over $500 billion in charitable giving combined.
“The interesting thing that's gonna happen post the Anthropic and OpenAI IPOs is if you think about the amount of charitable giving or DAFs that are going to be funded by both of these two IPOs, it's going to be north of $500,000,000,000 between the two companies, and it could be even more than that.”
Palihapitiya predicts AI euphoria will continue for another three years before market correction.
“I honestly, I don't see this stopping. We're at the beginning of the beginning. I think it's gonna be like this for another probably three years.”
Palihapitiya predicts AGI-level capabilities will be matched by open source alternatives within three to four months.
“I also think that these capabilities will be matched or close to matched by a handful of closed source alternatives and open source alternatives probably within the next three or four months.”
Palihapitiya argues NVIDIA's Hugging Face deal will force major tech companies to compete across entire AI stack.
“And at the same time, what you're gonna do is actually create more rational competition up and down the stack. And I mentioned this last week. Everybody is becoming everything. NVIDIA will look like Amazon.”