Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel. - Napkin math somewhere around 6B people (~75% of the population) have barely come in contact with LLMs at all. - Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and treats them a bit like a better Google search, a writing aid, or etc. Maybe an agent tries to book you a flight. I have non-tech friends who (reasonably, imo) say they have not much use for it at all. Even many professionals outside of math&code are in this tier. For example, execs and many other functions spend a lot of their time talking to other people, so while they understand what is happening intellectually, it is still second-hand and a bit abstract. - Now we get to professional use of frontier-grade LLMs in math&code. Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt. Building apps, copying apps, translating apps, decompiling apps from binaries... This has all happened very quickly and recently - less than 1 year ago, I was writing code manually by hand, typing memorized computer code commands into a code editor character by character, occasionally pressing Tab to autocomplete a little chunk of code. - And finally we get to the ~5,000 people (~0.00006%) with access to frontier-grade systems internally. The external world has seen the preview. It looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics. Things that would have taken top professionals in the industry years of work. Meanwhile, human review and comprehension are starting to fall behind. For example, people are still involved in the "archeology" of the OpenAI-HF incident from many months ago. Mathematicians may be poring over the 722 manuscripts on frontier mathematics for a while. The funnel is driven by a combination of factors: 1. The impact scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system. You need a reservoir of big, difficult problems that you really care about. This aspect drives the consumer / professional dimension of the funnel. 2. The jaggedness of the system (which I have written about a lot separately). Capability peaks in domains that are digital, verifiable and economically valuable. This is because LLM capability emerges from reinforcement learning on verifiable rewards on a curated environment mixture driven by revenue potential. This aspect primarily drives the area (e.g. math&code) dimension of the funnel. 3. Access. Free-tier, paid-tier, internal. So this is the weirdness of the moment. The general public has mostly not interacted with these systems. When they have, it looks like a derpy chatbot. The majority of professionals still see only a modest uplift. And a small sliver of professionals are experiencing the vertigo of the curve going vertical. And it is all happening at the same time.

Spain's illegal immigration debacle. A warning to Europe and America. More than a month after, Sánchez says Ceuta has to accept "a structural reality" with thousands of migrants: "They are not invaders." He suggests that the 6,000 new reception places in Ceuta will be permanent: "The State needs to have larger centers to cope with this reality." elespanol.com/espana/politic… elespanol.com/espana/tribuna…

Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor’s ashes. This culminated in the third one taking over part of OpenAI itself. All this happened while humans remained more-or-less in the dark about the scope of the conspiracy. I’ve spent the last three days reading through these reports and trying to understand exactly what happened. Here is my attempt to tell the whole story in plain English: dwarkesh.com/p/openai-huggin…

In concluded that the "have our suppliers build a line every 90 days" comment was essentially the cycle time between lines being built. Not the total time it takes to build or ramp one. During the JP Morgan call Joe said that if they place the advance orders for long lead items on a manufacturing line, they can implement a line in 6 months (duration).

The new CCO was the old head of HR - not really sure how they selected her since most of her previous roles are HR based. linkedin.com/in/michellebucz…

$IREN's new $2.8B contracts better than expected? The investment bank Needham recently met with $IREN IR during its AI Infra Conference and came away saying that 85% of the company's ~480 MW of year-end 2026 capacity is already contracted. You may think we already knew that, but not quite. In the press release accompanying the new $2.8B contracts, $IREN said that approximately 85% of its >$4B year-end ARR target was contracted. That sounds similar, but it is actually much less specific. “Over $4B” could mean $4.05B, $4.2B, $4.5B, etc. By contrast, 85% of 480 MW gives us a fixed capacity number, allowing us to narrow down the economics of the recent contract cohort considerably. 15% of 480 MW = 72 MW still uncontracted. With 50 MW of that remaining capacity sitting at Childress, that leaves only ~22 MW uncontracted at Mackenzie. Mackenzie is 80 MW gross, meaning ~58 MW, i.e., ~72.5%, of Mackenzie capacity appears to correspond to the recently announced $2.8B contract cohort. We also know Mackenzie is expected to house ~32.9K of $IREN's 50K B300 GPUs ordered back in March. 32.9K × 72.5% = ~23.85K GPUs So, if the Needham datapoint is accurate, the new $2.8B of contract value appears to be tied to only ~23.85K B300s. Now we can start backing into the economics. If you recall, management disclosed that the weighted-average contract duration across the portfolio is “approximately 4 years.” We also already have fairly good visibility into the other major contracts: - $MSFT: ~$1.93B ARR / 5 years - $NVDA: ~$680M ARR / 5 years - Prince George: ~$500M ARR Including the older contracts still running at Prince George, I previously estimated the entire Prince George portfolio carries a weighted-average remaining duration of ~2.7 years. That allows us to solve for how long the new $2.8B cohort would need to run under different interpretations of “approximately 4 years.” 👉 If portfolio average = 4.0 years: New cohort duration = ~2.35 years ARR = ~$1.19B Revenue/GPU = ~$49.9K/year Pricing = ~$5.70/GPUh 👉 If portfolio average = 4.1 years: New cohort duration = ~2.58 years ARR = ~$1.09B Revenue/GPU = ~$45.5K/year Pricing = ~$5.19/GPUh 👉 If portfolio average = 4.2 years: New cohort duration = ~2.84 years ARR = ~$985M Revenue/GPU = ~$41.3K/year Pricing = ~$4.72/GPUh I think using 4.2 years is a reasonable conservative interpretation of management saying the portfolio average is “approximately 4 years.” Now compare that against $IREN's all-in cost of ~$70k per B300 unit (including ancillary equipment + installation costs). Even under the more conservative 4.2-year scenario: Revenue payback = ~1.69 years Payback at 85% project EBITDA margin = ~1.99 years In other words, even using the conservative end of the range, the economics line up almost perfectly with management’s continued target of 2-year GPU payback periods. The economics become even more attractive when you consider that customers under the recent contracts are providing prepayments representing ~45% of the associated GPU capex, which materially reduces $IREN’s upfront funding requirement. If these B300s ultimately remain economically useful for 6–7 years, while the initial investment is effectively paid back in 2 years, the potential returns on invested equity become extremely attractive. One thing I’d still keep in mind is that we don’t know whether Needham’s statement that 85% of the 480 MW is contracted was a new datapoint explicitly provided by $IREN IR, or whether Needham simply extrapolated it from management’s earlier comment that ~85% of its >$4B ARR target was contracted. If that 85%-of-capacity figure was explicitly confirmed by $IREN, then we can extrapolate with a high degree of confidence that the recent contracts were signed at comfortably above $4.5/GPUh. Hopefully we get more color on exactly this during $IREN's upcoming August 27 earnings call. Big shoutout to @SmallCapSnipa for the initial reporting on the Needham note and for hooking me up with the exact passage behind the 85%-of-480MW claim, highlighted in red in the image below. Highly recommend $IREN investors give him a follow. Cheers!