Top Ten New York Times Bestselling Author
AI Summary. A benchmark designed to measure fluid reasoning rather than memorized knowledge — where earlier AI models scored near zero — has been solved at human-level performance by a new AI system. The benchmark's creator now expects true artificial general intelligence to arrive before 2030.
AI Summary. The share of AI queries requiring top-tier proprietary models has fallen from 60% to 25%, as open-weight models handle a growing proportion of tasks at lower cost.
Core argument: OpenRouter data show closed-weight proprietary models’ share of routed queries collapsed from 60% to 25% in 2025, indicating open models now handle three-quarters of real-world AI traffic.
One way to see which way the wind is blowing on open-model versus closed-model usage is by looking at data from router firms like OpenRouter, the New York start-up that Stripe agreed to buy last month. AI routers work a bit like an AI query triage service with a toll booth strapped on. Clients rock up, basically model-indifferent, and rather than tie themselves to any given model, they just send their queries to the router, which then flips them on to the lowest-cost model that will produce a good enough response for whatever the task at hand. Asking really tough closed-end frontier-model-worthy questions? To a frontier model they go. Increasingly, the share of queries that are truly closed-weight frontier-model-worthy is declining.AI Summary. Deep-tech investment outside AI has exceeded $150bn since early 2024, surpassing the $133bn invested across the entire prior decade. Falling valuations for traditional software companies and outsized returns from early bets on capital-intensive ventures are pushing investors toward riskier, science-driven deals.
Core argument: Deep-tech investment excluding AI exceeded $150bn since early 2024, surpassing the entire $133bn deployed across the prior decade (through end-2019), as falling valuations for traditional software push venture capital toward capital-intensive scientific bets.
The AI boom is fuelling a resurgence in ambitious “moonshot” bets, as early SpaceX backers’ huge returns and falling valuations for traditional software companies force tech investors to embrace riskier and more capital-intensive dealmaking. Excluding the giant sums ploughed into AI start-ups, global investment in “deep tech” — companies whose products are rooted in big scientific or engineering advances — has exceeded $150bn since the start of 2024, more than the $133bn in the entire decade to the end of 2019, according to Dealroom. This year’s deep-tech investments have not yet surpassed 2021’s peak, which was propelled by battery and electric vehicle deals for the likes of Rivian and Northvolt — many of which turned sour, highlighting the risks involved in moonshot dealmaking.AI Summary. Long-term interest rates have risen 45–79 basis points across major economies, with the AI investment boom—not fiscal or monetary policy—driving the surge in demand for capital. A comparable IT spending wave in the late 1990s coincided with even higher long-term rates despite low inflation and a budget surplus.
Core argument: Bruegel finds 30-year sovereign yields rose 45–79 bps across the U.S., Germany, France, Italy, the U.K., and Japan in the six months to Aug. 28, with the U.S. at 58 bps, suggesting a global demand-for-capital driver rather than U.S.-specific fiscal policy.
What [has been] driving interest rates higher [in the six months to August 28]? The European think tank Bruegel notes “US, German, French, Italian, UK, and Japanese 30-year yields [all] rose by 45-79 basis points in the six months to 28 August, with the US in the middle at 58bp.” It may not have much to do with policy at all, [but is instead due to] the surge in demand for funds as a result of the AI boom. We are in the midst of a surge in spending on IT that is on track to be even bigger than the boom of the late 1990s, [when] long-term rates were even higher then than they are now, even though inflation was low and we had a budget surplus.AI Summary. S. facilities, targeting 10,000 units annually by 2028 at mid-six-figure euro prices — a fraction of the $4m–$6m cost of comparable Tomahawk missiles. Lower unit cost enables large-scale barrages to saturate hardened targets such as weapons factories and military bases.
Core argument: Covenant’s Anthem missile, priced in the mid-six-figure euro range, costs roughly 10–25x less than a Tomahawk ($4mn–$6mn), enabling saturation-barrage tactics that offset its smaller 250kg warhead through sheer volume.
Covenant aims to produce 1,000 Anthem missiles a year at each of its German and US plants in 2027, and 5,000 a year at each site from 2028. [Anthem] would cost in the “mid six-figure” euros — a fraction of the price of Tomahawks, which are sold to overseas governments for roughly $4mn-$6mn apiece. Covenant declined to state the range of the Anthem but [Covenant's CEO] said the concept was partly a response to a German-British plan to jointly develop “deep precision strike” weapons with a range of more than 2,000km. Anthem will carry a warhead weighing up to 250kg half the 450kg payload of a Tomahawk. But Covenant said the lower cost and ability to be produced at scale would allow militaries to fire large barrages of the missiles to “saturate” targets such as weapons factories or military bases.AI Summary. AI-driven data-center expansion and related professional hiring have added roughly 1.05m jobs above trend since 2022–2023, spanning electrical contracting, equipment manufacturing, software development, and data science. The job gains exceed what broader construction, manufacturing, and professional employment trends would predict.
Core argument: AI-linked demand has generated roughly 730,000 above-trend jobs in engineering, software development, and data science since 2022, substantially outpacing near-term displacement effects.
[We] tracked five industries at the heart of the data-centre build-out, from electrical contracting to equipment manufacturing. Since 2023 employment in them has risen by roughly 320,000 more than broader construction and manufacturing trends would suggest. Not all of those jobs owe their existence to AI—grid upgrades and other factory building matters too. [We also] tracked employment in professional occupations closest to the AI boom—engineers, software developers, mathematicians and data scientists—and compared their growth since 2022 with professional employment overall. These roles have added roughly 730,000 jobs above trend in recent years. AI will not have created every single one of them. But it has almost certainly created quite a few.AI Summary. Net investment has fallen from ~40% of gross investment in the 1970s to ~25% today, meaning three-quarters of gross investment merely replaces depreciating assets. The shift toward faster-depreciating information technology assets requires larger gross investment increases to achieve any given gain in productive capital per worker.
Core argument: Net investment has fallen from ~40% of gross investment in the 1970s to ~25% today, meaning three-quarters of gross investment now merely replaces depreciating capital rather than expanding the productive stock.
The figure divides net investment by gross investment. Back in the 1970s, net investment was often around 40% of gross investment, but the share has been slumping over time. For the last decade or so, net investment has been about 25% of the gross–that is, about three-quarters of gross investment is just making up for depreciation of the pre-existing capital stock. The likely reason for the growing gap between gross and net investment is that modern investment is more likely to be related to information technology [which] depreciates more rapidly and thus needs to be replaced and updated more often. If we want the average US worker to be using a greater amount of capital on the job–which was one of the key drivers of rising labor productivity in the past–it now takes a bigger rise in gross investment to lead to a given rise in net investment.AI Summary. S. 575 between 2022 and 2026, the first sustained decline in relative demand for college-educated labor in four decades. AI exposure in white-collar occupations accounts for roughly 28% of that drop, as wage growth slowed disproportionately in high-AI-exposure jobs where college graduates are concentrated.
Core argument: The U.S. college wage premium fell from 0.626 to 0.575 between 2022 and 2026—the first sustained decline in relative demand for college labor after four decades of uninterrupted expansion.
After expanding for four decades, the U.S. college wage premium [dropped] sharply from 0.626 in 2022 to 0.575 in 2026. Current Population Survey data through 2026 implies an unprecedented drop in relative demand for college labor—the first sustained negative relative demand growth. Post-2022 wage growth slowed disproportionately in high-exposure occupations, which employ a disproportionate share of college graduates. By 2026, going from zero occupational AI exposure to full exposure had a negative effect on wages of−0.086. Combined with the college–non-college exposure gap, this mechanism accounts for roughly 28% of the total drop in the college wage premium from 2022 to 2026. While non-causal, these patterns indicate that task displacement in AI-exposed white-collar occupations plays a quantitatively meaningful role in the recent compression of the aggregate skill premium.