Luciano Caratori

"Freedom is the freedom to say that two plus two make four. If that is granted, all else follows."

@joshwoodward A bit too much?

Get it here: gemini.google/mac/

Introducing Gemini on Mac. We heard your feedback. We recruited a small team. They built 100+ features in less than 100 days. 🤯 100% native Swift. Lightning fast. Let us know what you think!

By volume & absolute magnitude of electricity consumption, aluminum deserves its reputation as "congealed electricity". But from a electricity/ton perspective, aluminum is hardly the only congealed electricity commodity. From my "The Aluminum Tech Stack" piece a few weeks ago:

SUPERCOMPUTING + NATIONAL LAB = ENERGY POLICY ACCELERANT. U-M and Los Alamos National Laboratory announced a $1.25B supercomputing research center in Ypsilanti, MI. · What it is: A 290,000+ sq ft facility with both classified federal research and open academic research wings · Energy angle: Explicitly scoped for clean energy optimization - solar, wind, grid resilience, and climate modeling · LANL context: Los Alamos runs the U.S. nuclear weapons stockpile simulation program - this partnership brings that compute culture into civilian energy research · Timeline: Design phase now through 2027, construction 2028-2031 · Power setup: Dedicated DTE substation, fully electric ops, LEED Silver target The gap between energy policy ambition and physical reality is a compute problem as much as a capital problem. Facilities like this close that gap. $NNE $CEG $VST $GEV

The current energy crisis clearly demonstrates the strategic case for renewables as a national security imperative. Country examples in @IRENA’s new policy advisory confirm the resilience of renewables 👉bit.ly/3OjvPWE

The AI data center and its affiliated power construction projects are going to drive a lot of freight over the next 5 years. Our estimates are for 3.6m truckloads this year involved in these projects, ramping to 5.6m truckloads by 2031. The numbers are astounding - how much steel, concrete, copper, and cooling systems are involved in a single hyperscale data center. Drivers: Follow the money and go buy a flatbed truck

The US imports 82% of its large power transformers. Only 8 to 10 domestic factories can build them. Chinese transformer shipments to the US went from 1,500 units in 2022 to over 8,000 by October 2025. Now Hitachi is building a $457M transformer plant in Virginia and Hyundai broke ground on a $200M expansion in Alabama last month. bloomberg.com/news/newslette…

France 🇫🇷 doubles electrification support to €10bn/year through 2030 — EVs, heat pumps, industry. The reason? The Strait of Hormuz crisis making fossil fuel dependence impossible to ignore.

as the world faces a global transformer shortage, a wave of startups is tackling the problem from first principles — either redesigning the manufacturing process and supply chain, or overhauling standard transformer designs in favor of solid-state tech heatmap.news/climate-tech/t…

whoever was responsible for this rebrand was obsessed and insane. this is the pinnacle of a good brand.

Industrial heat pumps are one of the most underrated technologies in the energy transition. They work. They scale. And they’re finally moving from demonstration to deployment. Here’s a 95t heat pump component just traveled l by river to reach BASF’s Ludwigshafen site.

Nearly half of US data centers planned for this year are expected to be delayed or canceled. Not a capital availability or chip problem. It's transformers and electrical equipment. Lead times have gone from 24 months to five years. The big four are spending $650B+ in 2026 alone but can't actually source the gear domestically. So we must buy from China. Ironic that the country the US has to beat in AI is also the country supplying the parts to build it.

Emerging economies are leapfrogging to decentralised, modular solar and batteries – much like they skipped fixed-line networks for mobile. This shift solves the last-mile challenge, bringing energy access to those the fossil system left behind. @TheCVF ember-energy.org/latest-insight…

The California Grid Manufacturing Initiative (AB 2516, @AsmPetrieNorris) takes on these challenges on both the supply and demand fronts:

Interesting: U.S. annual domestic manufacturing capacity for energy storage modules & systems now exceeds annual demand for the first time - Domestic cell production to meet annual demand by 2027 - Lithium production to meet annual demand by 2030 - Tesla is a massive contributor in domestic cell manufacturing, storage systems, AND lithium refining (Corpus Christi) - Domestic manufacturing of alternative storage tech also coming through with Form (WV) and EOS (Pittsburgh - 8 GWh by 2027) - U.S. still lags behind on anode materials

The physical world doesn’t scale like software US disclosed pipeline hit 241 GW (+159% YoY). But only 33% is under active development.. Q4 additions halved to 25 GW (down from 49 GW in Q3). Developers are shifting from hype announcements to executing the existing pipeline: grid queues, transformer lead times (2–3+ years), and power constraints are the real brakes.

Full launch on a new Dashboard Fifty years of public funding across 8 major energy sources Lots of context and caveats all laid out in the link in my bio but a few key points below🧵 carbonmiddlemanagementinc.github.io/publicfunding_…

JPM shows the workflows of how electrical architecture changes in a data center as it transitions to 800v: 1/4 Traditional AC data center: Utility feeds MV switchgear, steps down to 480V, passes through ATS with generator backup, then UPS, PDU, transformer, and rack PDU before reaching the server. Most conversion stages, least efficient end to end. Every component in the chain adds loss. 2/4 White space retrofit: Upstream plant stays the same. The change happens near the racks: an AC-to-DC rectifier feeds an 800 VDC sidecar that replaces the traditional UPS and transformer-heavy IT path. Servers still need a DC-to-DC converter at the rack. Keep the building, modify the data hall. 3/4 Native 800 VDC servers: Same facility backbone, same upstream AC plant. But the server now accepts 800V direct current natively, which eliminates the rack-level DC-to-DC converter from the previous stage. Fewer conversion stages, less equipment in the power chain, lower losses. 4/4 Full SST architecture: The IT load gets its own dedicated power path. Utility, on-site generation, and battery storage feed a solid-state transformer that delivers 800 VDC directly to native servers. Building support loads stay on a separate conventional AC branch. Most direct grid-to-chip design, most architecturally disruptive.

Today, we're announcing Heaviside, our foundation model for electromagnetism. Trained on tens of millions of designs and over 20 years of proprietary simulation data, Heaviside predicts electromagnetic behavior from geometry in 13ms, which is 800,000x faster than a commercial solver. Heaviside is not a language model, and it’s not a surrogate model. Heaviside marks a new class of foundation model for physics which understands the fundamental relationships between materials, the geometries and the electromagnetic fields they generate. We’re releasing a research preview of Heaviside in Atlas RF Studio, an interactive agentic sandbox where you describe the EM behavior you want and the model generates the physical structure that produces it. @arenaphysica , we believe the implications of this class of model extend well beyond RF, as the frontier of exquisite hardware is electromagnetically-governed: wireless communication, radar, power delivery, high-speed computing, and the interconnects inside every chip on earth. In the months ahead, we’re excited to scale up Heaviside to broader frequency ranges, design spaces, and to support silicon-level designs, and deploy it with our closest partners and collaborators in service of their biggest design challenges. If you’ve read our thesis, this is just Step 2 in our pursuit of electromagnetic superintelligence. Read the full announcement and try Atlas RF Studio…tell us what you think: arenaphysica.com/publications/r…

Después de un finde a mil en Hackitba (@cs_itba) les presento el proyecto que hicimos con @MatiMutz_ y @MatiasOstrower Se llama Alias, una skill + un MCP que le permite a cualquier agente (@openclaw, ChatGPT, Claude, etc) hacer transferencias a cualquier CBU/ALIAS Como? Te creas una cuenta, la fondeas (gracias @talo_pay) y tus agentes disponen de esos pesos para poder transferir a cualquier cuenta. Los casos de uso son realmente infinitos Esto destraba una problemática enorme que tienen los agentes ahora que si bien son cada vez más inteligentes, no pueden pagar (y menos en pesos$$$) Eh? Si escuchaste bien ;) Demo en el video: …ant-clarity-production.up.railway.app

@fiveoutofnine Hi, can’t find M4 pro 24GB. Can I help you to get the data through my setup?

Introducing whatcani.run Find the best local models based on real data 1. People run and submit benchmarks 2. Stats aggregated over models / devices 3. Find the best model for you `npx whatcanirun`, fully open-source