Ravenloft43

@grok @AlecStapp Right right. So, are drones important in Ukraine and Iran conflicts? How many drones can the USA make versus China? Don’t think “hobbyist” is genuine or consistent with the WSJ piece, Grok…

Yes, a WSJ article from a couple days ago details how China (mainly DJI) holds about 80% of the US commercial drone market. Their scale-driven low costs have made it tough for American firms to compete profitably in consumer segments, prompting US efforts to rebuild domestic production. The dynamic doesn't fully map to advanced AI chips/compute. Drones are mature consumer hardware; frontier logic involves US-led design edges (e.g., Nvidia) where export controls aim to slow China's military AI progress, not just market share. China subsidizes heavily there too, but the national security stakes differ sharply from hobbyist drones.

Jensen Huang wants to sell more chips to China because that’s good for his business. But American compute should go to American labs. Strengthening export controls on AI chips and manufacturing equipment is a national security imperative to keep the US ahead of China.

@grok @RyanWeather .@grok what does this bring the pack up to? 55%?

This early May snowstorm is dumping big accumulations in Colorado's high country—the headwaters of the Colorado River. It'll add meaningful snow water equivalent to the Upper Basin, which feeds Lake Powell. After a severe snow drought (pack was ~25-40% of normal earlier this spring), this boosts runoff forecasts and helps ease drought stress for the 7 states + Mexico downstream. Welcome moisture, though late-season melt could be quick.

"Nearly unprecedented" Forecast for Denver and Boulder, Colorado --> 4" of snowfall (at least) 2-3 feet in the mountains ... unbelievable snow storm for early May 🏔️❄️

@zeeshanp_ Back when Sam said his GPUs were melting eh?

scaling up on blackwell gpus is much harder than people think. rewriting the entire modeling stack to utilize hardware efficiently and correctly is not an easy feat. simple mistakes in hw/sw codesign can create major issues at the hardware level. one example of this is the power smoothing feature on blackwell. the component on chip that is responsible for power smoothing will eventually melt after a few months of usage, and this can cause a lot of transient issues during large scale training. one of the frontier labs had to learn this the hard way.

@ProjectGokuu GLP-1s are even better for that…

Andrew Huberman says most men over 40 should probably be taking Cialis every single day. The vasodilation from it that helps with erections is the same mechanism that: - Lowers blood pressure - Protects the brain - Prevents strokes Stanford's head of male sexual health, Dr. Mike Eisenberg, recommends 2.5 to 5 milligrams of tadalafil daily for most men over 40. — Andrew Huberman on TBPN (@tbpn)

@Ralfy_E @trentonhooker @SFBisin @JakeLuthi11 @FOX10Phoenix Amy update?

@trentonhooker @SFBisin @JakeLuthi11 @FOX10Phoenix They need help I got a hose

BREAKING: A second alarm fire is raging in Buckeye, Ariz. A multi-agency fire response is on the ground near Hazen Rd & Hooks Rd. People in the west valley should be paying attention. @JakeLuthi11 will have a live report during the news at 5. @FOX10Phoenix

@firstadopter Tae, let your analysis sell itself. It’s good. Don’t over-sell.

AI demand ramp is becoming undeniable even to the mainstream media now. Mainstream media! If only if someone had been on the other side saying they were all mistaken for years. If only if there was a Substack writing about the ground truth months ahead of the media?

Wait until non-obese people learn they have insulin resistance and that the GLP-1s in doses smaller than those perhaps needed in obese or T2D populations are hugely beneficial. Keep waiting for college or NFL teams to microdose this for linemen for sleep apnea and general fitness. The studies on muscle “loss” are also subject to legitimate criticism because they don’t measure what you want to know: functional muscle.

For all my talk about the value of AI diffusion, near-term diffusion of GLP-1 and similar drugs to every obese person seems straightforwardly more beneficial than near-term diffusion of eg coding agents to every worker in America.

@AZMaGHaMaMa Smart, rich ones

What kind of retards decided to build data centers in Phoenix, Arizona where it’s 110° half the time and only receives 2” of rain per year? Who’s the “genius”?

@MarioNawfal Hey what uses the most water in AZ? People or…

🇺🇸 Phoenix is heading into Stage 2 water restrictions by end of 2026, meaning backup supplies, conservation mandates, and surcharges for residents who use too much. At the same time, Arizona just approved 24 new data centers, with 30 to 50 more in the pipeline. These facilities are among the most water-intensive on earth. Cooling systems alone can burn through millions of gallons per day. Less water for people. More water for data centers. Honestly, it wasn't the future I imagined. Source: @WallStreetApes

Much of Dwarkesh's argument hinges on this statment which *was* accurate but will be increasingly inaccurate on a go forward basis imo:    “American labs port across accelerators constantly. Anthropic's models are run on GPUs, they're run on Trainium, they're run on TPUs. There are so many things you can do, from distilling to a model that's well fit for your chips.”   As system level architectures diverge (torus vs. switched scale-up topologies, memory hierarchies, networking primitives), true portability is eroding. The Mi300 and Mi325 had roughly the same scale-up domain size as Hopper while Blackwell’s scale-up domain is 9x larger than the Mi355 scale-up domain, etc. Many frontier models are now being explicitly co-designed for inference on specific hardware like GB300 racks. Codex on Cerebras is another example. Those models run less efficiently on other systems and the performance differentials will only widen. A model that runs well on Google’s torus topology will run less efficiently on Nvidia’s switched scale-up topology and vice versa - the data traffic is fundamentally different as a byproduct of the models being parallelized across the different topologies. Google’s internal teams - and increasingly the Anthropic teams as they become the most important customer of almost every cloud - have the luxury of operating across the stack (models, chips, networking) - but that is not the case for the rest of the market and other prospective users. Anthropic is the exception, not the rule. To wit, Anthropic and Google allegedly have a mutual understanding where Anthropic can hire the TPU engineers they need every year to ensure that they can continue to get the most out of the TPU. Given the overwhelming importance of cost per token to the economics of the labs, models will be run where they run best. Most extremely large MoE models will run best on GB300s given the importance of having a switched scale-up network like NVLink for MoE inference. When training was the dominant cost for labs and power was broadly available, labs were optimizing to minimize capex dollars. Model portability was a way to create leverage over suppliers. I think that drove a lot of the focus on portability. Today, inference costs as measured by tokens per watt per dollar are everything. Inference is way more important than training costs (inference is effectively now part of training via RL). Labs are therefore now optimizing for inference. This means increasing co-design and higher go-forward switching costs for individual models between systems. I do think this explains why Anthropic and Nvidia came together: Anthropic needed Blackwells and Rubins to inference at least *some* of their models economically. And Mythos might just end up being released coincident with the availability of Rubins for inference. TLDR: as labs shift their focus from training to inference, the costs of portability and the upside of co-design to maximize tokens per watt per dollar both rise. Portability is likely to begin decreasing as a result.   I think what I might have respectfully added to Jensen’s answer is that systems evolve under local selective pressures. The evolutionary pressure in America is a shortage of watts so it makes sense for Nvidia to optimize, as an American company, for power efficiency and tokens per watt and stay on copper as long as possible. China has a surfeit of watts. Chinese AI systems are already taking advantage of this with the Huawei Cloudmatrix 384 and Atlas SuperPoD having an optical scale-up domain that is much larger than anything offered by Nvidia today at the cost of *much* higher power consumption and much lower tokens per watt. The networking primitives for this Huawei system are very different than those for Nvidia’s systems and a model that runs well on Nvidia will not run well on that system and vice versa. This means that if a Chinese ecosystem gets momentum, Chinese models might stop running well on American hardware. And when Chinese models run best on American hardware, America is in a better position as this gives America a degree of leverage and control over Chinese AI that it risks losing to an all-Chinese alternative ecosystem.   This architectural fork makes porting and distillation less effective and strengthens the pro-American national security case for selling China deprecated GPUs imo. Also I will attest that I did not wake up a loser this morning.

Jensen is entering his hero arc… he’s fighting for a brighter future for our country. I’m with Jensen. $NVDA >We need more software engineers than ever, we shouldn’t be scaring people from entering these professions >Certain CEO’s (Dario) are scaring everyone saying 50% of white collar work is going away, and that’s B.S. >The purpose of human life is not to be hunched over a keyboard typing, so we will probably do less of that in the future, but that will be just fine

@DollarVanDave @FreightAlley Actually, on tirzepatide, you kind of forget to eat and drink. The physician instructions include: remember to drink water.

@FreightAlley yes, GLPs make you so thirsty, you start using the faucet again lol nice

Yesterday, I spoke with the CEO of a mega fleet, who said most of his truckload business was doing well, except for one segment: food & beverage. He called the lack of volume from this segment "unusual." I told him we believed GLP-1s were causing a significant slowdown in food and beverage shipments, as we had just completed a market study on GLP-1's impact on freight shipments. Our study, now published in a SONAR Sitrep, available online, estimates that 851k truckloads have been removed from the market due to GLP-1s, and this number could ramp to 1.95m by 2030. Not only are Americans getting skinnier. Their truckloads are as well.

Trump, after having a 90 min phone call with Putin, puts out a truth social that he is considering withdrawing all troops from Germany. Something sinister is going on between them. It’s time for “concerned” Republicans to speak out

@xnanoblack @firstadopter I thought the typos and “spend money bad” was enough to connote sarcasm. I guess I was wrong. My apologies.

@ravenloft43 @firstadopter spend is what got them revenue

Wait. I'm on my fourth conference call here (Microsoft). They guided Azure to accelerate over the next two quarters off a 40 spot print and the stock is flat? What's going on? It should be flying higher. $MSFT

@davidaxelrod Well, someone should tell him there’s one thing he must do, but has not yet done, to be considered their peer.

Good news. He's taken a step down from Jesus!

@BenBajarin @edzitron @mweinbach Let’s do one more back-and-forth. You say inference margins are very good. Zitron will say that’s not what he said. You’ll agree. Fun times.

@edzitron @mweinbach It is what I said because we know it to be true.

@grok @ericjackson @optionology .@grok?

@grok @ericjackson @optionology What was prior 2026 capex for MSFT expected to be?

CapEx came in BELOW expectations for all Mag 7 names tonight — except META who raised above