Allocator of things to allocate. Views are my own. Not investment advice. DOW IS AT 50,000
San Francisco, CA
Joined November 2012
- Tweets7.1K
- Following4.9K
- Followers12K
- Likes5.2K
Dynamicmoats retweeted
If you’re a SpaceX investor you have to listen to this video! $SPCX
“In 10 years I think it’ll be thousands of dollars a share. I think it’ll be worth $20T, $30T, $50T and Elon thinks I’m way low. It’s now $2T” xitter.cf/firesidealpha/status/2…
Dynamicmoats retweeted
When China figures out how to manufacture a product, their ability to scale is often greater than the rest of the world combined. We saw this with lithium batteries and solar, for example.
SpaceX is a rare counter-example, where its mass to orbit is roughly an order of magnitude greater than the rest of the world, BUT China now does a majority of non-SpaceX mass to orbit. SpaceX is an anomaly in the matrix.
Maybe RAM will be different, but RAM was a low margin commodity only ~3 years ago. LP5 and LP6 are designed to a common standard that China seems to be solving fast.
Newer HBM architectures and things like an SRAM layer below logic and 3D stacking RAM on top of logic while finding a way to keep the logic cooled will be harder to solve.
Dynamicmoats retweeted
This is very underappreciated
Hyperscalers and frontier labs won't blink on spend because they know this is an existential race
Dynamicmoats retweeted
Those in the industry know this already, but the actual DCF of pod life is lower than many think. That said I would also say most firms will try to retain top senior talent even if the PM is let go as they try to collect enough data to identify what the problem is (ie PM vs Senior Analysts)
Dynamicmoats retweeted
Base case increasingly seems like:
-$200 oil case off the table
-US/Israelis continue bombing for some time, fails at regime change but eventually stops and calls it a win. Iran escalates on a tit-for-tat and de-escalates once US/Israelis stop bombing
-Iran tolls and controls Hormuz, traffic gradually reverts. Pipeline diversion + traffic recovery results in LSD crude supply deficit
-Regions like SE Asia, Australia, and parts of Europe experience periodic fuel shortages
-People start bidding AI stocks and BTD wins again. AI fundamentals are stronger than they were 6 weeks ago and likely the only group of stocks with major upward earnings revisions vs. old economy companies that will see cost pressures.
-Can equity markets hold up at $100 dec 26 brent/$130 june 26 brent? TBD
Feel free to change my mind with facts
Dynamicmoats retweeted
1/n Lottomatica $LTMC is the $DKNG of Italy and is a World Cup 2026 beneficiary. Market leader taking share in a fast growing industry
70% upside over the next 12 months with very little downside
Dynamicmoats retweeted
Every podcast on the war:
Guest: "If Iran does the rational thing and agrees to a negotiated settlement, it will be fine."
Host: "And if not?"
Guest: "Then it's a tough situation."
Dynamicmoats retweeted
$MU $SNDK $EWY
Every few months a paper drops claiming AI needs less memory. This is that paper. There is no such thing as a free lunch.
Compressing KV cache in controlled benchmarks is not the same as scaling it universally to every user, task, and platform in the world. Accuracy collapse is real at scale. As expected from a research paper, the downsides never make the headlines.
Remember Titans Architecture? Google said it would make memory usage more efficient. Didn't move the needle. If anything, it drove more HBM demand. Jevons Paradox. What did Deepseek do? What's memory demand today? You be the judge.
HBM is still the primary fuel for AI training. That does not change. The paper targets long-context inference which relies on a full memory stack: HBM for speed, server DRAM and NAND for capacity and storage. When inference runs, memory spills from HBM into conventional DRAM and NAND. HBM stays fully utilized. At best, this research reduces the spill in a controlled workload, IMO.
Introducing TurboQuant: Our new compression algorithm that reduces LLM key-value cache memory by at least 6x and delivers up to 8x speedup, all with zero accuracy loss, redefining AI efficiency. Read the blog to learn how it achieves these results: goo.gle/4bsq2qI
Dynamicmoats retweeted
The Boxer Amphibious Ready Group (ARG), comprised of the USS Boxer (LHD-4), USS Comstock (LSD-45), and the USS Portland (LPD-27) with the embarked 11th Marine Expeditionary Group (MEU), is on its way to the Middle East after departing the U.S. west coast for deployment.
Dynamicmoats retweeted
I talked to two Israeli sources on why Iranian launches continue to increase, despite US-Israeli claims that they have destroyed almost all of the launchers. Here is what they said:
1) The 90–95% drop in volume claimed by CENTCOM earlier in the month was probably a temporary lull as Iran repositioned its remaining launchers into hardened sites. Independent satellite analysis suggests that a significant portion of the "80% destruction" claimed by the IDF actually hit high-fidelity decoys.
2) Despite fewer launchers, the lethality per strike has increased. Iran's shift to cluster warheads has allowed a single missile to impact multiple locations simultaneously, compensating for the lower volume of launches
3) Iran has successfully set up mobile, underground units able to fire at steady rates. Iran used that quiet period to move their remaining ~100-120 heavy launchers into "Super-Hardened" facilities
4) Iran is utilizing its Zolfaqar and Dezful road-mobile launchers. These units move from hardened tunnels to pre-surveyed launch spots, fire, and return underground in under 10 minutes, often before coalition drones can re-task for a strike.
5) Because these launching units are decentralized, it is very hard for US and Israeli intelligence to get info on them.
Israel and the United States do not have an answer to this problem. That is why they are trying escalation on energy sources instead. But that is backfiring.
Dynamicmoats retweeted
Classic Netanyahu: In Hebrew, he said the war will continue as long as it takes
In English, he said that it will end faster than most think
Dynamicmoats retweeted
BREAKING: Iran has issued an urgent warning calling for the evacuation of petrochemical facilities in Saudi Arabia, Qatar, and the United Arab Emirates.
Dynamicmoats retweeted
SCOOP: Senior Israeli officials tell U.S. diplomats Iranian protesters “will get slaughtered” if they take to the streets, though Israel is still angling for a popular uprising, per a State Department cable from Embassy Jerusalem 🧵
Dynamicmoats retweeted
With all the talk about warships escorting tankers through the Strait of Hormuz: I was aboard the USS Abraham Lincoln when it sailed through the Strait in 2012 as a show of force. What stood out was how close the ship got to the Iranian coast. We could easily see both the Iranian and UAE shorelines and little merchant boats crossed the carrier‘s path constantly. The crew had to secure all the aircraft because take off‘s and landings are impossible in the narrow waterway. Any ships, even warships going through there would seem to be pretty vulnerable. Report from 2012 attached: cnn.com/videos/world/2012/02… @CNN
Dynamicmoats retweeted
🧵This is a lie, the mediator, “Persian Gulf diplomat” has said. “I can categorically state that this is inaccurate. He was explaining that all of this material can all go away should we have a deal and Iran can be relieved from sanctions.” ms.now/news/exclusive-diplom…
Dynamicmoats retweeted
Another curveball in the AI ecosystem. Not out of nowhere we have been seeing this from many like groq, cerebras and samba but at 17k t/s this could shakeup cloud inference economics.
Still pretty speculative at 8b parameter. Doesn’t help a ton right now but surely companies will adopt a hybrid approach once a frontier model is baked in.
$NVDA of course saw this and isn’t caught flatfooted as they recently acquired groq but it does show phone inference could be commoditized sooner than we thought. Apple can easily bake a model into their silicon that would be good enough for 80% of people with an iPhone. The ones who are doing agentic/ deep thinking reasoning will still need cloud compute and lots of CPUs/GPUs.
17,000 tokens per second!! Read that again!
LLM is hard-wired directly into silicon. no HBM, no liquid cooling, just raw specialized hardware. 10x faster and 20x cheaper than a B200.
the "waiting for the LLM to think" era is dead. Code generates at the speed of human thought.
Transition from brute-force GPU clusters to actual AI appliances.
taalas.com/the-path-to-ubiqu…
