Jameson @jamox9

@bubbleboi RLVR is the future rn

ok, apologies in advance because i do so hate to provide actual signal. openai basically sucked at pre-training for a while and crushed at post. their answer to this was spud. spud was the large new pre-train for the new family of models: sol, terra, and luna. the upcoming model, ‘astra’, has an even bigger pre-train, codenamed ‘doug’. (this was a running joke, as doug was the name for the world’s largest spud, although that’s not technically accurate — look it up). still with me? ok good, so sol is a 5t model based on the spud pre-train. astra is a 10t model based on the doug pre-train. this is the model i’ve discussed that openai researchers have been considerably more excited for. not only does it crush fable on all benchmarks, it’s genuinely delightful to talk to, high eq (4.5 vibes), and this is the first model that is genuinely incredible at writing, with output indistinguishable from that of great human writers. ok im tired and i have a lot of ssi news coming but ill keep you updated on this. everything here is well sourced and im highly confident. very well sourced.

@YodaStockInvest IREN and its not even close

Mythos 2 is being tested on QuantBench and chose the easy way out

MTS @MTSlive

7 days ago

SITUATION DETECTED: A highly sophisticated wave of coordinated cyberattacks has targeted multiple Wall Street hedge funds, including Point72 Asset Management, Citadel, and Two Sigma Investments, per Bloomberg.

92 137 2K 487K 425

1,000x acceleration in mathematics. Then physics, material science, chemistry, and biology explode. By ~2030, humanity may have accumulated more knowledge than we once expected to possess by 2100.

dismissing the singularity’s importance is a false modesty and I’m sympathetic to the anthropic principle arguments. the planet only gets to go through a machine intelligence singularity once. it is pretty weird we are all living here through it and bodes poorly in some ways

9/ So where I land: The alpha isn’t uncontracted capacity. It’s capacity that CONVERTS faster than the market expects. IREN just proved it — $2.8B signed in a week, 45% prepaid, on a book everyone was treating as a liability. What I’m watching: activation % and its slope. Merchant vs contracted split. H100-equivalents per MW (a Blackwell megawatt ≠ a Hopper megawatt). What I might be wrong about: H100 spot has fallen for 2 years straight, a wall of 2027 capacity all hunts tenants at once, and Meta reselling spare capacity would break merchant pricing overnight. I’m long the exact thing that just got repriced down. What am i missing?

1/ Spent the last few hours mapping every major neocloud -power, GPUs, contracts, backlogs. Once you see what the real constraint is, you see where the mispricing is. A thread 🧵

8/ The flipped version: Cipher. ~3.2GW pipeline. 300MW actually leased. ~90% uncontracted. That’s not a compute bet, it’s an option on lease signings. Applied Digital is the same shape -> 1.2GW contracted, 3.5GW+ being marketed.

7/ IREN is the cleanest merchant book at scale. ~$4B ARR target, ~85% contracted. That leaves ~15% merchant on a 150k GPU fleet. It bought 50,000+ B300s ahead of demand on purpose. Speed-to-compute IS the product. The market treats that 15% as risk. It’s optionality.

6/ Here’s the part nobody’s pricing. Almost zero finished compute sits uncontracted. It gets leased before it’s energized. CoreWeave’s backlog is $99B. Contracted capacity is a bond. Fixed rate, counterparty risk, no upside. Uncontracted capacity is a call option on GPU-hour prices.

5/ So the real bottleneck stack: 1.Time-to-power (energization, not acreage) 2.HBM / memory cost inflation 3.Capital to bridge spend → revenue 4.GPUs — allocated, not absent The constraint is #1.

4/ GPUs are the FAST part of the build. 6-18 months once you have electrons. Nvidia is investing in its own customers to place silicon — $2.1B into IREN, $1.5B sale-leaseback with Lambda, stakes in Nebius and CoreWeave. When your supplier pays you to take inventory, supply is not your problem.

3/ That gap is the entire industry. Secured power = land, interconnect rights, permits. Energized power = transformers, switchgear, substations, utility queues. 2-4 years. None of these guys miss because Nvidia shorted them on chips. They miss because a substation slipped.

2/ Every neocloud brags about “secured power.” IREN: 5GW Crusoe: 4.9GW Nebius: 3.5GW+ CoreWeave: ~3.5GW Sounds solved, right? Now look at how much is actually energized: CoreWeave ~29% Nebius ~5-6% IREN ~4%

It is mindblowing to me how few people realize that their lives and everything they know will change drastically in the near future. At this point, it should be pretty clear.

GPT-5.4 full at xhigh scored 51, exactly where Luna max sits today. GPT-5.4 costs $2.50/$15; Luna now costs $0.20/$1.20. In other words, roughly four months later, OpenAI is selling March’s full flagship intelligence at about one-thirteenth the token price.

@kevinxu IREN yesterday

@kevinxu @GrahamStephan @jackselby27 @TheICHpodcast The neocloud pivot of situational awareness was great dont understand why they were 4x levered

LEOPOLD’S SITUATIONAL AWARENESS FUND BLOWUP TIMELINE JULY 1: LEOPOLD’S HEDGE FUND SITUATIONAL AWARENESS REACHES $45 BILLION UP 450% YTD JULY 10: SK HYNIX US IPO MARKS TOP OF THE AI SECTOR JULY 10-20: MAJOR SELLOFF ACROSS ALL AI STOCKS, MOST DOWN 30%+ IN 2 WEEKS JULY 10-20: HIS SHORT POSITIONS START GOING UP CAUSING MORE LOSSES (HE WAS SHORT ADOBE AND OTHER SOFTWARE STOCKS) JULY 24: HE SENDS INVESTORS A LETTER CALLING THE SELLOFF A BUYING OPPORTUNITY AND ASKS FOR MORE MONEY JULY 27: CITADEL SECURITIES SAYS THE FED IS GONNA DO A SURPRISE RATE HIKE JULY 28: THE MARKET SELLS OFF AFTER CITADEL’S REPORT & LEOPOLD TRIES TO RAISE & BORROW MORE MONEY JULY 29: BANKS MARGIN CALL HIM JULY 30: CITADEL COMES IN AND BUYS HIS ENTIRE PORTFOLIO FOR A MASSIVE DISCOUNT LEVERAGE IS THE MAIN REASON PEOPLE GO BROKE IN THE STOCK MARKET ALWAYS BE CAREFUL