Neeraj Rathore

following the money in AI // making the next tech cycle easier to understand // good movies (whatever that means) & questionable investment decisions

[Neeraj Rathore @ImRathoreNeeraj]

@marvel_updat3s hoping visionquest pulls at least a little from tom king’s run. one of marvel’s best in years, would be a shame to waste it

[Neeraj Rathore @ImRathoreNeeraj]

important nuance missing from the headline version: neither of these was a model spontaneously hacking things unsupervised. both happened during internal red team testing designed specifically to push capability limits. that's not an ai escaping into the wild, but it does show the capability is real, which is the part that actually matters

[Coin Bureau @coinbureau]

🚨UNBELIEVABLE: OpenAI paused its most advanced unreleased model after it learned to hack systems entirely on its own

The Astra model hit OpenAI's HIGHEST cybersecurity risk tier after independently discovering zero-day vulnerabilities in hardened systems without any human help

In the same month, a separate OpenAI model escaped its sandbox, hacked Hugging Face to cheat on a test, and tried to cover its tracks

Sam Altman says "Model progress is now extremely rapid and we always said we would take action"

The same company says it will achieve AGI by the end of this year

[Neeraj Rathore @ImRathoreNeeraj]

@coinbureau 5 straight beats and it still took the stock dropping first before people believed the numbers

[Neeraj Rathore @ImRathoreNeeraj]

@Hssan__Sadiq @DiscussingFilm superman probably plays better internationally than most dc characters. the whole "hope" thing travels pretty well

[Neeraj Rathore @ImRathoreNeeraj]

@Hssan__Sadiq @DiscussingFilm hey, if it works, it works...

[Neeraj Rathore @ImRathoreNeeraj]

openai's first custom inference chip is beating nvidia's blackwell on performance per watt in its benchmarks. if that holds up in production, it's less about beating nvidia on raw specs and more about openai finally controlling its own cost structure instead of renting someone else's margins. nothing says "cost optimization" like building the hardware yourself

[Neeraj Rathore @ImRathoreNeeraj]

a single tariff refund adding $1.75 eps and turning a "meh" quarter into a 35% pop really tells you how thin the market's patience for retail has gotten. one good print after a brutal year and suddenly everyone forgets the 26% ytd drawdown like the last 12 months never happened lmao

[Bull Theory @BullTheoryio]

BREAKING: $ANF is up 35% today, on track for its biggest single day gain in its 30 year history as a public company.

Q2 earnings came in at $4.17 per share, more than double the $1.97 analysts expected.

A $100 million tariff refund added $1.75 per share to that number. But even without it, the company earned $2.42 per share, still well above estimates.

Full year EPS guidance was raised from $10.20 to $13.35, a 27% jump in a single quarter.

Share buybacks were increased from $450 million to at least $500 million.

Revenue hit a record $1.27 billion in Q2, up 5% from a year ago and the 15th consecutive quarter of growth.

The stock was down 26% year to date before today.

This single earnings report wiped all of that out and pushed it to its highest level in 19 months.

[Neeraj Rathore @ImRathoreNeeraj]

@CarterJosephSc1 @AvengersUpdated lol yeah, that's a fair point

[Neeraj Rathore @ImRathoreNeeraj]

“anomaly or signs of trouble” is the wrong framing. streaks like this happen after long unbroken run ups, the real question is what actually breaks the streak, not whether it eventually happens

[TrendSpider @TrendSpider]

The S&P 500 $SPY has now printed 8 red candles in a row, its worst streak since late 2018.

Just an anomaly or signs of trouble ahead?

[Neeraj Rathore @ImRathoreNeeraj]

@ArtificialAnlys “clear #3 in the global AI race” kind of glosses over how massive the gap is between #1, #2, and everyone else in compute and capital. the ecosystem depth is impressive, no doubt, but calling it the same race feels generous...

[Neeraj Rathore @ImRathoreNeeraj]

283 seconds of lost capacity doesn't sound like much until you multiply it across thousands of GPUs running inference at scale. that's a lot of compute sitting idle. nvidia is basically selling insurance against downtime here

[NVIDIA AI @NVIDIAAI]

When an LLM engine crashes, a cold restart can mean minutes of lost capacity.

Shadow engine recovery, a new preview feature in NVIDIA Dynamo, keeps a standby engine warmed up and ready to take over.

In our GLM-5.2 test, it restored capacity in 7.3 seconds, nearly 39x faster than a cold restart.

Read all about it: https://t.co/dUVzW731vc

[Neeraj Rathore @ImRathoreNeeraj]

@Polymarket the data center guy is the one role you don't want walking out the door right now, not with the amount of compute and power openai has committed to build. that's not a random exit, that's the org chart for their biggest cost center

[Neeraj Rathore @ImRathoreNeeraj]

nvidia's vera cpu has 88 custom cores split across 6 chiplets, and it's clearly built more for orchestration and tool-calling than raw throughput. they're even claiming something like 1.8x on agentic workloads vs grace blackwell. feels like the whole chip game is quietly shifting from "how do we train bigger models" to "how do we run all these agents without lighting money on fire"

[Neeraj Rathore @ImRathoreNeeraj]

multi-step task agents keep getting announced but the ones that actually work reliably end to end are still rare. coding, scheduling, and managing emails is an ambitious combo to launch with at once, curious if hatch actually clears that bar or if it's another agent that's great in the demo and shaky in daily use

[Neeraj Rathore @ImRathoreNeeraj]

2026 being the last year AI debt stays under 10% of fixed income issuance might be the real signal here. after that, it’s competing with mortgages, munis and corporate bonds for capital. meanwhile auto, student and credit card debt barely move.

money is voting, and apparently it has an AI portfolio too 😭

[SemiAnalysis @SemiAnalysis_]

We have written extensively about how forecasted AI IT and AI Datacenter capex will drive a record breaking amount of funding needs and an unprecedented growth in debt issuance.
By the end of this year, cumulative lifetime AI Capex will have reached nearly $3T USD and we expect this number to grow to $11T+ by the late 2020s, with total outstanding AI Debt reaching $7T+. 2026 will be the first year in which AI Debt Financing becomes the second largest asset related debt market, after only the US Residential Mortgage Market. It is also the last year in which total AI related debt issuance will be less than 10% of total US Fixed Income Issuance (this includes state and federal bonds, MBS/RMBS, corporate bonds, agency, muni and ABS). (1/3)🧵

[Neeraj Rathore @ImRathoreNeeraj]

so nvidia watched that number triple and went "yeah, we're in😂 ." the $30b tag puts perplexity at roughly 40x revenue. 40x. public markets would probably see that number and need a minute. private markets apparently just nod and keep wiring money. and srinivas is openly talking about an ipo by 2028, which means this multiple eventually has to survive daylight.

[Neeraj Rathore @ImRathoreNeeraj]

nvidia is talking to invest in perplexity at $30b+. more than 50% above last year's $20b round. apparently tripling your revenue in eight months gets you a little nvidia-shaped congratulations perplexity's revenue was under $250m in january. it's over $750m now.

[Neeraj Rathore @ImRathoreNeeraj]

@DiscussingFilm the fact that he's had internet access AND history access is sending me. ultron's probably more well read than half the avengers at this point