Sabitlenmiş Tweet
Kevin Simback 🍷
7K posts
Kevin Simback 🍷
@KSimback
COO @delphi_labs - building + investing in AI and crypto. Ex @IBM, @McKinsey, @CarnegieMellon, reformed CFA
Puerto Rico Katılım Mart 2015
905 Takip Edilen18.7K Takipçiler
@ICYDuali Many places in Piedmont/Lombardy that didn’t have AC, most restaurants didn’t seem to have it and we had to carefully select Airbnbs after one we rented didn’t have it
English
Just departing Italy amidst this heat wave, I really don’t get why AC is even a debate
Literally everyone - locals included - are complaining
Sweltering indoors just makes people irritable, install the AC already
English
We can now clearly see the memory demand for LLMs is growing exponentially on the heels of agentic coding and agent loops
But we’ve not really factored in memory for humanoid robotics
The context windows need to be huge to handle multi-modal data
So much more memory!
Serenity@aleabitoreddit
Very interesting statement today: $MU CEO predicts a multi-decade memory demand cycle driven by humanoid robots. "Humanoid robots, he says, will require roughly ten times more memory than today’s Level 2+ autonomous vehicles." "And that demand wave is set to begin before the decade is out." Something as well as was "Over time, we expect the value of on-device AI combined with pent-up unit replacement demand to drive memory demand growth" Which is also another trend (Apple Intelligence is currently dog, but I'm sure we'll see innovations with localized/edge AI). Feels like all the industry leaders from $TSM Chairman, $TSLA Elon Musk, to $MU CEO see humanoids as the next major trend so physical AI is probably next. I wonder if the world is going to have enough memory. Or if we'll see enough breakthroughs to shrink memory usage.
English
@KSimback Are you still working on the development of Looper ?
English
Great explainer on Looper - a design tool to help you build better and more cost effective loops
Give it a try!
Shann³@shannholmberg
a poorly designed loop burns tokens and hands you more slop design the loop before you run it, then hand the spec to claude code, codex or any other harness mapped everything you need to know about looper in the image below
English
We just crossed $100M annual run-rate. I know many AI companies are capturing much more $$$ these days, but still proud of the milestone!
Maximizing short-term revenue has never been our priority. In fact, we're proud to manage to store and serve hundreds of petabytes of models and datasets while keeping HF free and open-source for 97% of our users. As a platform, we’re happy to hopefully create orders of magnitude more value for the community than what we capture. To me, that’s the very definition of a platform.
And it has helped us build one of the most loved platform in tech, with network effects, a defensible position and a sustainable business which is quite unique in AI.
Many many thanks to all the community members for building with us, we wouldn't be anywhere without you! Can’t wait for what’s next, especially as more companies start to see the value of open and local AI! Next milestone $1B?
English
@mdudas AI loves to use “substrate” and “cleanest frame” when explaining things
English
what are your favorite words or phrases to confirm that an e-mail, memo or presentation was ai-generated?
mine is "through-line"
English
Paris CDG Airport is built like a hothouse. Even in winter, when it’s sunny, you burn up.
English
I see new announcements like this almost every day, and I have no doubt the product helps solve problems companies face deploying AI agents
But I think distribution is going to win out here
Building a great AI product for companies is the easy part, getting it in the companies is hard
Andy Berman@berman66
Today, we’re announcing Runlayer has raised $30M from Felicis and Khosla Ventures to help companies go all in on AI. Runlayer is the golden path for AI: enablement, security, and control in one platform. So, how does it give your team the right tools for AI? 🧵
English
@pradeeepk Sounds like a good take for someone who’s sidelined
English
I have no position in memory
I read $MU SCAs details & it looks bearish to me
$MU is entering 5y deals with current market price as ceiling
$MU is locking in current market prices for 50% of its revenue
is this really bullish?
English
sharing my personal library on evals 1/n
i put together the highest quality blogs, podcasts, papers, and projects on evals. additions are welcome!
English
Just finished digesting the $MU earnings and transcript
It was the mother of all beats - crushed it on every metric, guided higher, and locked in demand via strategic agreements
Everyone playing with AI agents knows how memory hungry they are, and we’re just scratching the surface of adoption
So much more demand coming and nothing short term that displaces HBM, this is the most obvious play since NVDA circa 2023-24
Many will look at the run-up this year and say it’s too expensive - based on FY2026 guidance it trades at ~10x earnings, not revenue, EARNINGS
These are not bubble metrics, just pure business demand
I’d be a buyer on any dips
Kevin Simback 🍷@KSimback
“we need a metric fuck ton of memory” Yep We’re just entering an era of agentic loops becoming normalized business practice, the amount of memory this will consume is orders of magnitude larger than what exists today
English
@ceterispar1bus If you’re in prediction markets, pivot to memory
English
are they launching kalshi memory or something?
*Walter Bloomberg@DeItaone
KALSHI IN TALKS TO RAISE FUNDS AT $40BN VALUATION: FT
English
Micron $MU absolutely crushed it
AI needs so much memory, regardless of whether frontier or open source wins - memory is what makes it work
Algofinix Inc@algofinixai
MICRON $MU Q3’26 EARNINGS HIGHLIGHTS 🔹 Revenue: $41.46B (Est. $35.5B) 🟢 🔹 Adj. EPS: $25.11 (Est. $20.4) 🟢 🔹 Adj. Gross Margin: 84.9% (Est. 81.8%) 🟢 Q4 Guide: 🔹 Revenue: $50.0B ± $1.0B (Est. $43.4B) 🟢 🔹 Adj. EPS: $31.00 ± $1.00 (Est. $24.3) 🟢 🔹 Gross Margin: 86% (Est. 83%) 🟢 🔹 OpEx: Approximately $1.65B Q3 Segment Revenue: 🔹 Cloud Memory: $13.76B (Est. $10.69B) 🟢 🔹 Core Data Center: $11.52B (Est. $6.8B) 🟢 🔹 Mobile & Client: $11.52B 🔹 Automotive & Embedded: $4.63B (Est. $3.51B) 🟢 Other Metrics: 🔹 HBM4: In high-volume shipments for lead customer platform 🔹 HBM4E: Volume production expected in calendar 2027 🔹 G9-based PCIe Gen6 high-performance SSD: In high-volume production 🔹 High-capacity 245TB QLC SSD: Shipments commenced Financials: 🔹 Adj. Gross Margin: $35.199B 🔹 Adj. Operating Income: $33.681B 🔹 Adj. OpEx: $1.518B 🔹 Investments in CapEx, net: $7.1B 🔹 Cash, marketable investments, and restricted cash: $30.2B 🔹 Operating Cash Flow: $25.39B 🔹 Adjusted FCF: $18.3B
English
For those wanting to dig deeper into the Inference market, this is well worth a read from @KSimback
x.com/KSimback/statu…
Kevin Simback 🍷@KSimback
English
@0xgilbert Have not but heard latest Cohere model was good
Have high expectations for @NousResearch models as well
English
@KSimback Have you tried GigaChat Ultra at all? Mistral and Cohere are really the only non-US/China models I can name off the top of my head.
English
If you’re not US or China, I don’t know how you sit back and become dependents on them for the intelligence that will define the future
You’d think it would be a national priority to have a strategy here
Of course, easier said than done - you need big $ and talent to compete with the frontier labs
The easiest first step is to build sovereign data centers and host the open weights models or post-trained versions - at least you own some form of the intelligence
The UAE and Saudi Arabia seems to be going this path and making moves towards training
The most impactful next step would be focusing on decentralized training
If you can’t build a cluster capable of training a capable model, decentralized training may be a viable option
Not for today, but as an option to hedge the risk of being locked out or priced out of the US and Chinese models if that day ever comes
Doing a deep dive on decentralized training with a report coming out soon
Let me know if there are any projects you want to see covered, I’ve got a pretty good list already but open to include others if they’re legit
English
@bnivanovX Anyone operating at the forefront of AI is likely challenging their cognitive capabilities quite a bit, there’s a constant learning motion you have to sustain
English
@KSimback I would also say you need to assess the "friction" they also put in place to protect their teams from long term cognitive-degradation!
English
If you’re doing diligence on a new startup, you now need to assess their “AI nativeness” alongside team, product and market
Why? If they’re not at the forefront of using AI, it puts them at risk of not executing fast enough
It’s not a hard pass filter, but it is signal - here’s a few early tells:
1. Did they give you a .md file, alongside a deck, that you can feed to your LLM?
2. Did they build an interactive HTML deck or just send a PDF/docsend?
3. Do they talk about proprietary skills, data, evals or methods they’re using to build their product?
4. Do they ship significant product updates between meetings?
Not an exhaustive list, just some anecdotal signs that I’ve noticed
English