Andrei Stan

What hardware can run Kimi 3 well? I read 8 H100s but that seems not much for ~3T parameters, no?

BREAKING: Michael Burry has disclosed new positions. He is short: Nvidia $NVDA Tesla $TSLA Caterpillar $CAT Applied Materials $AMAT

"GFHK: MRVL is expected to generate $2.8B / $13.0B in revenue in 2027 / 2028, respectively, driven by Google CXL-related business." Did u listen to us, anon?? I'm pretty sure we were the first to break it down

Pre Mkt Thoughts - 29 Jun 26 Asia markets had some early jitters before coming back to almost unchanged. KOSPI, NKY both traded with very strong correlation, selling off early in the morning before inching back into the close. Lets talk about the memory names. $285A had a dip to 82.5k levels before ending about -4% for the day. This represents a high r/r level for longs. $285A also has triggered some of the prior patterns i noticed in its price action. Meanwhile on the other side, news broke on $SAMSUNG and $SKYNIX planning to build 2 new massive chip fab sites in South Korea's southwest region as part of a "national project". Let's rerun back to Leopold's Situational Awareness article - the manhattan project. The early innings are coming through. Countries are treating this as existential. The sovereign buildout thesis is manifesting in real time. Aside, direct implications for Korea is that the rally in KOSPI should increasingly broaden out. I am keeping my eye out for Korea small caps. I believe that for small accounts, these can turn out to be some of the best opportunities. Personally, i am afraid of the illiquidity in those tickers and will probably sit it out. Coming into the US session, i indicated in my Sunday note that for end of quarter rebalancing, there is about 165b of equities to be sold into the close. Some have asked, how i would allocate into such a situation. There is no right answer. It hinges on 2 key pivots. 1 - What is your risk appetite? 2 - How much do you think flows have been front run? For those who are aggressive, that could mean allocating into one clip right into the close on tuesday. For others, that could mean a 2 day twap into the markets. I have been banging the drum in the last few notes that i see the best technical set up into July for longs. Let me take the time here to list it out again. First of all, the fundamental picture looks extremely strong. We had 2 articles dropping over the week. First was news that $AAPL was lobbying for the white house to allow them to buy from $CXMT. While this might sound bearish for memory. Please remember that the memory market is global. Paraphrasing Semianalysis Dylan - "It literally doesnt matter? CXMT cannot satisfy chinese demand this decade". Next, let me emphasize that $CXMT is selling at ~5% discount. After you account for quality differences, there is almost no difference. So why is $AAPL doing this? It's not about price, its that the supply is simply not there. If anything, this is a direct affirmation that indeed...supply is simply not there. Next, an article that $GOOG is capping $META's use of Gemini AI. How insane is this? $META cannot get enough compute that it is sourcing it from $GOOG. Crazier is that... $GOOG simply cannot meet the demand. Look, the hyperscalers will not stop capex any time soon, even as their equity prices take a hit. Because they see the revenues on the horizon. CAPEX hits the bottomline on t0, revenues and backlog hit the bottomline in the future. As these revenues arrive, the hyperscalers will start making money hand over fist. Remember...Anthropic's gross margins on inference is ~70%. By building now, they are securing a moat for the future. Where does this lead to? Again, the 2 themes i keep rehashing is compute (neos) and memory pooling. While i think memory will continue to be strong, the easy part of the rally and the valuation gap has been covered. On compute, $NBIS stands out as the leader of the pack. $SHAZ is an interesting up and coming player backed by Situational Awareness. i am expecting a 13G to hit the timeline tonight. On memory pooling, tickers include $ALAB, $CRDO, $PENG, $MRVL. On $CRDO, friday's sell off was very much influenced by the Russell rebalancing (as per my sunday note), i think the 240s area offer exceptional value for risk reward. In my notes last week, i talked about $BABA - I strongly think that now that even Burry has sold $BABA - sentiment cannot get worse than now. These levels also offer good risk reward. Sep calls especially are cheap. Good luck!

MU is not NVDA!! MU goon crash to 700 soon (now 1200). Remember we had this at 80!!! RAM is NOT GPU fools !!!

ConnectX-10 NVLink-C2C なるものが登場しています。 Vera の PCIe は、Gen6 なので、1.6Tbps と想定する ConnectX-10 は Gen7 。 となると、Vera + ConnectX-10 構成の BlueField-5 では、1.6Tbps を流せません。そこで、ConnectX-10側にNVLink-C2Cを追加することで1.6Tbpsを実現するの？

Great tidbits from Lip-Bu in the pod 1) Almost became CEO in 2021, he declined 2) Lip-Bu boosting CAPEX (he said before that's tied to external customers) 3) Intel building an ARM CPU or making ARM AGI? 4) Lip-Bu hiring top CPU architect soon 5) 10 year plan, big company $INTC