Pre Market Thoughts - 15 Jul 26 Yields unchanged across the curve despite a goldilocks print for CPI yesterday. Going into the trading day today we have Warsh testimony as with some other minor events. (You can view this on the event calender at http://terminal.gambit.zone) Asia trading got off to a brisk start. Asia index moves now are largely correlated with the memory factor. In KR, KOSPI is up 6.2% today while NKY is up 1.5%. $0660 is up 8% today (as of now, the spread between SKHY and $0660 is at about ~35%). $0660 is also 20% higher following the bounce from yesterday's level. Again - this is why you should not use leverage when trying to buy bottoms - buying bottoms are fundamentally betting on a mean reversal - markets do mean reverse but momentum persists in the short term. You need to be able to survive if you want to bid bottoms. $285A is also about 6% higher. Again, this stock has been remarkable in absorbing all the sell flow from Bain's exit. I wouldnt read too much into Bain exiting, because post their last sale at 9K JPY, $285A went on to still do a 5x. For memory, the story is simple. The KR and JP names are trading at 4-6X 2027 EPS, if the memory crunch lasts till 2028 (which most think so), then it would have earned back half of the market cap already. Post that, it becomes a story about how much demand there continues to be. I.E what you are underwriting is the unknowns in demand. IF you believe that robotics and agentic AI will consume alot more memory, then you will be able to underwrite a thesis calling for a structural shortage till 2030 or beyond. Next, I see the story for neoclouds in the similar vein as with memory. This was a sector that was hot in June, then it got tripped over by the entire momentum sell off. Yesterday, neos also sold off because of headlines that NY is banning DC constructions. First of all, in any DC construction - the land is cheap. Next, with fuel cells, DCs have more geographical autonomy and need not be constrained as much by power as before. Hence, it is illogical for neos to sell off on this. This reminds me of $0660 selling off on monday that analysts were lowering revenue estimates, because $0660 was signing LTAs and they had more HBMs vs DRAM. That to me, was stupid. Because the whole shortage in memory is on HBMs AND LTAs reduce the cylical nature of the industry which should increase the PE ratio given. However, because the market was gripped by fear, it sold off hard. Likewise, the price action of neos in relation to this news - is similar, we likely have people who are scared and will sell at the first sign of a negative headline. This opens up the opportunity for dip buying. However, like what i said, buying bottoms require you to not take leverage because it is hard to know where the exact low is. Fear has a way of spiralling. Buying the fear however, is what gets you paid. Why neos though? Simply because i see this as the next memory. Neoclouds at current levels of compute pricing are able to have a payback period of 2 years. Given the amount of financing that NVDA has now started to backstop, we can effectively have insane IRRs for neocloud projects. This will feed into EPS and real revenue. The last 2 years, compute demand was mainly driven by training. Yet with the new class of models, we are going to see inference demand starting to drive compute pricing. This is why past chips like the H100s are getting renewed 3 years on at 90% of their initial price. The H100s have already been fully paid back at this point. The business is insanely profitable at this stage. While all the VCs are funding frontier labs and AI - guess what is the picks and shovels play? Everyone needs compute. However, why there is an opportunity - is simply because neos are not well understood. Much like people thinking the top was in for memory in Q12026 - You have people like Jim Chanos saying that Neos need $4 in capex for every $1 of revenue. That is simply not true. Besides his whole argument being filled with wrong numbers and bad accounting (honestly, my opinion of him has dropped alot). At this point, every $2 of capex generates $1 of revenue. Of which depreciation is $0.4 on a 5 year schedule. Any time that the GPUs have economic value after 5 years is PURE profit. A100s have been out for 6 years - they are still being rented. H100s are out for 3 years, still getting recontracted in long term contracts at 80-90% of original rental rates. However, what i anticipate for how neos will move is likewise to CRWD. CRWD earlier in the year was sold off hard with the IGV basket. However, after all the sellers were absorbed, CRWD rerated up almost 2.5x. In neos, i see positioning as heavy. The funniest thing for markets to do is to make everyone puke out their positioning before moving up. Everyone wanted to buy NBIS at 290, yet at 190 everyone is scared. Markets never change. Good luck!(degentrading)