Exited 80% of $MU long from $224.5 to $232.

While memory supply is likely to remain tight throughout 2026 and probably also 2027, I think the tight supply is going to be a problem soon.

You can think of this as a classic prisoner’s dilemma, where the first party to betray the “agreement” gets all the benefits and the last party suffers.

Yes while memory makers are experiencing supernormal profitability now due to sky-high prices, their growth is also limited due to sold-out supply.

So far, the memory makers have moved in tandem at least in public by not expanding supply. However, there will definitely be pressures growing, from both internal and external, to expand capacity to capture the excess demand, at abnormal prices.

However, I think it is inevitable that the prisoner’s dilemma is eventually broken by a few greedy players that will expand their capacity to capture all the excess for themselves. YTMC and MU have already hinted at this. This will then spur on an arms race for capacity expansion — which will ironically be the death knell for the memory supercycle.

The counterpoint here is that memory might not be as cyclical as previous cycles, due to explosion in AI demand, and thus capacity expansion will not matter as much. However, this needs to be proven out first before we decide on being massively long memory again as I don’t want to jump to the conclusion that “this time is different”.

Markets are forward looking, so just like how the share price of memory makers started moving up even before memory (DRAM, NAND) spot prices moved up, the share prices will stall even before any official announcement of capex expansion or decline in memory spot prices. Hence I believe this is the reason for the weakness in the memory prices these past few weeks despite the bullish news.

So what are we long instead? I have shifted memory exposure to $LRCX @ $147.5, $AMAT @ $230, $KLAC @ $1149. These are semicap players that will benefit from capacity expansion from memory makers, so also an indirect exposure to the growing importance of memory - but one that should outperform in the coming months.

Shorted $MU @ $225 and $SNDK @ $227

“SK Hynix plans to ramp production of its 1c DRAM about 8x in 2026 as AI demand shifts from pure model training to large scale inferencing.”

And the race to the bottom begins

Shorted NVDA and neoclouds’ spike. Fading the rally as explained above. Mid-size positions, leaving some ammo to add just in case of further spikes

$CIFR @ $16.93
$IREN @ $52.42
$CRWV @ $80.02
$NVDA @ $191.54

Updated short basket. Avg price is entry price, last is current price. Largest shorts at the bottom.

TPed on 35% of the new shorts today. Have a feeling we might bounce

CIFR IREN CRWV SNDK MU

Wow, went to sleep after taking partial profits on the shorts and woke up to a sea of red.

Bad news is market is still really weak, but good news is our core longs (INTC AMAT LRCX COHR LITE FIX 9988.HK) are starting to look very reasonably priced.

I’m out of all low caps except a few Chinese moonshots, and mostly in high conviction forward-looking plays, so I think (and hope) ultimately we will be alright!

$BMNR starting to look attractive as a trade.. maybe just a bit more capitulation.

Thinking more on this subject, I think point 3) above is wrong. In fact, I want mgmt to absolutely HATE magic internet money, so that they will be more likely to dump the coins to buyback the stock when the company is trading at a discount to NAV.

I think there will be a race to dump the coins soon. Again, prisoner dilemma at work here: the last ones to dump the coins will have to sell at the lowest price. Tom Lee is an intelligent tradfi guy, I think he will cave first.

I guess counterpoint is that they own too much of the ETH supply to signal a shift to dump the coins without immediately crashing the market.. leaving them unable to exit their remaining holdings. Doesn’t help that the blockchain is completely transparent.

Closed $SNDK and $MU shorts for 8-10% gains

If the market can reclaim green by open, after the pre-market panic right now, I would be much more constructive on the bull case. Signals capitulation and changing of hands to me.

Kind of the opposite of fading the NVDA gap up

Closed $IREN & $CRWV shorts earlier for 10-15% gain

So far so good. Let’s not jinx it.

Also added more $LITE $AMAT & $COHR earlier too

Very preliminary thoughts, but I think a good framework to be long/short the AI “bubble” over the coming months could be:

Long Google and TPU related stuff
Short OpenAI, NVDA and GPU related stuff

Back in $MU long @ $211

This was created by Google’s Nano Banana Pro. One of the reasons for re-entering $MU long

Trimmed 40% $COHR @ $150 to enter fresh $CIEN long @ $186.5, forming the optics basket alongside $LITE and $COHR.

Optics is clearly the next major bottleneck in scaling AI training and inference, as GPU/TPU clusters get larger and DCs get further apart — all while latency and bandwidth requirements rise.

Trimmed 30% of $MU and allocated more % of portfolio to re-enter $SNDK long @ $225.5. Don’t think the memory cycle is over yet on second thoughts, and SanDisk is one of the highest beta plays.

So far we have traded the memory stocks quite well, hoping it continues.

$SNDK added to S&P500, +9% AH

The timing on this… 🤗