If you haven't already sold because of DAT overhang, quantum concerns, odds of clarity act passing reaching <20%, or relative price underperformance, what would cause informed participants to take down their BTC allocation now (outside of a broader macro sell off)?

I think the market ultimately has no choice but to go sell memory, long optical in the "short term." Actually, some hedge funds already seem to have this position on. There are three main reasons. 1. With Korean leveraged ETFs effectively dead, LPs are in a redemption rush, which could bring out additional sell on flow. 2. Nvidia is nerfing Rubin Ultra's HBM and responding with optics, tying multiple racks together, so that even if Rubin Ultra's per rack performance is not superior to Rubin, at the cluster level optics let the Rubin Ultra cluster hold an edge over the Rubin cluster. This holds even if Rubin Ultra's HBM nerf is a supply problem rather than a demand problem. 3. Consensus is forming that memory prices will peak within the next two quarters. Medium to long term I am still a memory bull, but short term I am somewhat bearish on memory. I currently have no memory position.

I believe semiconductor & memory stocks present one of the most attractive trading opportunities we've seen in the market in a long time. Previously, I talked extensively about the fundamentals of memory stocks and how demand for compute continues to be underestimated. Today, however, the trade offers an attractive setup following a massive positioning unwind. The positioning reset wasn't isolated to one group of investors. It occurred across multiple parts of the market simultaneously: Global hedge funds reduced semiconductor exposure from record highs. Korean retail investors experienced widespread liquidations. One of the most prominent AI-focused hedge funds was forced to deleverage. Importantly, this positioning reset wasn't accompanied by a comparable deterioration in fundamentals. If anything, the opposite happened. Throughout July, the fundamental outlook for AI infrastructure continued to improve. Hyperscalers reaffirmed aggressive AI investment plans, while memory companies continued to emphasize that demand remains structurally stronger than supply. Recent commentary from both hyperscalers and memory vendors suggests that these supply-demand imbalances are expected to persist well into 2027, driven by accelerating AI deployments and continued HBM supply constraints. In other words, positioning changed. The fundamentals didn't. Demand for compute continues to outpace supply. AI infrastructure spending continues to rise. Long-term supply agreements remain in place, pricing expectations are constructive, and the industry's long-term earnings power appears stronger than ever. To me, this combination of improving fundamentals, compressed valuations and a significant positioning reset creates one of the most attractive risk/reward setups currently available in the market.

500k people using agentic AI And we are already in a compute shortage

Prediction: Nvidia Nemotron's market share will 5-10x from now until end of this year. Open source AI is having a real moment. Nemotron is best set up to capture it, especially inside large enterprises. It won't show up cleanly on OpenRouter stats or similar leaderboards, because the install will live within on-prem environments. This is not because Nemotron is the "smartest" model per se. Many other open/close models are "smarter". But it is the most open! Weights are open. So are pre and post-training data, plus how the model is built. Only other model that is as open is the K2 models from @mbzuai, an academic institution. (See @ArtificialAnlys openness index, an increasingly important chart.) None of the open source Chinese labs open up beyond just the weights. Many do write great papers to share methodology and innovation. None share the datasets that go into pre/post training. This data transparency part is becoming increasingly important for large companies, who want to verify the models they deploy are not trained on data that could present security vulnerabilities, then post-train them further with its own data and IP. Only Nemotron fits the bill. It also has all the hyperscalers plus Palantir as partners to make deploying, post-training, plus continuous improvement on-prem in perpetuity manageable. This is huge undertaking!

UBS has sharply raised its memory price outlook for H2 2026. New forecasts show DDR contract prices rising +32% QoQ in 3Q and +18% in 4Q (previously +17% and +12%). NAND is seen at +30% in 3Q and +12% in 4Q. The firm expects DRAM to remain undersupplied through at least 2028. 2027 demand growth is projected at +36% YoY, still well ahead of supply growth at +19%. Memory industry revenue could reach $392 billion in 2026 and $513 billion in 2027. The key risk is affordability, hyperscalers need continued capital access to sustain the buildout. How long do you think this pricing power lasts before new capacity catches up? #DRAM #NAND #AI #Semiconductors

Samsung to Raise Q3 DRAM Prices by Up to 20%, AI Demand Remains Solid Korea's memory semiconductor industry is understood to be pushing to raise the average selling price (ASP) of commodity DRAM in the third quarter of this year by as much as 20% versus the prior quarter. With supply shortages persisting across the entire product lineup on the back of AI infrastructure investment, memory makers are interpreted to be extending a strategy of maximizing profitability. The pace of price increases will slow thereafter, but industry sources say an extremely high profitability trend will continue into next year as well. According to the industry on the 3rd, Samsung Electronics is in negotiations with customers targeting a Q3 DRAM ASP increase of up to around 20% versus the previous quarter. DRAM prices have shown a sharp upward trend, driven by aggressive AI infrastructure investment from global big tech firms. This is because supply shortages have intensified across the entire product lineup, spanning not only server DRAM and high bandwidth memory (HBM) but also low power DRAM (LPDDR), which is drawing attention in the AI inference space. Samsung's DRAM ASP increase is especially pronounced relative to SK Hynix. The industry assessment is that commodity DRAM, which carries high price volatility, accounts for a large share of Samsung's total output, and that the company has been the most aggressive in raising prices. In fact, Samsung's first quarter DRAM ASP rose by the low 90% range versus the prior quarter. The second quarter is estimated at around 50 to 60%. Furthermore, it is targeting an increase of around 20% in the third quarter as well. SK Hynix, which has a relatively high share of HBM production, is estimated to come in somewhat below this level. A semiconductor industry official said, "Samsung is negotiating very aggressively on pricing in the third quarter this year. We understand it will also raise LPDDR, which has recently seen severe bottlenecks in both server and mobile, by more than 20%," adding, "That said, it is not certain whether customers will accept all of this." DRAM prices are seen as highly likely to remain stable going forward. Although the pace of DRAM price increases is gradually easing, the share of long term agreements (LTAs) signed with key customers is steadily expanding. For example, Micron of the United States disclosed through its earnings release late last month that it had signed a total of 16 long term agreements with customers. These contracts carry binding commitments to purchase a certain volume and set a price floor that guarantees very high margins. This is interpreted as reflecting customers' judgment that memory supply and demand will remain tight over the medium to long term. The recently raised possibility of Meta entering the cloud business is also not expected to weigh on memory demand as a negative. Some had read Meta's move to sell its internal surplus computing resources externally as a sign that its AI production capacity may already be sufficiently built out. However, Meta has maintained an aggressive investment stance, in April raising its annual AI infrastructure investment plan from an initial 115 billion to 135 billion dollars to 125 billion to 145 billion dollars. Another industry official explained, "With the expansion of LTAs that include price floors and HBM price renegotiations, there will be no sharp decline in the DRAM market next year either," adding, "In Meta's case, it is more accurate to view this as a way to efficiently utilize its internal computing resources."

UBS raises its DRAM and NAND price forecasts DRAM prices are now expected to rise 32% QoQ in Q3 and 18% QoQ in Q4. NAND prices are expected to rise 30% QoQ in Q3 and 12% QoQ in Q4.