Once again, the GPU depreciation bear thesis is dead. Elon on the $SPCX call: "The value of compute per watt will increase over time, not decrease... that's why you see things like even old computers like H100s renting for more than they rented when they were new." When supply is this constrained, GPU rental prices are set by the value of the marginal token, not the replacement cost of the chip. Models keep getting better, and better models raise the value of every installed watt faster than new GPUs deflate the old ones. Various datapoints all stack up: -> $SPCX says compute contracts are renewing "at higher dollars per watt, not lower dollars per watt" -> 1-year H100 contract pricing went from ~$1.70/hr to ~$2.35/hr between Oct 2025 and Mar 2026 (a 2022 chip!) -> $CRWV signed an A100 contract running into 2029 (a 2020 chip) and said older-gen pricing is at / above a year ago, and raised list prices ~25% in July. -> $NBIS's first Blackwell auction cleared 15% above its previous high and as a reminder....they told customers H100 on-demand goes up ~17% (from Oct 1). -> Zuckerberg: $META is getting offers for its compute "at a significant premium over what we paid for it". -> Lenders are now lending against GPUs out to 2030 e.g. $NBIS borrowed against its GPUs at SOFR+250 in July. Feels like the 2-3 year GPU useful life bear thesis keeps getting setback. Ofc, the one thing that ruins the party is the S/D gap closing -- Rental prices fall back toward replacement cost + older GPUs get repriced on perf/W. Which won't happen anytime soon.

Elon: "In a nutshell, what this implies for SpaceX annual recurring revenue by the end of (20)27 is somewhere between $500 billion and $1 trillion." Bruh?????

Great earnings from $SPCX Revenue: $14.6B vs Est. $13.0B AI Revenue: $7.56B, EST. $6.3B EPS: $0.35 vs Est. $0.16 EBITDA: $8.8B vs Est. $6.6B CAPEX: $19.59B, EST. $19.6B CFO Commentary: We delivered another strong quarter in Q3. Revenue grew an impressive 161% year-over-year, fueled by continued momentum in Connectivity and a clear inflection in our Super Intelligence business. The combination of our launch leadership, recurring Starlink revenue, and the significant expansion of our SI infrastructure continues to drive meaningful operating leverage. We are building the infrastructure of the future with discipline — focused on capital efficiency, strong returns on invested capital, and maintaining a robust, investment-grade balance sheet — while moving with the velocity to capture the opportunity in front of us.

The Anthropic IPO roadshow is expected to begin the week of November 9th. The public S1 has to be filed on EDGAR at least 15 days before the roadshow. There are no EDGAR filings on weekends. A November 9 Anthropic roadshow start would require that the public S1 filing take place no later than Friday. The S1 will validate what Anthropic reportedly told key investors on October 14th: Q3 revenue was $16.8 B ($70 B ARR) and they expect their year-end ARR to be $110 B. OpenAI completed a $122 B funding round in March. The company met with long only funds and sovereign wealth funds on a new funding round of $30 B. OAI reportedly needs the capital ​to fund infrastructure commitments and bolster its balance sheet in case its IPO is delayed. It wants to complete the round before Anthropic pre-IPO financial disclosures. OpenAI says GPT-6 Astra accelerated its growth and pushed its ARR to $50 B. They expect a $70 B ARR by year end. Anthropic seeks a $2 T IPO valuation on $110 B year end ARR, an 18.2X multiple. Last week its investors pressed for a $1.5 T valuation, a 13.6X multiple. Anthropic reports gross revenue. OpenAI reports net revenue. Anthropic pays 16% to sales partners. Adjusting $110 B year end ARR for that puts net ARR at $92.4 B. $1.5 T IPO valuation then has 16.2X multiple. 16.2 X multiple and OpenAI year end ARR puts its IPO valuation at $1.1 T. Labs should take what they can get. SpaceX reports after the close. Total revenue is expected to be ~12.8 B, with AI revenue half of that. Musk has shorted a monthly release cadence of from scratch new Grok models. GrokBot is a hit and may drive a SpaceX AI revenue beat. Grok 4.8 is due this month, 4.9 in November, and 5.0 by year end. A SPCX AI beat and Grok 4.8 release before OpenAI new funding round could unnerve its backers. Ditto for Anthropic investors if 4.9 drops before its pre Thanksgiving IPO.

Current stock prices will look incredibly cheap once every single AI-related company raises EPS guidance over the next few weeks. It's crucial to remember that the current market choppiness is mainly due to macro...not because the AI trade is "broken".

For AI/Semis stocks, it feels like every earnings season becomes signficantly more important than the last. This time is the same, but for different reasons which has led to market chopiness: - Macro instability (Iran, oil, yields, inflation) - Midterms - $CRWV credit stress - AI ROI concerns via frontier lab ARR's (& pending hyperscaler earnings on top). Today: $SPCX Wed: $META, $TXN, $LRCX Thu: $BE, $WDC, $NOK Let's see if good EPS will be enough to turnaround sentiment.

If my prediction ends up being right, I will parade it around for the next 12 months. (If it's wrong, I will delete the post)

Hedge funds are now >8% underweight tech vs. the MSCI World Index... Since the end of July, net exposure has come down almost entirely through new shorts while longs have remained steady... In other words 1 -- HFs haven't completely dumped long positions. In other words 2 -- HFs have hedged (who would've guessed it)...rather than totally capitulating on tech. In other words 3 -- Semis positioning is probably much cleaner for Q3 earnings vs. Q2. (Not totally clean though, Semis L/S @ ~3x vs. SaaS L/S @ ~1x). Interesting note via JPM today -- "We don't expect the rotation out of Tech to end abruptly, though its pace could moderate."

Fulltech Fiber in Taiwan ($1815.two) has almost doubled since TSMC presented this slide on 11 August.

Bank of America: "We continue to lean risk-off next 3-4 weeks given 3% real yields, need for Fed to hike Oct 28th, US/Iran/Russia/Ukraine/oil, rising risk of "wave-like" DEM sweep midterms, price action AI spenders & lenders." Based on these (and other PB) comments, it seems like there's currently ~low confidence in Q3 earnings acting as a big enough catalyst to re-rate Semis...

Feels like shorting Taiwanese AI stocks is probably a good hedge right now. Positioning remains extremely elevated = highly exposed to short-term downside. Just look at how the market responded to $TSM's awesome earnings/guidance....-2.5%.

Korea / KOSPI positioning: Corporate buybacks have been the only meaningful source of inflows in recent weeks, while all major investor groups have been net sellers. Foreign investors have led YTD outflows, primarily driven by selling Samsung Electronics and SK hynix. Chart 1: Leveraged ETF AUM has collapsed back to the Iran war market lows in March. Chart 2: Trading activity in leveraged ETFs has collapsed since new regulations were introduced in July. Chart 3: Margin loan balances are below the June peak, while margin loans as a % of market cap are at ~0.6%. This highlights that market caps are still relatively elevated. [GS Research]

Memory risk/reward seems attractive again. IMO, memory stocks are currently priced as if supply > demand by 2028. This seems highly unlikely - feels like the market is currently ignoring the fact that Chinese memory players will face significant hurdles to achieving mass sales. That said, I wonder what tailwinds drive memory stocks higher from here? Maybe it's as simple as the market being more risk-on than it is now, rather than one single catalyst.

BofA on Memory: "Overall, the recent correction in memory stocks appears largely disconnected from memory fundamentals and reflects excessive concern over an AI-memory boom-bust cycle. We also see the potential for the global memory TAM to double from near-US$1.0tn in 2026 to US$2.0tn+ in 2030, with OP potentially reaching US$1.0tn and the industry's market cap consequently reaching US$10tn, versus only a few trillion dollars currently." $DRAM