Adrian Morris

Remember what I said re: $MSTR and MSCI If | when a de-listing happens, we may see a drop after an official announcement. But likely priced in, not a terminal event. Ignore the FEAR | FUD

一台 Mac Studio 加一张 RTX 5090，用雷电扩展坞连起来当一个推理引擎，跑 Qwen3.8 Flash-Next 单路约 210 tok/s。 项目叫 Apple NVIDIA Sidecar，11/8 刚开源。机器是 M5 Max 64GB，5090 装在扩展坞里，权重用的是 NVIDIA 官方的 FP4 版本。 同一个模型，别人公开过的数字大概是这样，测法各不相同，只能看个量级。 1. 一台 DGX Spark 约 43 2. Jetson AGX Thor 约 47 3. M4 Max 峰值约 83 4. 5090 加 128GB 内存的 PC 跑 NVFP4，114 到 116 5. 72GB 的 RTX PRO 5000，中位 159 短板作者自己列得很清楚。 1. 8K 提示的 prefill 约 1,900 tok/s，好一点的 PC 能到 3,900 以上 2. 一次只服务一个会话，1 到 16 路冷启动，总输出一直在 61 tok/s 左右 3. 雷电链路实测单向 6.79 GB/s，反向 5.53，标称 7.88 4. 驱动是作者本地改过的，没放进仓库 想复现，得先自己把驱动搞定。Mac 和 N 卡拼起来跑同一个模型，这条路算是有人走通了。

The Jensen Huang story keeps getting crazier. US intelligence reportedly discovered that Nvidia’s CEO was romantically involved with a suspected Chinese spy. Here's where things get insane: 1.) The woman, Alice Huang, was an executive at Megaspeed, a company under investigation for allegedly helping Chinese firms circumvent US chip export restrictions. 2.) US intelligence believes she was recruited by China's Ministry of State Security, its primary intelligence agency. 3.) Intelligence officials reportedly suspected her objective included gathering information about Huang and potentially influencing Nvidia to push for looser US chip export controls. 4.) Huang has spent the past year repeatedly lobbying Washington to loosen those exact restrictions. 5.) US officials reportedly prepared an intelligence memo detailing the relationship and circulated it among senior government officials shortly before Trump's summit with Xi Jinping. 6.) Trump had reportedly already been briefed on the suspected espionage operation and chose not to act on the information. 7.) Huang has met with Trump repeatedly, reportedly even accompanying him aboard the presidential aircraft on a trip to China. 8.) Megaspeed's Malaysian affiliate reportedly purchased at least $4.8 BILLION in equipment containing Nvidia chips between June 2024 and February 2026. 9.) Federal authorities have been investigating Megaspeed since 2024, with the Justice Department reportedly taking over the Commerce Department's investigation this year. 10.) Investigators reportedly found Nvidia equipment in boxes that appeared ready for shipment rather than installed in operational data centers. 11.) Alice Huang's name was reportedly removed from Megaspeed's corporate documents as scrutiny intensified. Her continuing involvement with the company remains unclear. 12.) Nvidia maintains that it investigated Megaspeed's chip deployments and found no evidence of illegal diversion. 13.) Three congressional committee chairmen are now demanding that Huang appear for a closed-door briefing on November 18. We're talking about the CEO of the world's most valuable company, a suspected Chinese intelligence operation, billions of dollars in advanced AI chips, and the US government's efforts to control China's access to the technology. There is still no public evidence that Huang knowingly cooperated with Chinese intelligence or that the alleged operation successfully influenced US policy. But the questions surrounding this are extraordinary.

Firstly, I should note that my latest v20 models explain 72.8% to 85.2% of variance. Secondly, I find it interesting that (the A.I.) concedes the short term variance results (using outdated 63% R2 stats) and then pivots to an unproven "longer term structural effect” that depends on BPS or CEBE being the factors that set value or the multiple. However, there are a lot of flaws there, specifically around the claim that dilution effects reducing BTC per share (or CEBE) lowers the mNAV or the price for any given total NAV, even if it doesn’t show up in the regression. Also, declines in BPS and mNAV “compression” are used as evidence of a structural effect that my variance work can't account for. Ironically, BPS has been one of the most consistently modeled factors in my work and I think the A.I. is assuming that  BPS | CEBE are valid metrics for what the market is pricing; or for the “residual claim” the market is valuing, but they aren't. After today's latest run (with a fresh as of 10-30 date), when assessing BPS growth as a factor across all models from 2021 to 2026, its price variance contribution is just 0.04% to 0.33%. Getting more specific, in my full Capital Structure model (which explains 85.2% of price variance), the variance contribution from BTC Per Share is 0.04% (+R2=0.000) and it is statistically insignificant (t=-0.948,p=0.3438 ), just as the ATM itself is statistically insignificant (t=0.961,p=0.3375 ) with a 0.67% attribution. Even in the broader 2021–2026 models where BPS shows significance, the t-stat is negative (t=-5.73 to -7.01) which in plain terms means that generally, as BTC Per Share went up, share price went down. I think that is a rather concrete answer about whether or not the KPI shows up as some missing long-horizon driver of price. @Strategy itself has defined BTC Yield as a narrow KPI for per share accretion on $BTC holdings and has repeatedly stated that it isn't a measure of valuation, financial performance, or returns. Specifically, Strategy’s own Q3 2026 financial release (filed October 29, 2026) explicitly states in its KPI disclosures that BPS (in Sats) "does not represent (i) the Company's ability to satisfy its financial obligations, or (ii) the Company's book value per share" and that BTC Yield "is not a measure of the return on investment the Company's shareholders may have achieved historically or can achieve in the future by purchasing stock of the Company". BTC Per Share is essentially an accounting reality that doesn’t represent  a redeemable claim nor a redeemable residual after senior claims. If we move onto CEBE; this is an attempt to net out senior claims, but it serves as just another formulation of the same accounting residual and is not what the market is forward pricing. Based on years of analysis what I see tells me that the market is pricing the BTC capital base, the optionality within the equity, the adversarial positioning around the equity and the sentiment | positioning around the resulting structure (mNAV). With mNAV acting as a multiple that is assigned independently of the current BPS | CEBE reality. I think this is somewhat  evidenced by persistent rises in BTC Per Share (for many BTC-TC’s) while share price fell. Now the question that I have is this: If the company that created the metric, defined the metric, and was the first to consistently report the metric, is saying that it's not a measure of valuation or related to stock price performance; perhaps there is a reason for that. No? At the same time @saylor himself has described mNAV expansion as driven by market sentiment | confidence, with BTC Yield and BPS rising as a result of that sentiment rather than being its cause. The data, the results of my modeling, and the official corporate framing are well aligned. Coming back to the analysis that I have done in more detail, the modeling isn't a simple “correlation” analysis and it isn't missing a slow structural effect that is captured in BPS | CEBE or dilution effects . Rather, using orthogonalized factors, Newey-West HAC standard errors, rolling 90D | 180D Shapley windows, and BVAR sign restrictions, its showing that the factors that actually move price variance are overwhelmingly $BTC direction (accounting for 41.2% – 51.2%), followed by options positioning (GEX, DEX, Max Pain Distance, PCR), shorting pressure and related positioning, volatility (VIX and MSTR IV), volume (Turnover vs. ADV), capital rotation ($BTC Miners, Spot ETFs, and Mag 7), and the preferred capital structure ($STRK, $STRF, $STRD, and $STRC). The ATM is simply noise inside that process (0.24% to 0.67% of variance, even when tested across 5d, 10d, and 20d rolling sums) and that is consistent over days, weeks, months, and now years. In the face of all of that, to treat a change in an internal KPI as the missing long horizon driver we would also have to use the same investment framing and metric that the company itself already excludes as valuation, and the market pricing process doesn’t reinforce. By the way, I should note that the money printing comparison is a common one that doesn't really fit or apply here. Money printing expands the supply of the unit of account against a relatively fixed base. The ATM is issued against available liquidity that is then used to acquire more of an underlying asset; which isn’t really an expansion of the monetary base. Yes, the ownership percentage of an existing share falls, but the economics of that share depend on what the company receives and does with the proceeds. But I think this is making the same assumption that the share count is the scarce unit and that the market capitalizes a per share claim, but there is no evidence of that in the observed data or price trends. Finally, yes, an artificial buyer or seller can move the price, that is obviously true for individual trades, and my work never denied that. What I tested was how much issuance (and also buybacks) explain price variance over various time frames once I account for $BTC and include other factors. What I found was that “other factors” overwhelmingly dominate. So claiming otherwise is arguing against a claim I didn't make.

Not at all impossible, though there are many hurdles to them transitioning to a Bank, even a non-lending one. 2030 sounds about right

Then go by the number Strategy reported on the Q3 call: $5.1B of USD assets against $13.1B of preferred, 3.4 years of dividend coverage. That is what a 12% coupon gets priced on, and it is why STRC sits at $100.