Yabadabadoo

My take: one high-volume down day tells you about emotion and sell algos, not direction and it’s resolved both ways historically. The more useful question is the catalyst. Friday sold off on hike fears from a jobs number that was mostly a World Cup one-off. The scare was noise; the selling was mechanical. That’s the real “will this time be different.”

Leopold Aschenbrenner's Situational Awareness LP disclosed beneficial ownership of 12.41M Class A shares of Nebius $NBIS, representing a 5.6% stake.

So I was deep diving into $NASA ETF holdings to see what will heavily benefit from $SPCX IPO... And I came across $CPSH. At the start of this year, it was a solid US AlSiC play, but quickly became a multi-thesis bet. As of last week, $NASA ETF holds ~1.8M shares of $CPSH (I immediately started a relatively large position). This is important because $NASA is the only pure-play space ETF with direct SpaceX exposure pre-IPO. And a portion of every dollar flowing into $NASA for SpaceX exposure becomes a purchase of $CPSH. As SpaceX IPO momentum builds toward June 12, that flow accelerates. But let's go over why $CPSH even made the cut into $NASA's portfolio. At the beginning of 2026: > It was the only meaningful US/Western hedge against East Asian AlSiC supply (Denka, Sumitomo, JFC, etc.) > Existing customers: US Military, NASA, $LMT, $RTX, Northrop, General Dynamics, and more > AI optionality as $NVDA Rubin generation scales towards multi-thousand watt requirements (this angle alone deserves a whole new post) And the foundation has only gotten stronger... > Record 2025 annual revenue of $32.6M and Q4 2025 revenue $8.2M vs $5.9M prior year (+39% YoY). Q4 gross margin recovered to 14.6% from a Q4 2024 gross loss > New $4M hermetic packaging order announced post-Q1 > Navy SBIR office extended Phase I program for Amphibious Combat Vehicles > Potential US Navy destroyer ballistic shield contracts with Congressional funding already secured Now add the $NASA ETF layer... Large asymmetry here.

Thermal is a likely issue for 2027-2028 GPUs: As thermal targets escalate toward the 2000-3000W+ range, copper and aluminum becomes insufficient. But, AlSiC, a metal matrix composite, may become important. Here's why: Rubin VR200 GPU pointed toward 1800W TDP, AMD Instinct MI450X, reportedly reaches up to 2500W, and $NVDA Rubin Ultra's power reportedtly goes to 2300W. The move toward 2300W creates a heat flux problem of massive scale. Aluminum Silicon Carbide (AlSiC) is a metal matrix composite, used as for defense, space, and high-power industrial applications (eg. high-speed rail). Similar to how random toilet companies like Toto became critical for HBM, this material composite used for thermal management for aerospace and industrial might be used for AI as it can survive tens of thousands of thermal cycles without delamination. So, as AI accelerators like Rubin reach power levels comparable to industrial power modules or space, the adoption of AlSiC in semiconductors may become important. For AlSiC in 2300W systems there's: - (Junction-to-Case): Includes the silicon die, the underfill, and the first layer of TIM - (Case-to-Sink): Includes the heat spreader (lid) and the second layer of TIM (TIM2) - (Sink-to-Ambient): The thermal resistance of the cold plate or liquid heat exchanger. Case-to-Sink is likely the application for AlSiC. And as power densities will likely continue to climb toward the 3000W mark there becomes a mechanical and thermal crisis that likely causes a material change from traditional copper packaging. So there's four different parts to this: 1. Internal/External Thermal It does look like SiC interposers (internal) are probably used for $NVDA Rubin gen GPUs. Then AlSiC (external) may be used for the Microchannel Lid (MLCP) or heat spreader that sits on top of the die interposer assembly. 2. 3D vertical stacking (SoIC) These complex packages are highly vulnerable to warping during thermal cycling. A SiC interposer is a brittle crystal and it cannot provide structural rigidity. AlSiC acts as a "stiffener" that prevents the substrate from bowing under the high clamping pressures 3. Rubin Ultra NVL576 rack likely reaches high KW of power density This density creates a weight-loading bottleneck that SiC interposers cannot solve. Rubin Ultra NVL576 rack cumulative weight of the thermal stack plus liquid manifolds can exceed the floor limits (and AlSiC may become necessary to reduce the "dead weight" of the thermal management by 60%+ without compromising heat transfer). 4. Net-Shape Manufacturing of Microchannels Traditional copper lids must be etched or CNC-machined, a process that has reportedly encountered high mass-production difficulty for Rubin volumes. AlSiC is manufactured using "Quickset Injection Molding" to create a ceramic preform that is then infiltrated with aluminum. This allows for the creation of complex internal geometries. AlSiC Microchannel Lids and Silicon Integrated IHS looks like the alternatives for copper for thermal management. We might be seeing this addressed earlier in 2026 as SemiAnalysis in 2025 reported that Nvidia’s Blackwell (B100/B200) faced yield issues specifically due to warpage in the CoWoS-L packaging. TLDR thermal is a likely a bottleneck in 2027-2028: Some beneficiaries are potentially SiC interposers (high-purity SiC powder) and AlSiC composite for thermal management in 2027-2028. This is all ongoing research, but maybe we'll see some extremely niche and random small railway or space AlSiC supplier be used up for AI in 1-2 years time like toilet makers for HBM.

I bought $RDDT on Friday. I mean, they will surely get disrupted by Meta, so that seems like a solid way to lose some money.