Calvin

The $CBRS–$CRWD today announced a strategic partnership. This partnership may reveal an important new growth market for Cerebras. Cerebras has mainly been viewed as high-speed infrastructure for AI inference. But that same advantage—extremely low latency—could also be valuable in AI cybersecurity. CrowdStrike plans to use Cerebras to help power Falcon AIDR, which protects enterprise AI applications by monitoring prompts, AI agents, model interactions and sensitive data flows. The key advantage is not simply “stopping hackers faster.” It is allowing more advanced AI security models to inspect and respond to threats in real time without adding too much delay to normal AI workloads. This suggests Cerebras may not be limited to serving model providers. Its architecture could also be well suited for AI security, fraud detection, financial risk control and other applications where real-time inference matters. One partnership does not yet prove a major new revenue stream, and the contract size has not been disclosed. But CrowdStrike is a high-quality reference customer. If other security companies adopt Cerebras for similar workloads, this could become an important new vertical for $CBRS. The long-term opportunity may be broader than the market previously assumed.

Mooni insight

The $CBRS–$CRWD partnership may reveal an important new growth market for Cerebras. Cerebras has mainly been viewed as high-speed infrastructure for AI inference. But that same advantage—extremely low latency—could also be valuable in AI cybersecurity. CrowdStrike plans to use Cerebras to help power Falcon AIDR, which protects enterprise AI applications by monitoring prompts, AI agents, model interactions and sensitive data flows. The key advantage is not simply “stopping hackers faster.” It is allowing more advanced AI security models to inspect and respond to threats in real time without adding too much delay to normal AI workloads. This suggests Cerebras may not be limited to serving model providers. Its architecture could also be well suited for AI security, fraud detection, financial risk control and other applications where real-time inference matters. One partnership does not yet prove a major new revenue stream, and the contract size has not been disclosed. But CrowdStrike is a high-quality reference customer. If other security companies adopt Cerebras for similar workloads, this could become an important new vertical for $CBRS. The long-term opportunity may be broader than the market previously assumed. $CBRS $CRWD #AI #Cybersecurity #AIInfrastructure Disclaimer: This post is for informational purposes only and does not constitute financial advice or a recommendation to buy or sell any security. Please conduct your own research.

@damnang2 @Semicon_player I’ve read your posts, and honestly, it seems like you already know pretty much everything covered in this book.

Just picked up the bestselling book on semiconductor investing in Korea. Looking forward to reading it😍 @Semicon_player

AI & Memory Stock Outlook The next signal may come from Alphabet. On July 16, TSMC provided a more important fundamental signal than the short-term share-price reaction. TSMC raised its 2026 capital expenditure guidance from $52–56 billion to $60–64 billion. Management said capital expenditure over the next three years would be meaningfully higher than during the previous three years. TSMC is currently constructing or advancing 13 leading-edge process and advanced-packaging facilities in Taiwan. TSMC does not rely solely on forecasts from its direct semiconductor customers. Management explained that it also communicates with its customers’ customers, including cloud service providers and AI data-center operators. TSMC combines bottom-up customer forecasts with top-down end-market analysis. It reviews customers’ multiyear product roadmaps, the actual progress of AI data-center construction and deployment, and whether the chips it produces are genuinely being installed and used rather than accumulating in inventory. TSMC therefore does not blindly accept the forecast of any single customer. It cross-checks demand across multiple layers of the supply chain before committing capital that cannot easily be reversed. C.C. Wei also noted that a new process technology such as A14 now requires approximately five to seven years from technology development and capacity preparation to high-volume production. There is no shortcut. This means TSMC would not decide to build multiple fabs simply because orders looked strong for the next one or two quarters. The capacity decisions being made today are intended to serve demand several years into the future. For me, this is an important fundamental confirmation. AI-related demand has not disappeared simply because AI and semiconductor stocks have declined. After checking demand with both direct customers and downstream customers, TSMC is still willing to significantly increase capital expenditure and commit to long-duration leading-edge capacity projects. That suggests management believes AI demand is a multiyear trend extending toward the end of this decade, rather than a temporary boom lasting only one or two quarters. This does not mean every AI stock has already bottomed. It also does not mean memory prices will no longer experience cyclical corrections. But it does mean that the bearish assumption that major technology companies are about to broadly stop AI capital expenditure is not currently supported by TSMC’s actual investment decisions. The next important signal may come from Alphabet. Alphabet is scheduled to report Q2 2026 earnings after the U.S. market closes on July 22, or during the early morning of July 23 in UTC+8. I will be watching three things: 1. Whether Alphabet maintains or raises AI and data-center capital expenditure; 2. Whether Google Cloud demand remains strong; 3. And, most importantly, whether the market responds positively to good news. The final point matters most. A fundamental bottom is not confirmed simply because earnings beat expectations. In recent weeks, Broadcom, Samsung and TSMC all demonstrated that exceptionally strong results can still be followed by falling share prices. A more convincing bottom signal would require the market to begin buying good news again. If Alphabet reports strong cloud demand and sustained or higher AI capital expenditure, and the stock rises and holds those gains, it could signal a transition from “sell the news” to “buy the news.” That would not guarantee that the correction is over. But it would be one of the first signs that market psychology is beginning to improve. For now, my base case remains unchanged: AI infrastructure investment is still expanding; The memory cycle may last longer than previous cycles; But strong fundamentals alone are not enough. The next stage of confirmation must come from price action. ⚠️ Disclaimer: Not financial advice. The analysis above is for informational purposes only based on publicly available IRS guidelines. Please do your own research (DYOR) before making any investment decisions.

AI & Memory Stock Investment Review Dates are based on UTC+8. Over the past six weeks, I have been trying to distinguish between two very different things: 1. Whether the AI and memory fundamentals are still strong. 2. Whether the market is still willing to reward strong fundamentals. The answer, so far, appears to be: The fundamentals remain strong, but market confidence has become much more fragile. On June 4, Broadcom reported fiscal Q2 2026 results after the U.S. market closed on June 3. Those were still exceptionally strong numbers. However, Q2 revenue was slightly below Wall Street expectations, the Q3 AI semiconductor guidance failed to meet the market’s extremely aggressive unofficial expectations, and management merely maintained its target of more than $100 billion in fiscal 2027 AI semiconductor revenue instead of raising it again. Broadcom subsequently fell 12.6%. That was the first important signal to me. The company was not forecasting a slowdown. It simply failed to beat already extreme expectations by a wide enough margin, yet the stock suffered a double-digit decline. I concluded that market confidence in AI stocks had become fragile, so I reduced part of my AI-related exposure. In retrospect, that decision was correct from a risk-management perspective. Later in June, after memory stocks had already declined for a period, I bought Micron again ahead of its fiscal Q3 earnings. I believed Micron had a high probability of beating expectations. On June 25, Micron reported: Revenue of $41.456 billion, up approximately 74% sequentially and 346% year over year; GAAP net income of $28.243 billion; Non-GAAP EPS of $25.11; Next-quarter revenue guidance of $50 billion, plus or minus $1 billion; And non-GAAP EPS guidance of $31, plus or minus $1. More importantly, Micron disclosed that it had signed 16 Strategic Customer Agreements, or SCAs. These were not ordinary long-term supply agreements. The contracts generally extend from 2026 through the end of 2030, while automotive agreements generally run for three years. They include binding minimum-volume commitments and take-or-pay provisions, although the pricing mechanisms differ between agreements. Fourteen of the agreements represent approximately $100 billion of cumulative minimum revenue over their remaining terms, based on minimum committed volumes and minimum pricing. Micron also expects approximately $22 billion of cash deposits and related financial commitments, including about $18 billion of cash deposits. This led me to believe that the memory industry may be undergoing a structural change. The cycle will not disappear, but the risks of sudden order cancellations, abrupt price collapses and undisciplined capacity expansion may be lower than in previous memory cycles. Based partly on Micron’s results and the continued tightness in memory supply and demand, I bought SK Hynix ordinary shares in Korea ahead of its U.S. ADR listing. On July 7, Samsung announced extraordinary preliminary Q2 2026 results: Operating profit equal to roughly 19.1 times the level recorded a year earlier; And quarterly operating profit exceeding Samsung’s combined full-year operating profit from 2023 through 2025. At the exchange rate at the time, Samsung’s quarterly operating profit was approximately $58.4 billion, higher than Nvidia’s most recently reported quarterly operating profit of roughly $53.5 billion. Yet Samsung fell as much as 10.1% intraday and closed down 6.9%. SK Hynix also fell around 6%. I did not sell. This was where my risk standards became inconsistent. When Broadcom delivered strong results but was punished for failing to exceed extreme expectations, I interpreted the price action as a warning and reduced exposure. When Samsung delivered even more extraordinary numbers but still suffered a major sell-off, I did not respond in the same way. In retrospect, that was a mistake. SK Hynix’s ADR was priced at $149 and began trading on Nasdaq on July 10. Each ten ADRs represent one Korean ordinary share. The ADR opened at $170 and closed at approximately $168 on its first day, up around 12.8% from the offering price. I assumed that such a strong U.S. debut would help SK Hynix’s Korean shares stabilize or rebound during the next Korean trading session. That assumption was wrong. On July 13, SK Hynix fell 15.4% in Korea, marking its largest one-day decline in nearly two decades. Samsung also fell sharply, while the KOSPI dropped roughly 9% and triggered a temporary trading halt. The ADR subsequently rebounded by around 27%, and its premium over the Seoul-listed ordinary shares reportedly widened to approximately 51%. This showed that strong U.S. ADR demand did not automatically imply that Korean ordinary shares would rise. The two markets were being driven by different liquidity conditions, investor bases, supply constraints and arbitrage limitations. My broader view has not changed: I still believe this memory cycle may last longer than previous cycles. However, my review also shows that being correct about an industry does not automatically mean being correct about timing, position sizing or short-term price direction. The main lesson is that fundamentals and price action must both be respected. When the market repeatedly sells off on exceptionally strong news, that is itself an important risk signal. But then came what may be the most important fundamental signal so far: TSMC’s Q2 report. ⚠️ Disclaimer: Not financial advice. The analysis above is for informational purposes only based on publicly available IRS guidelines. Please do your own research (DYOR) before making any investment decisions.

@ChrisCamillo I actually just did a deep dive on this from an IRS tax law perspective. The short reports are missing a massive mathematical loophole regarding the MACR threshold. Detailed breakdown here: 👇 x.com/choilokchao/st…

Is it theoretically possible that $BE is lying about scandium and that Brookfield, Oracle and Nebius all failed to uncover a fatal supply constraint before committing to multibillion dollar projects? Yes. It’s also extraordinarily unlikely. Bloom has now stated in an SEC filing that it has sufficient non-China scandium supply for its current demand and backlog, with visibility to support 25 GW of annual production. Brookfield just expanded its partnership to $25B. Nebius committed up to $2.6B. Oracle contracted an initial 1.2 GW under an agreement covering up to 2.8 GW. Outside scandium estimates are not precision exercises. Bloom’s material intensity, manufacturing yields, inventory, supplier contracts and future efficiency gains are not publicly known. Read the short report. Read Bloom’s response. Use AI. Do your own work. Make your own decision. I’m not selling.

⚠️ Disclaimer: Not financial advice. The analysis above is for informational purposes only based on publicly available IRS guidelines. Please do your own research (DYOR) before making any investment decisions.

Conclusion: Reaching the >40% PFE cost threshold using just Scandium is mathematically absurd in a commercial SOFC system. Short reports are intentionally confusing "supply chain origin" with "tax law violation." $BE’s credits are completely fine. Know what you hold.

Short sellers are pushing a panic narrative that Bloom Energy ($BE) will lose its US tax credits due to the use of Chinese Scandium. But they completely missed the IRS math. Even if $BE uses 100% Chinese Scandium, their subsidies are mathematically safe. Here is why 🧵👇

@GongYouchai 一支穿云箭，就把韭菜骗

#Bitcoin 一支穿云箭，后一句是什么来着？

@PixelMoai @alexlam219 @c34750323 @YanPang14

#NFTGiveaways WIN 2 PixelMoaiClub NFT To enter 1. Follow @PixelMoai 2. Retweet, Like + Tag 3 friends 3. Join discord (Show Proof) : discord.gg/xT53VjH6Td Winners will be picked in 72h! Winner will be announced in discord! Good luck! #NFTCommunity #Giveaways #NFT

@GongYouchai 現在還是看空嗎?

从3000usd到40000usd，做了5天…我应该算是言行合一的博主了