There are billions of people worldwide who can invest in the U.S. stock market. Most people don’t have a technical background. Of those who do, many don’t have investing experience. And of those who have both, most don’t have time to keep up with everything. The market for tech investing research is huge. The question is mostly about marketing and distribution. For example, Burry ranks second in Finance, with tens of thousands of paid subscribers. I’m going to bring my research subscription price to $45/month starting Oct 1st, it's closer, but still lower, to Burry’s and other research publications’ prices . I also don’t want a low price to make the research seem cheap. Existing members will always keep the price they subscribed at. You were the first people to believe in and trust my research, and you’ve become my friends along the way. I can’t thank you enough for it.

Always on agent adoption keep going, semi demand will continue to accelerate. AMZN is upgrading Seller Assistant with “workflows” that can continuously monitor and act on pricing, inventory, listings and account health, even when a seller isn’t logged in. Sellers can describe a task in plain English, set guardrails, and let the agent run in the background. It can flag rating drops, adjust pricing, refresh listings or prepare restock plans, with approval controls and full audit trails.

Both frontier labs cut prices on the same day. Anthropic launched Claude Opus 5.5 at $4 input and $20 output per million tokens, 40% cheaper to run than Opus 5 while matching Fable 5.1 on most tasks. Roughly an hour later OpenAI answered with GPT-6 Sol at $2 and $10 and Luna at $0.10 and $0.50, at least half off GPT-5.6 promo rates. The contest has moved to cost per completed task rather than benchmark crowns, and Anthropic itself says benchmark margins are a less reliable guide to real world differences. Cache reads tell the story: both landed at $0.20 per million, 60 to 90% below prior rates, since Anthropic says cache reads are most of the agentic coding bill.

Samsung just became the first to validate LPDDR6 on a Snapdragon. The 16GB chip hit 10.7Gbps on Qualcomm's Snapdragon 8 Elite Extreme Gen 6, with power efficiency up 21% over LPDDR5X. Samsung says sample supply and quality validation are done and it's ready to ship on customer schedules. Meanwhile China's CXMT announced LPDDR6 in mass production earlier this month at 12.8Gbps, faster than the 10.7Gbps baseline, and Xiaomi's 18 Fold with its own XRING O3 will be the first platform to use it. On the Galaxy side, Samsung may split memory tiers: testing Micron's 1 gamma LPDDR5X for the S27 and S27 Plus while keeping LPDDR6 for the Ultra and Pro

Alibaba is now playing the full stack game. At Apsara on Sept 22 the company confirmed Qwen 4 is in training and said Qwen 4.5 and Qwen 5 will scale to 5T to 10T parameters, up from 2.4T on today's Qwen3.8-Max. Existing Zhenwu chips already serve more than 650 customers, and Alibaba claims V900 clusters can scale to 500,000 cards. New cloud regions are coming to Türkiye, Finland and the Netherlands plus more capacity in Malaysia, Germany, the UAE, France and Hong Kong. Keep in mind every performance number is Alibaba's own claim, the 3x figure is measured against its own M890.

Yuanta Securities raises Samsung Electronics target 530,000 → 630,000 won on prolonged memory upcycle (Sept 23, 2026) Yuanta Securities researcher Baek Gil-hyun raises Samsung target price from 530,000 won to 630,000 won, Buy maintained; Q3 estimates: operating profit 99.86T won (+720.8% YoY, +11.6% QoQ), revenue 212.19T won (+146.6% YoY); Device Solutions division operating profit ~100T won in Q3; DRAM ASP +18% and NAND ASP +16% QoQ on full-scale HBM4 sales. What happened: Yuanta says tight memory supply and demand will persist into 2027 and 2028, as HBM production keeps eroding general-purpose DRAM capacity while AI inference spreads demand for high-capacity memory. Baek also flagged possible increased shareholder returns as the upcycle lengthens, arguing free cash flow could structurally expand and "concerns about peak-out adjustments present a buying opportunity." Why it matters: this is a sell-side name extending the memory upcycle call a full two years further out — if DRAM and NAND pricing stays tight through 2028, the whole memory chain's earnings power gets repriced higher.

Glass substrates are arriving sooner than planned. NVIDIA is pressing its substrate makers to speed up development, which could bring adoption to before 2028. BOE has 24 layer substrates over 100 by 100 mm already validated in house and being tested with real server CPUs and GPUs, with a mass production investment decision by 1H27. Samsung Electro Mechanics and LG Innotek are targeting 2027 to 2028. Glass replaces the conventional core inside the package with a much flatter and more stable material, so AI chips can be packed denser and run cooler. Packaging is turning into the next capex fight after CoWoS. -TrendForce

Hi all, When I started sharing my investment research, I charged $5 a month. Then readers asked to see my trades, so I added portfolio updates for $20 a month. I want people to be able to follow along without paying a huge fee. But then I see many writers charging $50, $100, even $200 a month without sharing their portfolios, and honestly, it makes me feel a bit stupid for charging so little while putting so many hard decisions and research articles out there for everyone to see. A few bad decisions and my reputation could be gone, while others don’t take that same risk. So I just feel silly to keep doing it at this price. I’ve decided to raise the price. The price will rise to $45 per month on October 1. The current price is available through September 30. By the way, existing members will keep the price they signed up at. It won’t change. Past results don’t guarantee future returns. I share my portfolio and decisions so you can judge my research by what I actually do with my own money, including the calls I get wrong. That work is shaped by decades of experience in investing and engineering. Thank you all for sticking around and following this journey with me.

Hi all, When I started sharing my investment research, I charged $5 a month. Then readers asked to see my trades, so I added portfolio updates for $20 a month. I want people to be able to follow along without paying a huge fee. But then I see many writers charging $50, $100, even $200 a month without sharing their portfolios, and honestly, it makes me feel a bit stupid for charging so little while putting so many hard decisions out there for everyone to see. A few bad decisions and my reputation could be gone, while others don’t take that same risk. So I just feel silly to keep doing it at this price. I’ve decided to raise the price. The price will rise to $45 per month on October 1. The current price is available through September 30. Past results don’t guarantee future returns. I share my portfolio and decisions so you can judge my research by what I actually do with my own money, including the calls I get wrong. That work is shaped by decades of experience in investing and engineering. Thank you all for sticking around and following this journey with me.

One of the most time-consuming tasks for me is calculating my portfolio performance across many different accounts, with all kinds of option premiums, cash, margin, etc. There may be an error here or there, but the numbers are quite close. I do it because, first, it’s fun. Second, I think it’s easier said than done. You can talk and post a lot, but what you’ve actually accomplished matters. Say you’re bullish on a stock but don’t own it, or it makes up only 1% of your portfolio. If the stock triples, that 1% position adds only 2% to your portfolio. But if you go all in, that’s a 200% gain. The opposite can happen as well. It can be a life-changing decision, which is why it’s much harder to do than to say. I started sharing my research on Patreon, and later Substack, mostly to document my investing journey. I priced it at $5 at first, then $20 when I added trade updates. I was doing it for fun and didn’t really know what it would become. My thinking was simple: if the research helped a $1 million portfolio beat the market by just 1%, that would be $10,000 in extra returns, far more than the subscription fee. The results have been far beyond what I imagined. Even excluding margin and option premiums, my stock picks and position sizing have delivered 5.68 times QQQ’s return since I launched the Patreon research. That’s why I keep sharing the work. The research is interesting, but putting it to the test with my own portfolio, and showing the results, is what makes it meaningful to me. Thank you all for joining me along the way. It’s been a fun ride so far, and I’m excited to see where we go next.

My portfolio just hit a new ATH, +127.7% YTD intraday today, so I thought I’d review current semi valuations and share some updates on what I’m holding. My performance since the end of 2022 is only an estimate, but it honestly looks ridiculously high at around +4,345%. A big part of that came from using margin during 2023 and 2024. For practical purposes, I think it makes more sense to share performance without margin and focus purely on the results from my stock picks. Pure stock picking and position sizing would be a fairer comparison with QQQ. Since launching my Patreon research, the performance has been +405% vs. +71.3% for QQQ, or roughly 5.68 times the return. Below are my current portfolio allocations, recent trades, and an important update on my cash position. After that, I’ll share my review of semi stock valuations and the impact of tech trends such as META MUSE adoption.

Jensen on ASIC: I worked for the original company LSI Logic who invented the idea of ASIC. As you know LSI Logic is not here anymore, right? When the market size is not very large, it's easy to have somebody uh be a contractor to help you put the packaging of all that stuff together and do the manufacturing on your behalf and they charge you 50 60 points of margin. But when the market gets large for an ASIC, there's a new way of doing things called COT, customer owned tooling.