Autonomous Investor

NVDA is up over 12x in the last 5 years. Sounds easy if you can just 'buy and forget', right? Actually, living through the journey did come with some mild turbulence along the way. 💎🦾 📈 Sep-21 to Nov-21: 63 weeks, 11.67 to 34.58, +196%, Covid helicopter money, gaming and crypto boom 💰 📉 Nov-21 to Oct-22: 46 weeks, 34.58 to 10.80, -69%, transitory is dead, the fastest rate hike cycle ever 💀 📈 Oct-22 to Mar-24: 76 weeks, 10.80 to 96.73, +796%, ChatGPT fired the AI super cycle starting gun, NVDA 2023 = CSCO 2000 chart from FT 🚀 📉 Mar-24 to Apr-24: 3 weeks, 96.73 to 75.58, -22%, Growth scare 🩸 📈 Apr-24 to Jun-24: 9 weeks, 75.58 to 140.72, +86%, Druckenmiller sold all NVDA between 80 and 95 🥶 📉 Jun-24 to Aug-24: 7 weeks, 140.72 to 90.56, -36%, Salm rule triggered, Yen carry trade unwound 💴 📈 Aug-24: 3 weeks, 90.56 to 131.22, +45%, Jackson Hole 🕊️ 📉 Aug-24 to Sep-24: 2 weeks, 131.22 to 100.92, -23%, Worries about capex overspend 🏭 📈 Sep-24 to Jan-25: 18 weeks, 100.92 to 153.11, +52%, Trump rally 📉 Jan-25 to Apr25: 13 weeks, 153.11 to 86.61, -43%, DeepSeek, H20 ban, Michael Burry bet half of his portfolio on NVDA puts, Liberation Day, Howard Marks nobody knows (yet again) 🤮 📈 Apr-25 to Aug-25: 18 weeks, 86.61 to 184.48, +113%, Mega sovereign AI deals, China trade deal, Jensen single handedly reopened the China market 📉 Aug-25: 3 weeks, 184.48 to 174.18, -10%, AI bubble headlines, data center revenue no longer grows at triple digits and slows down to a shocking 56%, Alibaba's rival chip 😱 The lessons? 📑 Know what you own, and why you own it. 🎢 Great return comes with great volatility. 🐻 Bear markets are painful, but temporary. ⏳ The stock market is a device to transfer money from the impatient to the patient. 🎯 Everyone makes money in a bull market, but spotting and holding onto the power law winner makes a huge difference to the returns. ⌚ You either sell too early or too late. 🙅‍♂️ The biggest risk is not taking any risk. Now bring on the weak seasonality, rate cuts, and more drama! 🪖✂️🍿🤞 #ai #tech #geopolitics #markets #investing #trading #stocks #compounding $nvda

to be clear, NVIDIA is NOT a car

xAI engineer Sulaiman Ghori got fired by Musk after a podcast talk. ->The xAI engineer revealed core secrets of the MacroHard project on the podcast, including internally packaging AI as "colleagues" for testing, with some people going to workstations to find their colleagues only to find empty desks; ->xAI is betting on a small-model approach to pursue extreme speed, with MacroHard already reaching 8 times the speed of humans, and considering renting the idle computing power of approximately 4 million Tesla vehicles in North America for deployment; ->The engineer also exposed xAI's flat organizational culture and extreme execution capabilities, with Colossus 1 being built in 122 days using a "temporary land lease," after which he was fired by Musk.

J.P.Morgan Global Memory Market.. REALLY GOOD TO READ.. $MU $SNDK SK hynix Samsung

The most bullish AI data point in a long time

Nvidia is so far ahead that even its 2 year old chips still outsell the rest of the industry combined.

Thoughts on NVIDIA Rubin and AMD MI455X Let’s dive into a comparison between Nvidia’s Rubin and AMD’s MI455X, both unveiled today. Starting with Rubin, it utilizes an 8-stack HBM4 configuration. It boasts a memory bandwidth of 22TB/s, leveraging memory with a per-pin Fmax of around 10.7Gbps. On the flip side, the MI455X opts for a 12-stack HBM4 setup. However, it delivers a bandwidth of 19.6TB/s, using memory with a per-pin Fmax of roughly 6.4Gbps. Considering the current JEDEC standard for HBM4 is 8Gbps, the difference is stark: Rubin is utilizing top-tier, high-spec HBM4, while the MI455X appears to be relying on HBM4 that falls below the standard spec. This highlights a distinct divergence in corporate strategy: Using top-tier components vs. Brute-forcing capacity. AMD likely adopted this approach because securing top-speed HBM4 volume is challenging for them. However, this strategy carries two significant risks. First, the cost and yield implications. Mounting more HBM stacks requires a larger interposer area, which directly drives up unit costs. Furthermore, a larger footprint inevitably lowers the yield for 2.5D packaging assembly. In other words, the strategy of using more units of lower-spec HBM4 could paradoxically end up being more costly than Nvidia’s strategy of using fewer units of high-spec HBM4. Second, the impact during memory shortages. This approach exacerbates supply chain bottlenecks. A 12-stack configuration consumes 50% more HBM chiplets/stacks per GPU compared to an 8-stack design. The tighter the global HBM4 supply, the more AMD’s shipment volume becomes capped by memory availability. Of course, in the early stages where yields for high-spec HBM4 are low, this isn't a major issue—low yields for top-bin parts naturally result in an abundance of lower-binned supply. But what happens as the yield learning curve improves? As yields for high-spec HBM4 rise, suppliers will have more incentive to allocate wafers to the higher-margin chips destined for Nvidia. This makes it increasingly difficult for AMD to source large volumes of low-performance HBM4 at low prices. Furthermore, with Samsung performing well in the HBM4 space, AMD won't be able to pick up inventory at "clearance" prices like they did during the HBM3E cycle. Ultimately, AMD is facing an inherently more disadvantageous cost structure at the chip level compared to Nvidia's Rubin.

Morgan Stanley Picks for 2026

‘The Thinking Game’ documentary has just passed 200M views on YouTube in just 4 weeks! 🤯Perfect holiday viewing if you’re interested in a behind-the-scenes look at how an AGI lab works, or what goes into making a Nobel Prize winning project like AlphaFold happen.🧬🚀

Nvidia paid 3X Groq's September valuation to acquire it. This is strategically nuclear. Every AI lab was GPU dependent, creating massive concentration risk. Google broke free with TPUs for internal use, proving the "Nvidia or nothing" narrative was false. This didn't just demonstrate technical feasibility, it revealed that Nvidia's moat was shallower than markets believed. When a hyperscaler successfully builds custom silicon, every sophisticated buyer starts running" should we build our own?" calculations. This drops Nvidia’s TAM. Jonathan Ross (Groq’s founder) is the inventor of TPU. He understood the architectural principles that made non-GPU AI acceleration viable. His LPU architecture targeted inference workload where GPUs are actually over-engineered. This matters because inference is where the real money is long-term. Training is one-time capex, but inference is recurring opex that scales with usage. If Groq proved LPUs could hit competitive price-performance on inference, every cloud provider would white-label their architecture. Nvidia would get squeezed into "just training" while losing the annuity stream. It is safe to see this deal as Nvidia insuring against Groq enabling an entire ecosystem of Nvidia alternatives. But what is more interesting is the second-order effect, the customer lock-ins. Now, Nvidia owns both the incumbent standard (CUDA + GPU) and the most credible alternative architecture (LPUs). This is MSFT buying Github-level strategic. Any AI lab evaluating "build vs buy vs alternative vendor" now faces: - Option A (Nvidia GPUs) - Option B (Nvidia <> Groq LPUs) - Option C (start from scratch) Turning a competitive threat into a customer segmentation tool, Jensen is the master of trades. They can now price-discriminate: premium customers pay for GPUs, price-sensitive inference gets funneled to LPUs, and Nvidia captures both. If Nvidia doesn't integrate LPUs in its roadmap, this was a pure defensive play. If they do integrate it and start offering "GPU for training, LPU for inference" bundles, this becomes a textbook moat-widening acquisition. The most expensive thing in technology isn't building the future, it's preventing someone else from building a future without you.

$GOOGL is gaining ground on OpenAI. How will this look like at the end of 2026?

Bernstein after meeting with $NVDA IR: Nvidia is still waiting on formal US licenses to ship H200s to China and hasn’t received codified terms on the 25% govt. revenue share. Mgmt says it is about 2Yrs ahead of $GOOGL's TPU program and sees GPUs as the better fit for cloud AI.

$nvda dominance continues 🤣

@amitisinvesting If you guys are together all the time…manager is probably not the right title 😂😂😂

live with my new manager with announcements for 2026 letsgoooooo

Bofa 2026 views

@USA_Polling What is your point in comparing $nvda to a bunch of pharma cos?

This is totally not a bubble ready to burst, not at all

Gemini has always had exceptionally strong multimodal capabilities. Gemini 3 Pro is an incredible vision AI model and is SOTA across all main vision & multimodal benchmarks. It’s great for document, screen, image, video & spatial understanding tasks - try now in the @GeminiApp!

The real frenzy starts once the most hyped companies go public. Space: SpaceX Defense: Anduril Al: OpenAl, Anthropic, xAl Robotics: Figure Al, Apptronik Fintech: Stripe, Revolut Crypto: Kraken, Ripple Chips: Groq, Cerebras Cloud: Databricks We haven't seen anything yet...

Dec 2010. Dec 2025.

How Netflix's acquisition of Warner Bros changes the media company pecking order More charts of the week: a16z.news/p/charts-of-th…

Good deep dive on Asia tour and customer conversations from Morgan Stanley. "customers biggest anxiety for the next 12 months is their ability to procure enough NVIDIA product generally, and Vera Rubin specifically"