JPMorgan just dropped an ABSOLUTELY INSANE forecast for AI compute. According to JPMorgan, global AI compute demand is expected to reach roughly 530 GW by 2031, while available supply is projected to reach just 250 GW. That leaves us with a staggering 280 GW shortfall, even after the industry spends hundreds of billions of dollars expanding capacity. We're talking about global AI compute supply potentially QUADRUPLING between 2025 and 2031, yet the shortage is expected to get even WORSE. In fact, JPMorgan estimates the gap between supply and demand could expand from roughly 50 GW today to nearly 280 GW by 2031. And here's why this matters so much. When demand consistently outpaces supply, the companies controlling scarce compute capacity gain enormous pricing power. This is especially important for neoclouds like $NBIS and $CRWV, which are building out GPU infrastructure to serve customers that simply cannot secure enough compute capacity through traditional hyperscalers. If JPMorgan's projections are even remotely accurate, GPU rental pricing could remain stronger for much longer than investors currently expect, supporting higher utilization rates, long term contracts, and potentially stronger returns on deployed infrastructure. But the opportunity extends FAR beyond neoclouds. Companies like $NVDA and $AMD benefit from the continued need for accelerators, memory manufacturers like $MU benefit from increasing demand for high-bandwidth memory, and networking, storage, power, and cooling providers all become increasingly important as the industry attempts to close this gap. And this is where I think the market is making a BIG mistake. Investors are so focused on whether AI companies are spending TOO MUCH money that they're overlooking the possibility that the industry STILL isn't spending enough to satisfy future demand. The real constraint isn't necessarily a lack of customers willing to pay for compute. It's the ability to manufacture enough chips, secure enough electricity, build data centers fast enough, and actually bring that capacity online. And that is EXACTLY why I continue to believe the biggest winners in AI won't necessarily be the companies building the models, but the companies supplying the infrastructure required to run them.

According to Korean media, Microsoft, Google, Amazon and Meta are speeding up their in house AI chips. That could open up a SECOND group of HBM buyers beyond $NVDA. Microsoft's Maia 300 is expected to ship more than 1 MILLION units next year, according to Reuters as cited in the report. Each chip is said to pack eight HBM4 stacks totaling 288 GB. Google's next TPU and Amazon's Trainium3 are also pushing HBM capacity per chip higher. Counterpoint expects HBM bit demand from custom chips to rise about 35x from 2024 to 2028. SK Hynix already supplies HBM3E for Maia 200. Samsung wants a bigger piece of this custom chip wave, including foundry overflow if $TSM stays full. Less dependence on Nvidia does not mean less memory but rather means MORE HBM sockets. $MSFT $GOOGL $AMZN $META

$AAPL and $META are teaming up on AI and I think this partnership is a MUCH bigger deal than people realize. For the first time since 2017, these two longtime rivals are publicly working together and it's all about bringing powerful AI models directly onto Apple's hardware. Meta is releasing Muse Spark Open, an open weight AI model that developers can run locally using just 2 to 4 connected Mac Studios. Now, why is this important? Traditionally, running powerful AI models required access to expensive cloud infrastructure packed with GPUs. Every time someone interacted with these models, much of the computing happened inside a data center. But with this partnership, developers can run a powerful AI model on their own machines without having to send every request to a remote server. This could mean lower ongoing inference costs, better control over sensitive data and less dependence on cloud providers for certain workloads. And I think this is where things get REALLY interesting. For Apple, this creates another reason for developers to invest in its high end Mac hardware. Apple's unified memory architecture allows its chips to access large amounts of memory, which is particularly useful when running large AI models locally. For Meta, it's about distribution and getting more developers to build with its models. Making powerful AI accessible outside expensive data centers could accelerate adoption, especially among developers building AI agents, coding tools, and other applications but there's also a much bigger trend happening across the AI industry. We're starting to see AI computing expand beyond centralized data centers and move closer to the actual user. And before anyone starts worrying about what this means for $NVDA and the broader AI infrastructure industry, I don't think this obviously lowers the need for data centers whatsoever. Training frontier models and serving millions of users simultaneously still require enormous computing resources. Local AI simply creates another way for people to access that intelligence. In fact, I think the long term opportunity is that cheaper and more accessible inference will encourage MORE people to build and use AI applications, ultimately increasing total computing demand. My biggest takeaway is that the AI industry is becoming much bigger than just the race to build the largest model. It's now about who can make these models cheaper, more accessible, and easier to deploy and Apple and Meta clearly want to be part of that future.

This morning I went discount shopping and initiated a position in $NBIS at $164 and bought more $SPCX at $141 I also sold my $AAPL at ATHs to fund these buys, so I still maintain my cash position I realize there are risks w/ higher rates/inflation & Anthropic financials But I continue to believe that the demand for compute is much much higher than the market realizes. I also believe that this compute remains scarce for years ahead, which means the pricing power of compute builders/owners remains strong Regardless of the markets random ARR estimates of Anthropic or OpenAI, the secular demand to turn energy into intelligence is not stopping. Whether this comes from frontier model providers, agent harnesses, open models or elsewhere, we will use more AI tomorrow than today And the more companies use AI, the more efficient and profitable they will become (we are seeing it in so many companies already). It will not be in a straight line but it will continue to be up and to the right... I know this for sure This is why I am continuing to build positions in those who are capable of turning energy into intelligence By the way, you can track all of these moves in real-time and see what else I hold in my portfolio inside Milk Road PRO here: Nebius is the one neocloud that has a strong balance sheet and could whether a big storm. They have $8B of cash and could even acquire some of the many neoclouds who will likely fail at discounted prices Even more so on this is SpaceX, with more than $100B in capital sitting on their balance sheet, they will imo be the biggest buyer of compute if things fall apart. Their numbers already look good at current build prices, if they were to get discount prices from distressed companies, that would be a bonus Higher rates makes financing compute a little more difficult, it doesn't make it impossible, not even close. which means compute businesses who manage their capital well can continue to flourish during these times. I think Nebius and SpaceX will be those two companies Maybe we go lower, maybe we don't. I dont know in the short-term, but in the long-term I think both of these companies are going much higher

OpenAI just lost the man in charge of its entire COMPUTE buildout. Sachin Katti, who led OpenAI's infrastructure and Stargate data center push, announced he is leaving to build something newm, his last day is November 15 and he will stay on as an adviser. His replacement is Uday Ruddarraju, OpenAI's CTO of compute and the former head of infrastructure engineering at xAI. The timing is what makes this BIG, OpenAI is targeting roughly $600 BILLION in compute spending through 2030. Katti's team signed a big share of those deals. That includes about $300 billion with $ORCL, 6 gigawatts of $AMD GPUs, 10 gigawatts of custom chips with $AVGO, 10 gigawatts of $NVDA systems and up to $22.4 billion with $CRWV. In his own words, they secured DOUBLE DIGIT gigawatts of new capacity in a single year. Signing deals was phase one. Now comes the hard part, which is EXECUTION. The first AMD and Broadcom gigawatts are scheduled to start coming online right now. He also isn't the first to go. Several of OpenAI's top data center leaders have left this year. None of this means AI demand is slowing. It means the companies building this out are more dependent than ever on OpenAI delivering on time. With chip stocks down 7 straight days and CoreWeave bonds yielding near 15%, the market is already nervous about how this buildout gets FUNDED. Ruddarraju now owns the most important construction project in the United States tech industry.

Oracle got crushed and it's no longer just a neocloud problem. $ORCL fell 8.8% on Wednesday to close at $130.55, after the hot CPI report pushed up bets on another Fed rate hike. The stock is now down about 58% from its high of $308.87 a year ago. Debt is the issue because S&P cut Oracle to BBB minus in July, one notch above junk. Yields on its 2032 bonds rose to 7.24%, and the cost of insuring Oracle's debt against default hit a RECORD high. Oracle is the only one of the big AI spenders without an investment grade rating from S&P. Alphabet, Amazon, Meta and Oracle have issued more than $228 billion in bonds so far this year, more than double all of 2025. BUT the bull case is the backlog, Oracle has about $638 billion in contracted business, and Mizuho just reiterated Outperform with a $320 price target after an update showing construction on its New Mexico data center is moving forward. Higher rates are hitting the AI names that are building on borrowed money hardest, and Oracle is the biggest of them.

What if the AI boom is real but the RETURNS never show up? That's the base case from Jefferies Chris Wood, and it's going viral again tonight as Asian chip stocks crash. His weekly GREED & fear note says the most likely long term outcome is "massive capital destruction in the US," with market share going to cheaper open source Chinese models. He has numbers to back it up because the top Chinese models processed 36.39 trillion tokens on OpenRouter in the week to July 19, up from 4.37 trillion in late April while the top United States models processed 7.39 trillion. Spending isn't slowing down either, Jefferies expects the four biggest United States hyperscalers to spend about $695 billion on capex in 2026 and $870 billion in 2027. Wood says the trigger for unwinding all of this would be credit markets refusing to keep funding it. The bull answer is that cheaper tokens mean MORE compute and demand for chips keeps rising whichever model wins. The real question is who captures the profit.

Elon Musk replied to my post about $SPCX competition hurting the valuations of Coreweave & Nebius I explained how SpaceX is moving so fast that they're taking scarce resources and capital that Neoclouds need to succeed My guess is for Elon to reach 10GW by EOY'27 its not just from BUILDING compute, but from BUYING compute from distressed neoclouds Maybe this is why he replied "yikes" to @chamath's post about Coreweave yesterday too. Elon is on a mission to acquire as much compute as possible, one way or another As I've said many times, access to compute is the biggest moat for any company right now and not only is SpaceX going after as much as possible on earth, they are moving towards building compute in Space too Another product that SpaceX has a clear moat in, both in the transportation of GPUs into space, but also in the ability to beam it down back to earth via Starlink I explained this in the post that Elon replied "interesting" to I don't know exactly what part of these posts were specifically interesting to Elon, but it's clear that Elon is building to become the largest owner of compute in the world His plan is to either rent that out to the many enterprises who need access to compute (remember, compute = intelligence and intelligence = labour) with the optionality to use it for Tesla's Robotaxis and Humanoid Robots or SpaceX's Grok as the demand is needed His plan is to acquire as much compute as possible and then have the optionality to rent it or use it for the products his companies build What Elon is building with SpaceX and Tesla is by far the most ambitious endeavour in the world and I think these two companies (soon to be combined imo) will become the largest company in the world, by a large amount as this all plays out Both $TSLA and $SPCX remain core assets in my Milk Road PRO portfolio. Yes, I know every analyst will tell you they are both way overvalued, but I think they are wrong. In a world of AI, almost no one has a true moat. However, Tesla and SpaceX both have multi-year leads across the industries they compete in (or are inventing), they have the best engineers in the world and a leader that's relentless and always multiple steps ahead of others I think both of these companies are a must own in any portfolio. If you'd like to see my entire portfolio in real-time, you can track it here:

$SPY fell 1.1% today to around $769 after dropping as low as $766.71, as a hot inflation report pushed rate hike odds higher and hit AI stocks. Here are the top stories you need to know. 1. September CPI came in at 3.7% from a year ago with gas prices up 9.1% in a single month, and the odds of a Fed rate hike this month jumped to about 50% from about 22%. 2. $CRWV fell about 15% after its 2032 bonds dropped below 80 cents on the dollar and their yield jumped to nearly 15%, and Elon Musk replied "yikes" to a chart of the move. 3. The AI debt scare spread to $ORCL and $NBIS, with Oracle down more than 7% as investors question how much of the buildout is being funded with borrowed money. 4. Chip stocks kept sliding, with $AMD down more than 5% and both $NVDA and $AVGO down nearly 4%. 5. $ASML raised its 2026 sales outlook to 44 to 45 billion euros and said AI demand is so strong that customers want to place orders for 2029 and even 2030. 6. Barclays downgraded $DELL, $HPE and $NTAP to equal weight, saying it sees no meaningful AI inflection yet, and Dell fell about 10%. 7. Anthropic is showing its third quarter financials to institutional investors for the first time today as it prepares for an IPO that could value it around $2 trillion. 8. $GFS will run two new chip fabs in New York backed by up to $21 billion from Japan, and the United States Commerce Department will take a roughly 4% stake in the company. 9. $GOOGL rolled out Gemini 4 Argon to all paid users, with API pricing starting at $2 per million input tokens and $10 per million output tokens. 10. $NVDA committed $250 million to support AI safety research in the United Kingdom.

CoreWeave's CEO just told his own employees the market is acting IRRATIONALLY. In an email seen by Bloomberg, Michael Intrator wrote that "when you are building a new type of technology company, there will be periods of time when the market misunderstands and acts irrationally." He said CoreWeave's contracted backlog "more than supports" its debt obligations, and that he's confident in its business model, customers and execution. He also pointed at rivals. He said the weakness and delays around several competitors' attempted IPOs and debt deals are "bleeding into the narrative around CoreWeave." The bond market isn't convinced yet. CoreWeave's unsecured bonds due 2032 traded below 80 cents on the dollar today, according to Bloomberg Law. That's a yield above 15%, and those are levels usually linked to DISTRESS. But you have to understand the DEMAND. CoreWeave ended June with about $104 billion in backlog, and that doesn't count more than $25 billion in new customer commitments signed early in the third quarter. Q2 revenue more than doubled from a year earlier to $2.58 billion. Intrator says the company is still largely sold out for years. The bear case is the cost of building it. Interest expense hit $640 million last quarter, up from $267 million a year earlier, and CoreWeave lost $626 million in the quarter. It spent $14.1 billion on property and equipment in the first half of 2026, compared with $3.9 billion in the same period last year. That's the whole debate in two numbers. More than $100 billion of contracted demand on one side, and a 15% cost of money on the other. I am still bullish on $CRWV here.

Is SpaceX competition hurting the valuation / credit risk of Neoclouds? $CRWV and $NBIS down 10%+ today, while $SPCX down just 2% (and UP on the month) The obvious trigger is credit risk, with CoreWeave's unsecured bonds yielding 15% But in the last week, Elon also confirmed that SpaceX will reach 2GW by November and potentially 2.6 GW by EOY. That's .6GW ahead of schedule and quickly becoming much bigger than Coreweave and Nebius Almost no one building compute is doing it ahead of schedule, but SpaceX is proving to be superior here on speed and is expecting close to 10GW by 2027 (And significantly more in years ahead, especially as we move to compute in Space) At the same time, Elon also mentioned earlier this week about the difficulty in executing this build out (said even SpaceX engineers are finding it hard) but also in the ability for companies to raise capital... the issue Coreweave is having now. But we know SpaceX and Elon has no issues in raising capital (and generating revenue from other sources too). Raising capital is what Elon excels at over anyone So I wonder if the market is starting to see that SpaceX is becoming a very big competitor to these neoclouds, taking demand sure... though I think there is plenty of demand to go around.. but more so taking resources and capital that other neoclouds need to scale SpaceX also has optionality to either rent out the compute or use it for xAI / Grok, which lowers risk even further to fund their compute buildout. It's also why SpaceX can charge a premium on their compute vs the other neoclouds, lowering risk even further I'm not saying SpaceX is the full reason for Coreweaves issues, but you gotta think this evolving competitive landscape of Elon entering the ring has to have something to do with it Even with this said, I'm still bullish on neoclouds as I think the demand remains endless, but I think SpaceX becomes the dominate neocloud by next year

Paul Tudor Jones says this market STILL feels like 1999 but the numbers tell a very different story. Jones described a two speed market that is perfectly bifurcated between everything adopting AI and everything that is not, with the two sides moving in opposite directions.He is right that the market is split, but the comparison to 1999 starts to fall apart once you look at valuations. At the March 2000 peak, the largest tech stocks traded at a market weighted P/E of about 126 times earnings but today, the Nasdaq 100 trades at roughly 22 times forward earnings, which is about in line with its 20 year average and nowhere near bubble territory. The bigger difference is that today's AI leaders are actually making ENORMOUS profits. $NVDA reported $96.2 billion of revenue in a single quarter, up 106% from a year ago, with 75% gross margins and $59.7 billion of net income. In 1999, a huge number of the companies driving the Nasdaq higher had no earnings at all, and the market was paying for promises instead of profits. The buildout is also being funded very differently, because $MSFT generated $182.9 billion of operating cash flow in fiscal 2026, more than enough to cover $115.9 billion of capital spending. And the demand behind that spending is already signed, with Microsoft, Oracle, Alphabet and Amazon now reporting nearly $2.4 TRILLION of combined backlog in contracted revenue that has not been recognized yet. The supply picture might be the clearest difference of all. In 2001, the United States was sitting on a massive fiber glut, and Merrill Lynch estimated that only 2.6% of the fiber capacity that had been laid was actually in use. Today, the problem is the exact opposite, because Micron says more than 75% of its 2027 output is already committed, Microsoft says its cloud growth is constrained by capacity through at least the end of 2026, and the New York Fed's John Williams said today that AI demand is pushing up chip prices with supply unlikely to catch up for years. A glut destroys pricing power, while a shortage creates it, and that is why memory and power companies are seeing prices rise instead of collapse. Good try to scare the market for the 7635363th time.

$SPY finished basically flat today, closing at $778.01 for a 0.01% loss after dipping as low as $775.95 earlier in the session as chip stocks sold off. Here are the top stories you need to know. 1. Memory stocks got hit again on worries about China's CXMT, with $MU falling as much as 5.8% and $AMD dropping more than 3% at one point. 2. South Korea's KOSPI fell 5.5% as Samsung dropped 7.4% and SK Hynix sank 10%, pushing the index's losses over the last five trading days past 15%. 3. $AVGO CEO Hock Tan compared many AI investors to casino gamblers, but he also said chip supply should stay tight for years and pointed to OpenAI's $50 billion revenue run rate as proof of demand. 4. The IMF raised its 2026 global growth forecast to 3.1% from 3.0%, crediting AI spending and calling it the largest investment boom in United States history. 5. Bain estimates AI infrastructure spending could reach $1.5 trillion a year by 2031, which means the industry would need about $6 trillion in annual revenue to justify it. 6. $GOOGL rolled out its new Gemini 4 Argon model globally across the Gemini app, Search and its developer platform. 7. $AMD agreed to provide credit support for the first 222 megawatt phase of Fermi's AI power campus in Texas, and $FRMI shares jumped more than 20%. 8. $CRWV bonds climbed to yields near 15% for the first time as investors grew more worried about its debt load. 9. $ASML reports earnings tomorrow, and analysts expect about €11.7 billion in revenue, up more than 50% from a year ago. 10. $NVDA CEO Jensen Huang kicked off his European tour in Paris and announced investments in four French AI startups.

The Fed just admitted that AI demand is now big enough to move INFLATION across the entire United States economy. New York Fed President John Williams said today that one more rate hike may be appropriate this year but he currently sees no need for any further increases in 2027. That is notably less hawkish than the market, which had been pricing a December hike plus two more hikes next year. Bonds liked what they heard, with the 10 year Treasury yield falling about 2.5 basis points to 5.219% and the 30 year dropping about 3.4 basis points to 5.565%, which is a small relief for $TLT holders. But the part that really caught my attention is what he said about AI. Williams said AI infrastructure demand is pushing up chip prices, construction costs, and inflation, and that supply is unlikely to catch up for SEVERAL YEARS. This is not a new view from him either, because back in July he said the economy is in a race between available supply and surging demand, with sharp price increases in semiconductors, power transformers, and other technology essential for the AI buildout. In September he went even further, saying that so far demand is winning, and that those higher costs are starting to show up in the prices consumers and businesses pay for other products. When a central banker is telling you that chips and power equipment are scarce enough to show up in national inflation data, that is the AI supply shortage thesis being confirmed at the highest level. It is exactly why memory names like $MU, power infrastructure names like $VRT and $GEV, and optical names like $COHR and $LITE have pricing power right now, because customers cannot get enough of what they sell. Now, there is another side to this, and it is worth taking seriously. Higher rates are tightening borrowing conditions, Williams himself noted that higher long term yields are doing some of that work, and a 10 year yield above 5% makes it more expensive to finance the massive data center buildout. New York Fed survey data also shows one year inflation expectations climbing to 3.9% in September. If AI driven price pressure proves sticky, the market could be right to price more hikes than he does, which would be a real headwind for $SPY and richly valued names like $NVDA. Williams still expects AI productivity gains to eventually help bring inflation down, and he says the broader economy is not overheating and labor markets remain stable. My take is that the Fed is now openly saying AI demand is outrunning supply, and as long as that stays true, the companies selling the scarce stuff should keep winning.

Google just released its MOST powerful AI model ever and it is going straight after OpenAI and Anthropic. Sundar Pichai announced today that Gemini 4 Argon is live globally for developers and enterprises in the paid tier of the Gemini API and the Gemini platform. Ultra subscribers can use it in Google Antigravity, and both Pro and Ultra subscribers can use it in the Gemini app and Search. Google says Argon is now the NUMBER ONE model on the Vals Index, which measures economic impact across finance, coding, legal, and tax work, with each sector weighted by its contribution to United States GDP. Argon scored 68.9% on that index, ahead of Claude Opus 5.5 at 67.0% and GPT-6 Astra at 63.1%. On DeepSWE, a test of long, real world software engineering tasks, Argon set a new record at 77.9%, compared to 74.2% for Opus 5.5 and 74.1% for GPT-6 Astra. The long context results are where the gap really opens up, with Argon scoring 84.2% on GraphWalks between 256K and 1M tokens while Opus 5.5 came in at 66.8%. Google also raised the output limit to 1 MILLION tokens, up from just 64K before, which means the model can keep reasoning and writing through a huge task in a single run instead of stopping halfway. It is not a clean sweep, though, because Argon still trails Claude Opus 5.5 and GPT-6 Astra on terminal heavy coding tests like Terminal Bench 4.0 and FrontierSWE. The pricing might be the most aggressive part of this launch, with an introductory rate of $2 per million input tokens and $10 per million output tokens, and cached input at 95% off. After the introductory period, the price rises to $4 and $20, the same as Anthropic's Opus. Google says more than 100,000 of its own employees battle tested Argon over the last month, using it for everything from daily debugging to large codebase migrations. For $GOOGL, this is a big deal because Google owns the chips, the cloud, the distribution, and now arguably the strongest all around model in the world. The frontier race is far from over, but Google just made it very clear that it is not falling behind anyone.

Apple just made its BIGGEST push yet to put AI in every room of your house. It starts with HomeView, Apple's first smart home display and it has a 7 inch screen, runs on the A18 chip, costs $299 and ships November 13. HomeView can recognize up to six people by their faces and show each one their own Messages, Calendar and Reminders and it can also stream up to nine security cameras at once and control locks, lights and thermostats. The new Apple TV 4K is getting the A19 chip, with up to 50% faster CPU and 2x faster GPU performance. It also adds far field microphones, so you can talk to Siri without the remote. It ships November 13 at $299, up from $199.HomePod mini 2 costs $149, up from $129, and ships October 30. Pair four of them with the new Apple TV and you get Apple's first four speaker surround sound setup. Apple also launched Apple Home Collection, a line of cameras, thermostats and a lock built with LG Electronics and Schneider Electric and it starts arriving in December. The real star is Siri AI. Apple calls it an entirely new version of Siri with personal context, broad world knowledge and the ability to take actions across your apps. Apple says it runs on Apple Foundation Models, which Google says are built on its Gemini models under a multiyear deal announced in January. This is Apple's AI strategy in action. Apple is not trying to win the model race but rather putting a Gemini based Siri into hardware people already trust. A screen in the kitchen that knows who is standing in front of it is exactly the kind of always on assistant that keeps people using AI every single day. And with more than 2 billion active Apple devices, every HomeView that needs an iPhone to set up makes the ecosystem even harder to leave. These products won't move Apple's revenue much on their own, but they show Apple is turning Siri from a punchline into a PLATFORM.

$AMAT just opened the largest joint chip research facility in the world, and all the major chipmakers are in it. The EPIC Center in Sunnyvale, California covers about 600,000 square feet, with nearly four acres of cleanroom space. Spending is expected to reach about $5 billion as customer projects get going. $TSM, $INTC, $MU, Samsung and SK hynix are all founding partners. $AVGO, Advantest, SCREEN, Kioxia and Besi are joining as partners, and Stanford and UC Berkeley will take part too. The idea is to put chipmakers, equipment makers and researchers in one building so new materials and processes can be developed side by side. TSMC's CEO says it will help the company get next generation equipment early and strengthen its advanced packaging research. To be fair, $5 billion is a big bill, and it will take years before this shows up in Applied's revenue but here's why it matters. First, Applied gets an early look at the problems its customers are trying to solve in logic, memory and packaging. Second, its customers can get to volume production faster and third, it makes Applied part of the roadmap, not just a seller of machines. As AI chips get harder to make, the companies building the tools become MORE important.

The chip sell off just spread across ALL of Asia. South Korea's KOSPI is down about 6% and has dropped below 6,000. Japan's Nikkei is down about 2.6%, and MSCI's Asia Pacific index outside Japan is down about 2%, according to Reuters. Not one of the six major Asian indices is green right now. SK hynix is down more than 8% and Samsung about 6.7%, both at their lowest levels since August, according to CNBC. It started on Wall Street, where the Philadelphia Semiconductor Index fell 2.66% after China's CXMT claimed progress on a denser DRAM design. Brent oil near $106 is adding to the pressure on inflation and interest rates. To be fair, AI chip stocks had a huge run this year, so some of this is just investors taking profits but here's what we are watching next. First, whether chip stocks find a floor in Asia today. Second, the direction of oil prices. Third, $TSM earnings on Thursday, which will be the first real check on AI demand. One bad day does not change the demand story, but it shows how CROWDED this trade has become.

Mark Zuckerberg just gave one of the best answers yet to the GPU depreciation bear case. He says owning its own data centers is a financial advantage for $META. Meta will probably keep running its GPUs for a while even after they are fully depreciated on the books. And he says even an $NVDA H100, a chip that is years old now, is worth MORE today than it was last year. The bears keep saying GPUs lose their value in a couple of years and that this will crush AI margins. But if old chips are still in demand and still doing useful work, every GPU these companies buy pays off for much longer than the market thinks. That is good news for $META, and it is good news for $NVDA.

$SPY finished nearly flat today, down about 0.06% and closing near $778 after dipping as low as about $777. Here are the top stories you need to know. 1. $MSFT announced Maia 300, calling it its biggest bet yet on custom silicon, and said Maia will become the primary inference system for its own products starting in 2027 with multi gigawatt deployments. $NVDA fell about 1.3% on the news while $MSFT rose. 2. Memory stocks sold off hard after China's CXMT said it made progress on a denser 4F2 DRAM design and plans DDR5 server memory by year end. $MU dropped about 4% and slipped back under $1,000 while $SNDK fell close to 7%. 3. Every name in the Philadelphia Semiconductor Index traded lower at one point, with the index down more than 2% as AI valuation worries kept pressure on chips. $AMD closed down about 2.3% and $AVGO fell about 1.9%. 4. $AMD unveiled its Helion rack at the Open Compute Project Summit, an open design for agentic AI workloads that can run CPU only or mix in up to 16 Instinct MI350P GPUs. Partners including Dell, HPE, Cisco, and Supermicro plan configurations available from November. 5. $NVDA and $CSCO introduced a new Ultra Dense Small Form Factor Pluggable optics module for AI data centers, targeting much higher bandwidth density with better power efficiency. $NVDA also opened more of its Vera Rubin infrastructure designs at the same summit. 6. Anthropic reportedly signed an unannounced multiyear cloud deal with $IREN, adding to roughly $530 billion in data center leases the lab is lining up. Shares jumped overnight on the report before giving some of those gains back. 7. $META released Muse Wonder and Muse Wonder Full, calling them its largest and most capable agent models yet, with Active Memory, a 1 million token context, and strong scores on Meta's personal agent benchmarks. Mark Zuckerberg also told the Financial Times that Meta has caught up with OpenAI and Anthropic at the frontier. 8. $NVDA is in early talks to deepen its stake in or acquire Reflection AI, the open weight coding startup it has already backed with about $800 million. Reflection's Beam model and a reported $25 billion valuation put the talks in the spotlight as $NVDA pushes further into AI software. 9. $TSM reports earnings on Thursday and is expected to post a sixth straight quarter of record profit of about $23.4 billion, up roughly 65% from a year ago. Morgan Stanley expects $TSM to raise its full year growth forecast into the low to mid 40% range. 10. Reports say $AVGO is arranging tens of billions in financing for OpenAI custom chips on top of the $42 billion facility already tied to Anthropic. That underscores how chipmakers are now helping fund the AI labs that buy their silicon.