Newsletter · 2026-07-25 · Ashutosh Agarwal

Intel's Comeback Quarter as Memory Stocks Fall and DRAM Keeps Climbing - Semiconductor Podcast Briefing - Week of July 25, 2026

Semiconductor Podcast Briefing for the week of July 18 to 25, 2026. Earnings dominated the tape: Intel posted its best quarter in about 15 years, TSMC and ASML raised capacity and capex yet their stocks fell, DRAM prices kept climbing while memory stocks dropped, and the AI scarcity thesis took its first real hits.

Semiconductor Podcast Briefing

Week of July 25, 2026: Intel's Comeback Quarter as Memory Stocks Fall and DRAM Keeps Climbing

This week the podcasts were dominated by earnings. TSMC, ASML, Intel, Texas Instruments, STMicro, Samsung, SK Hynix and Alphabet all reported, and the reactions were strange: several companies posted their best numbers in years and their stocks fell anyway. Underneath that, a real debate broke out: is the AI chip boom still about scarcity, or is the story quietly changing? Here is what the people on the podcasts actually said, with the numbers and the direct quotes.

TL;DR: Things That Mattered This Week

• Intel had its best quarter in about 15 years and it is suddenly the "comeback kid." Data-center revenue grew 59% to $6.3 billion, the foundry (contract manufacturing) business grew 31% to $5.8 billion, Apple was named as a new foundry customer, and management signed 10+ multi-year customer contracts. The stock is up nearly 300% since the US government took a stake last August. The Rundown
• TSMC and ASML both raised the roof, and both stocks fell. TSMC lifted its 2026 capital-spending budget by $8 billion to $62 billion (a 15% increase) and guided full-year revenue growth above 40%. Morningstar raised its TSMC fair value 27% to $534 per ADR and its ASML fair value to $2,050. The Morning Filter
• Memory is the week's paradox: prices up, stocks down. The average price of DRAM (the working memory in every server) rose about 20% in July, yet memory stocks fell around 20–24%. Micron trades at roughly 7x forward earnings on 350% revenue growth and 85% gross margins. Limitless
• Alphabet raised 2026 capex to $195–205 billion (from $180–190 billion) and the stock dropped about 6% to 7%, even though Google Cloud grew 82%. Higher capex is now read as good for chipmakers and bad for the spender. Squawk on the Street
• The "AI scarcity" story took its first real hits. SpaceX and Meta are renting out spare computing power, and the rental contracts are reportedly only three months long, which undercuts the idea that chips are permanently scarce. RiskReversal Pod
• Morgan Stanley now sees US AI-related capex at $1.2–1.3 trillion in 2027, up to $1.4 trillion by 2028, about 3.5% of GDP. Asian semiconductor exports to the US are growing 90% year over year. Thoughts on the Market
• China keeps doing more with less. A new Chinese model, Kimi K3, jumped to the top of the model rankings, is reportedly about three times cheaper to run, and had to freeze new sign-ups within 48 hours because it ran out of computing power. Motley Fool Hidden Gems Investing

1. AI Chip Demand and Hyperscaler Spending (NVDA, AMD, AVGO, MRVL)

The clearest single takeaway this week: the giant cloud companies are spending more, not less, and the market is now nervous about it.

On Alphabet's earnings call (quoted on Squawk on the Street, July 23), the CFO laid out the new number directly:

"We are updating our full-year 2026 CapEx guidance range to $195 to $205 billion, up from our previous estimate of $180 to $190 billion. The increase in the range is primarily due to an acceleration in the delivery of capacity to meet growing demand. As we previously shared, we continue to expect our CapEx to increase significantly in 2027."

Jim Cramer's reaction captured the tension: the cloud business was genuinely great (Google Cloud grew 82%, up from 63% the prior quarter, with backlog up 11%), but the spending scared people. "Don't say you have a great balance sheet... our balance sheet's not as good as it used to be." His bottom line on the stock: "I don't want to be in a stock that owes that much money." Squawk on the Street

Why does that matter for chips? Because, as Bloomberg Intelligence analyst Mandeep Singh put it on the Intel reaction shows, more cloud capex flows straight to the chipmakers: "The Google report with them raising CapEx, I'm expecting every hyperscaler will raise CapEx. Well, guess what? All the semiconductor names will catch a bid." Bloomberg Businessweek

Veteran tech analyst Dan Ives, on Bloomberg Intelligence (July 24), framed the whole thing as an arms race no one can afford to sit out. His memorable line: the hyperscalers are "building out the Vegas Strip in 1955... you get one spot on the Strip. If you pull back, guess what? You're two miles away. You're in Reno." He also predicted "a tidal wave" of equity raises from these companies "over the next 6, 12, 18 months," because "if they don't do it, despite whatever the stocks do, they know they're out of the game." He noted the phase has shifted "from CapEx to monetization." Bloomberg Intelligence

Nvidia's own CEO Jensen Huang made limited media rounds and told Axios (quoted on Squawk, July 23) that every industry is about to get more capital-intensive: "The new IT industry, the new software industry, and in the future, every single industry will be CapEx heavier. Not CapEx heavy, but CapEx heavier... But what you get as a result is incredible intelligence, incredible productivity, incredible growth." Nvidia's stock rose about 6 points that day on no fresh news beyond the interviews. Squawk on the Street

The bull case on Nvidia, argued by Rachel on Motley Fool Hidden Gems Investing (July 20): the stock trades at just about 22x forward earnings, has a huge backlog of "guaranteed revenue," just started shipping H200 chips to China, and remains "the toll booth for the entire AI industry." The bear pushback, from her colleague Matt: 22x only looks cheap "if the growth is going to continue," Nvidia has heavy customer concentration among a handful of hyperscalers "starting to make in-house chips... with the stated goal of reducing dependence on Nvidia," and the China approval is a reversible policy decision. Motley Fool Hidden Gems Investing

Cramer, for his part, called Nvidia "incredibly cheap... I could argue that it's below 18 times earnings." Squawk on the Street

The big-picture demand read came from Morgan Stanley's quarterly economics roundtable on Thoughts on the Market (July 21). Chief US Economist Michael Gapen: "We were thinking a little over a trillion for 2027. Now we're more like 1.2, 1.3 trillion, maybe as high as 1.4 trillion in 2028." He added an important caveat: because roughly 60% of that spending goes to imported computers and equipment, AI capex is only adding about 40 basis points (0.4 percentage points) to US growth this year. Chief Asia Economist Chetan Ahya noted the flip side: "they are seeing semiconductor exports growing by 90%" from Asia to the US, benefiting Korea, Taiwan and Japan most. He also stressed the cycle is broader than AI. For Asia, AI/semiconductor capex is about $380 billion in 2026, but energy capex is $900 billion, and he said he has "visibility for this cycle to be lasting for three, four more years." Thoughts on the Market

A quieter but important demand signal came from Cerebras CEO Andrew Feldman on The MAD Podcast (July 23): there is now a CPU shortage too, driven by "agentic" AI (AI that takes actions, not just answers). "As AI gets better and better at doing things... we're using more and more CPUs... the demand for CPUs is through the roof. And CPUs experience the same memory shortage." The MAD Podcast with Matt Turck

2. Memory: Prices Up, Stocks Down (Micron, SK Hynix, Samsung)

This was the most talked-about disconnect of the week. On Limitless (July 22), the hosts laid out the numbers plainly: memory prices "have skyrocketed to the tune... on average three to 500% over the last nine months," and just in July, "the average price of DRAM is up almost 20% on the month... and the stocks are down 20%, even though the demand has not changed one bit."

They walked through Micron's profile ("7x forward earnings on 350% revenue and 85%, 85% gross margins. There is very little businesses, especially in the hardware realm, that makes that type of margin") and one striking data point on SK Hynix: "between 13 to 15 customers secured around 40% of their projected profit for next year. So they've already sold out their supply for the entirety of 2027." Their view on when the shortage eases: "that bottleneck isn't going to get unlocked until around 2030." Limitless

Cramer flagged that Micron had pulled off something rivals could not, crediting an executive by name: "a big shout out to Sandra Merotra at Micron for being able to get something that nobody else has, which is DRAMs." He has repeatedly called Micron the best "commodity" chip play. Squawk on the Street

The contrarian view was sharp. On RiskReversal Pod (July 23), venture investor Rick Heitzmann called the memory names "so toppy," pointing to SK Hynix's $26.5 billion US listing and Samsung immediately signaling it would do the same "at a time where you couldn't think of growth being any higher... These guys are growing sales and earnings 100% year over year." His warning was historical: memory is brutally cyclical, "some of the memory companies were trading at two to three times earnings a decade ago," and he reached back to the dot-com bust: "Cisco took 26 years to get back to their February 2000 pricing." RiskReversal Pod

Part of the stock weakness is a Korea story, not a demand story. On The Morning Filter (July 20), Morningstar's Dave Sekera explained that over 50% of Korea's entire market cap is just two stocks, Samsung and SK Hynix, that the Korean market recently entered a bear market (down about 25% from its high), and that "a lot of margin accounts ended up getting called in," forcing selling. Even so, he noted, that market is "still up 50% year to date and it's doubled over the past 52 weeks." The Morning Filter The Limitless hosts made the same point, estimating Korean investors were "over-leveraged to the tune of about a billion dollars" and calling the sell-off "just oversold." Limitless

3. Chip-Making Equipment and WFE (ASML, and the "Fab Five")

ASML, the Dutch company that makes the lithography machines used to print chips, was the standout in equipment. On The Circuit (July 20), hosts Ben Bajarin and Jay Goldberg said ASML "reported very strong EPS for the quarter and guided revenue about a billion and change above expectation," is "raising prices pretty significantly," and is "expanding capacity this year and next year." One detail investors fixated on: ASML guided to roughly 85 EUV systems for 2027, and some had hoped for closer to 90, leading to talk that ASML may be sandbagging. As Goldberg put it, "they're probably can do more than they say. Just why would they tell us?" The Circuit

Morningstar's Dave Sekera explained why his team raised ASML's fair value so much. It was about future capacity, not the quarter: "they're increasing capacity by 30% in 2027... and potentially increasing capacity by another 30% in 2028." He said the 2027 capacity is "essentially already booked" and about 30% of the 2028 increase is booked too, which pushed his 2030 revenue forecast up to $70 billion from $60 billion. The new ASML fair value: $2,050 per ADR. The Morning Filter

The Circuit also flagged a genuinely new technical milestone: Intel is now using ASML's next-generation "high-NA" EUV machines (the most advanced, and most expensive, lithography tools). Bajarin noted the ASML machines needed for Intel's 18A process are "at least two and a half times the size of prior nodes," which has real implications: you need far more clean-room and building space than in the smartphone era. In Goldberg's blunt framing, "you got to build three more buildings to make less chips than you did in a smartphone." The irony they highlighted: Intel was famously late to adopt EUV, and is now first to adopt high-NA EUV. The Circuit

4. Foundry and Manufacturing (TSMC, Intel)

This was the deepest topic of the week.

TSMC. On The Circuit, the hosts said TSMC beat on revenue and EPS, guided next quarter above consensus, and, most importantly, "raised their CapEx forecast" to north of $60 billion, with "70 to 80% allocated to advanced process." Morningstar put exact numbers on it: TSMC lifted its 2026 capex budget by $8 billion to $62 billion (a 15% increase) and raised revenue guidance to over 40% growth, prompting a 27% fair-value hike to $534 per ADR. Sekera's read: "the CapEx increase this year is because they're already talking to their clients, already talking to the hyperscalers... I'm assuming those hyperscalers are also increasing their 2027 budgets." The Morning Filter Separately, Wallstreet Trapper reported TSMC's June sales surged 67% year over year, with first-half 2026 revenue of NT$2.4 trillion (about $75 billion), up 36% from 2025. Trappin Tuesday's

TSMC is also raising prices. On Mac OS Ken (July 22), host Ken Ray relayed a Nikkei Asia/AppleInsider report: TSMC will raise base chip prices 5% to 10% starting next year, with an extra 10% to 15% premium for customers who order more than they forecast, meaning a customer could pay up to 25% more "if they make a mistake on their estimates." A TSMC spokesperson called the increases "strategic, not opportunistic." Apple, as a major long-term client, is expected to face closer to the 5% end. Mac OS Ken

Intel was the week's big surprise. On CNBC's Fast Money (July 23), CNBC's Christina Partsinevelos summarized what she got directly from CFO Dave Zisner: Intel "nearly doubled earnings and raised its forward guidance," "locked in 10+ customers on long-term agreements on both price and volume" ranging 3 to 5 years, is "committing to ramping 14A [its most advanced process] into high-volume production by 2028," and has "line of sight to billions of dollars" in advanced-packaging business. Crucially, she said Intel is "undershipping... well into Q3," meaning, like the memory makers, it now has visibility into demand it can't yet fully supply. CNBC's "Fast Money"

The hard numbers (from The Rundown, July 24): data-center revenue grew 59% to $6.3 billion with CEO Lip-Bu Tan saying demand is outpacing supply; the foundry business grew 31% to $5.8 billion; Apple was added as a manufacturing customer; and the stock is up nearly 300% since the US government invested $8.9 billion for a 10% stake last August. The Rundown

On the CapEx detail, Bloomberg's Ed Ludlow (Bloomberg Surveillance) relayed what the CFO told him: spending will rise to about $20 billion this year (up from an $18 billion prior guide, and above the roughly $15 billion the Street expected), but stressed CEO Lip-Bu Tan "is a very cautious guy. He does not deploy capital and increase CapEx unless he's sure he's going to get a return on it." Bloomberg Surveillance

Cramer was effusive: "What Lip-Bu Tan has done at Intel is just nothing short of a miracle... Intel is my favorite chip stock." Squawk on the Street

The skeptics were loud too. Gene Munster of Deepwater (Fast Money) put Intel's success in perspective: its data-center/AI segment is "1/13 the size of Nvidia's GPU business," grew 56% versus Nvidia's expected 96%, and on 2027 estimates "Intel trades at 61 times... Nvidia trading at 17 times. AMD... 31 times." His verdict: "there's just so many better companies to invest in the infrastructure... Intel still is at the little kids' table." CNBC's "Fast Money"

Bloomberg Intelligence's Mandeep Singh raised the deeper question, whether Intel is selling a system or just a chip. "Nvidia is bundling everything. Google TPUs are bundling everything... Intel is not selling you a system. They're just selling you a chip." He also flagged the "surprising" puzzle that Intel says it is supply-constrained even though it owns its own fabs. Bloomberg Businessweek

Where Intel might actually have an edge: advanced packaging. On the Chip Stock Investor Podcast (July 21), Nicholas and Kasey Rossolillo argued Intel's packaging technologies (EMIB and Foveros) offer a real alternative to TSMC's CoWoS, the packaging that stitches chips and memory together. Their thesis: Nvidia "essentially monopolizes TSMC's advanced packaging capacity for training," so as the market shifts toward AI inference, customers like Google (for its TPUs) "may prefer Intel's simpler packaging solutions." Their caution: "the door's been cracked open... they have to execute on that. The financials have to catch up." Chip Stock Investor Podcast

Bloomberg Intelligence's Kunjan Sobhani added that 70% of Intel's revenue this quarter was "AI exposed or AI related," and credited the turnaround to execution rather than the government stake: "It has all been execution since the new CEO coming in... The yields are getting better, costs are coming down." Bloomberg Intelligence

5. Analog, Auto and Industrial Chips (TXN, ADI, NXPI, STM)

Coverage here was thin and mostly filtered through market reaction rather than deep analysis. On Squawk on the Street (July 23), Cramer walked through the group: STMicro was "down almost 18%"; Texas Instruments fell about 10% "and then the stock was down 10 despite a quarter a lot of people liked," though it was "making a comeback"; and NXP "was good. It was up five." Squawk on the Street

CNBC's Christina Partsinevelos framed the core problem for the analog names on Fast Money: unlike Intel's multi-year contracts, when "Texas Instruments says that the cycle is improving... it's not telling us that demand is going to start climbing at least for the next 8 quarters." In other words, the auto/industrial recovery lacks the visibility investors are now demanding. CNBC's "Fast Money"

6. China, Export Controls, and China's Chip Push (SMIC, Huawei, Kimi)

The China thread ran through several shows and centered on one idea: China keeps producing competitive AI at a fraction of the cost and compute.

The headline event was Kimi K3, a new model from China's Moonshot AI. On Motley Fool Hidden Gems Investing (July 20), the hosts said it "jumped to the top spot among AI models," ahead of Anthropic's and OpenAI's latest, is "rumored to be three times cheaper to run," and is "an absolute data monster with 2.8 trillion parameters." Tellingly, Moonshot "had to freeze new user signups just 48 hours after launching because their servers literally hit a physical limit," which the hosts used to argue that computing power, not model design, is still the real scarce resource. Analyst Rachel: "even when that software blueprint is free, running it safely at scale can be really difficult... a lot of the raw AI intelligence is becoming a cheap commodity... the cloud infrastructure required to run it, not so." Motley Fool Hidden Gems Investing

On RiskReversal Pod (July 20), Dan Nathan spelled out the strategic contrast: US frontier labs are "spending hundreds of billions of dollars to develop these models," while China does it "very differently on much less" (DeepSeek recently raised $7 billion at a $52 billion valuation, and Alibaba announced $50 billion in capex over three years), "and let's be clear, they're doing them without Nvidia's most advanced chips." He noted the Bloomberg headline "China's Moonshot Unveils AI Model That Narrows Gap With US Firms" was one of the reasons AI names sold off that day. RiskReversal Pod

Cerebras CEO Andrew Feldman was blunt about the stakes on The MAD Podcast: China is "an industrial adversary." He said China is "behind in chips" but has made big investments in power and the grid ("a real weakness in the US"), and that its "approach was at the next level as open source models... very good," if not quite at the level of GPT, Anthropic or Gemini. He also confirmed the full domestic stack is real: "Huawei send chips for deep seek." The MAD Podcast with Matt Turck

On policy, two data points surfaced. First, Nvidia has just started shipping H200 chips to China again after they were newly approved, a "nice regional tailwind" per Motley Fool, though the bears stressed it's a reversible policy decision. Motley Fool Hidden Gems Investing Second, Jensen Huang said Chinese models are "OK" but wants any US deployment kept domestic: "I want the data kept here, and I want tight cybersecurity." Squawk on the Street On RiskReversal (July 23), Heitzmann noted US companies like Airbnb and DoorDash are quietly testing Chinese open-source models even as Treasury Secretary Bessent talks about pushing back on "distillation" (Chinese models trained off US models). RiskReversal Pod

7. Earnings Reactions at a Glance

A recurring theme: strong results, weak stocks. The clearest summary came from Christina Partsinevelos on Fast Money: "TSMC posted its best quarter in history and fell. ASML, TXN raised outlooks and yet their stocks fell. Samsung printed record profit and yet fell. But Intel locked in these long-term agreements... and that's why they're raising CapEx." CNBC's "Fast Money"

One footnote worth knowing on Alphabet's headline number: The Rundown pointed out that Google's $9.11 EPS "way higher than the $2.88 Wall Street was expecting" was inflated by roughly $98 billion in paper gains on its stakes in SpaceX and Anthropic (Google owns about 5% of SpaceX, worth about $94 billion at quarter-end but about $65 billion now after SpaceX fell about 30%). Those are unrealized gains, not cash. The Rundown

8. Mergers and Deals

Tesla and SpaceX chatter got louder. On Brew Markets (July 23), Ann Berry noted a Goldman Sachs analyst report listing "key person risk" and the "internal control environment" as risks, and argued a merger "is now seeming ever more likely." She pointed to TerraFab, a $25 billion chip-fabrication facility in Austin that is already a Tesla/SpaceX partnership, as evidence the two are converging. On Tesla's own call, Musk said only: "we can't talk about combining companies on an earnings call. It's got to be done with the appropriate process." Cramer added on Squawk: "The more this goes down, the more I think it is natural." Brew Markets

Nvidia's $20 billion purchase of Groq was framed by Cerebras CEO Andrew Feldman as a strategic admission: "the GPU architecture could not do fast inference and that this market was large and growing quickly... they paid $20 billion for the number two" inference-chip company. The MAD Podcast with Matt Turck

Robotics land-grab. On AI Chat (July 22), host Jaeden Schafer reported Anthropic held (unclosed) acquisition talks this spring with Physical Intelligence, an $11 billion robotics startup, while Samsung is separately in talks to back French AI lab Mistral at a EUR 20 billion valuation. Both Anthropic and OpenAI are "racing to build robotic capabilities" ahead of planned IPOs later this year. AI Chat: AI News & Artificial Intelligence

Vendor financing to watch. The Circuit flagged that Nvidia has formalized a program to directly backstop "NeoClouds," up to $30 billion of off-balance-sheet guarantees where Nvidia agrees to buy or rent GPUs a customer can't sell. Goldberg's worry: "why does Nvidia have to work so hard to get these deals done?" if AI demand is truly so strong. The Circuit

9. Cyclicality, Peak/Trough, and the Bubble Debate

This is where the week's most interesting disagreement lived.

The "scarcity is breaking" argument. On RiskReversal Pod (July 20), Guy Adami and Dan Nathan argued the entire bull case rests on chip scarcity, and that story is springing leaks. SpaceX and Meta now have excess computing power they're renting out (Anthropic and Google each signed about $1 billion deals to rent SpaceX capacity), but "the problem with those contracts is they were 3-month" deals, not the durable long-term agreements the bulls assume. As Adami put it: "what you just described blows a hole in the whole bull thesis of scarcity." RiskReversal Pod

Rick Heitzmann extended it on the July 23 episode: the industry has moved from "grow at all costs" to an "efficiency phase," where tools like OpenRouter route each task to the cheapest capable model, and "the most premium tokens to do the same job are actually over 100 times more expensive." His parallel to the late-1990s telecom bust was explicit: "Cable and Wireless destroyed $40 billion of equity market cap in eight months... these capital-intensive equipment, chip, memory, infrastructure companies are incredibly cyclical. And people forget." RiskReversal Pod

The options-market warning. On Excess Returns (July 18), derivatives veteran Andy Constan described what he calls the "Fab Five," a handful of companies (memory makers, compute-chip makers, one raw-silicon maker, plus equipment suppliers) absorbing enormous capex-funded money. He argued the options market flashed a classic top signal: unusually high single-stock volatility plus "flat skew" (call options bid up more than protective puts). "It almost always happens at a bottom... It just doesn't happen at tops." His core thesis: "there is simply not enough GDP to earn money from all the investment that's being made... some people are going to be left without pie." He stressed the bubble isn't in the multiples (some chip stocks are cheap) but in the earnings expectations: "companies doubling, tripling their earnings year over year... and people expect it to continue." Excess Returns

The "this is just rotation, not a crash" counter. The Circuit's hosts made an important, more measured point: nearly every semiconductor stock fell over the prior two weeks. "good results, bad results, everybody's down... it has nothing to do with fundamentals. It's just the buy side is rotating into other areas." As Goldberg put it: "I don't think the bubble's over... but there's a lot less enthusiasm for semis in the investment class than there used to be." Money is hunting the "non-obvious names" further down the supply chain (PCBs, MLCCs, smaller components) rather than the crowded megacaps. The Circuit

The "serious people" defense. Cramer acknowledged the 1999 comparisons but drew a distinction: back then, telecom customers like Lucent and Nortel were quietly missing numbers, whereas today the big AI customers are not. "if OpenAI came out right now and said, look, things are soft, then... that would be a new narrative. We cannot afford that narrative. They're not saying that." His stated pain threshold is the bond market: he's watching whether the 30-year yield pushes toward the "six and a quarter" that he says "was the make or break in 1999." Squawk on the Street

There is also a growing view that the next scarce resource is power, not chips. The Limitless hosts noted US data-center power demand is "basically doubling... from 31 gigawatts to 66 gigawatts in 24 months," pushing data centers' share of total US electricity from about 1% toward about 3%, and argued money is now rotating out of memory and into the electricity trade. Limitless

What I'm Watching Next Week

• The big hyperscaler earnings: Meta, Microsoft, Amazon and Apple. Dan Ives expects all of them to keep raising capex ("no one's going to slow CapEx") and says the key question has shifted "from CapEx to monetization." Mandeep Singh's simple rule of thumb: if every hyperscaler raises capex, "all the semiconductor names will catch a bid." Watch whether investors reward the spending or punish it, as they did with Alphabet. Bloomberg Intelligence
• Nvidia's earnings at the end of August. On Charles Payne's podcast, Simpler Trading's Danielle Shay noted Nvidia typically rallies into its report and said she's watching for a move back above its 50-day moving average in the chip ETF (SMH). Charles Payne's Unstoppable Prosperity Podcast
• Intel's follow-through. Analysts want what the company still won't give: named foundry customers and a backlog number to model the 14A ramp. Mandeep Singh: "if Intel really has that visibility... they should give a backlog number." Bloomberg Businessweek
• The Korean memory market. Whether the forced selling (margin calls) in Samsung and SK Hynix stabilizes, and whether DRAM prices keep climbing while the stocks lag. The Morning Filter
• China policy. Follow-through on Nvidia's H200 shipments to China, any move on "distillation," and how fast Kimi K3 (and DeepSeek/Alibaba) narrow the gap. RiskReversal Pod
• Tesla and SpaceX "appropriate process." Watch for any formal step toward the combination Musk hinted at, given the TerraFab chip-fab tie-up already in place. Brew Markets
• Rates and oil as the swing factor. With Brent around $100 and the 30-year Treasury yield above 5% for 12 straight days (the longest stretch since 2007), Cramer's warning is that a further move higher is what would truly "tip negative" the AI/chip trade. Squawk on the Street