$MU $TSM $CXMT CXMT's expansion deserves a place in the DRAM bull case. Charles Kau says 10 fabs are being expanded, with capacity expected within two years. TechNews reported his warning on Nov. 10. DIGITIMES followed on Nov. 11, quoting his skepticism about TSMC entering 3D DRAM. The expansion figures remain Kau's account, pending independent verification against CXMT's project disclosures. My concern is how additional supply travels through the market. If CXMT supplies more Chinese demand locally, overseas suppliers could lose orders even without CXMT exporting those chips. That would put pressure on pricing if supply grows faster than demand. For Micron, I'd watch CXMT's qualified output, customer adoption and DRAM contract prices. The investment risk becomes measurable as those numbers move.

$CRWV $NBIS CoreWeave and Nebius are selling into the same shortage, and both have now priced short-term capacity above $40M per MW a year. Where they differ is funding. At the last balance sheet, CoreWeave carried $4.7 of debt for every $1 of customer prepayments. Nebius carried $1.4. Serenity's line on older GPUs is the one we'd underline too. CoreWeave said Ampere and Hopper rental prices rose 30-50% quarter over quarter in Q3 and are now above their original contract prices. A fleet that's already financed is earning more as it ages, and that's what lenders care about most. CoreWeave's position: $46.9B of debt at Sep 30 against $10.0B of prepayments. Q3 net interest was $846M, adjusted operating income $379M, and Q4 interest is guided at $1.05-1.15B. The way out runs through the old fleet. Management put 2027 renewals at close to $4B of extra annualized revenue at today's prices with no new capex, and more than $3B of the ~$9B in prepayments expected over the next three months comes from re-contracting infrastructure whose financing is already repaid. Its next delayed-draw term loan is expected at 7-7.5%. Nebius's position at Jun 30: $8.5B of debt, $6.0B of prepayments, $8.0B of cash. Its newest secured money is cheaper, a GPU-backed loan at SOFR + 2.50% and a Citi facility at SOFR + 2.75%, about 6.4% and 6.6% with SOFR at 3.88%. It also sold $2.8B of stock through its ATM at an average $223.6 a share, against Tuesday's $133.96 close. So on rates we're with Serenity. Nebius's harder test is delivery. It's targeting 5 GW of contracted power by year end and more than 1 GW a year of deployment from 2027. Its Microsoft contract lets Microsoft terminate a GPU service if Nebius misses the agreed delivery dates after a grace period and can't provide alternative capacity. Nebius reports Thursday before the open, with the call at 8am ET. We're watching four numbers: 1. Annual contract value per MW, against the >$40M short-term level it flagged in August 2. Prepayments collected, against the >$9B it expects for 2026 3. How much of the 5 GW is energized and on schedule 4. Debt on the balance sheet once the $5.75B of August convertibles are in

$CRWV Q&A Highlights | Financial Services Account for 10% of Backlog; Enterprise Software Customers Are Becoming Compute Customers When UBS asked about AI labs building their own compute capacity and customer diversification, management disclosed that 10% of CoreWeave’s backlog comes from financial services. Management said the company already counts 10% of Fortune 500 companies as customers, and more than half of new customers this year came from outside AI labs and the technology sector. Exact quote: “10% of our backlog is from financial services.” Management added that 70% of new cloud customers in Q3 were already Forge customers. Enterprises are using software and development tools first, then beginning to purchase and deploy cloud infrastructure, translating into actual customer conversions. Industry implications: Enterprise AI adoption has reached a stage where orders and purchases can be observed. Industries such as financial services are becoming quantifiable sources of compute demand. Conversion from software tools to production deployments provides a new growth path for broader infrastructure demand.

$CRWV Q&A Update | 2028 Capacity Not Yet Formally Offered for Sale In a follow-up response, the CEO clarified that 2028 capacity has not yet been formally brought to market; the company is waiting for construction to mature further before selling it. The earlier statement that “2028 capacity brought to market has already been sold” is corrected to: 2027 is essentially sold out; demand for 2028 is already several times capacity, but sales have not formally begun. The previous related Feed has also been updated. Exact quote: We haven’t brought 2028 to market because we are waiting for that to season further.

$CRWV Q&A | Queued buyers could absorb all capacity; purchase orders placed when cloud contracts are signed Management said current buyers in the queue could buy all of the company’s capacity. The company is choosing to wait patiently and bring capacity to market when some infrastructure development is more mature, seeking better economic returns. Exact quote: “We could sell all of the capacity that we have to current buyers that are in the queue.” Asked how it controls cost increases between signing and delivery, management explained that it places corresponding purchase orders and locks in pricing when signing customer contracts. The customer’s data center, site costs and operating expenses are already defined at signing. Exact quote: “As we sign customer contracts, we also place the corresponding POs and lock in the pricing.” Industry implications: Strong demand gives platforms capable of delivery room to choose when to sign contracts; customer contracts also translate into equipment purchase orders at the same time. For the AI buildout supply chain, tracking newly signed contracts and corresponding procurement is closer to actual order placement than looking only at capacity plans.

$CRWV | Earnings call Largely sold out through 2027; marketed 2028 capacity sold Management said CoreWeave is largely sold out through 2027. It clarified that 2028 is not fully sold out, but all infrastructure offered to the market for that year has been sold. This provides concrete sales visibility for future capacity amid investor concerns about industry oversupply.

$CRWV | Earnings call Parent-level capital needs expected below five billion through 2027 Management expects to raise less than five billion in incremental parent-level capital through 2027, after accounting for available funding sources. This quantifies the previously described limited need for parent-level financing.

$CRWV | Earnings call First customer signed for platform on customer-owned infrastructure Management said it has signed its first customer for a platform offering that runs on third-party infrastructure. The model lets CoreWeave monetize demand without funding the underlying assets, while customers retain direct infrastructure ownership. This introduces a less capital-intensive route to growth alongside CoreWeave's owned infrastructure deployments.

$CRWV | Earnings call Next DDTL expected within weeks at 7–7.5% yield Management expects to launch its next DDTL in the coming weeks, with multiple rating agencies indicating a high investment-grade rating. Based on early discussions and current rates, management expects a 7–7.5% yield while financing more than 90% of associated capital expenditure. The planned facility would fund most associated deployment costs at a lower yield than the previously disclosed upper end of loan pricing.

$CRWV | Earnings call Q4 deliveries pulled into Q3; construction in progress declines Management said Q3 capital expenditure of $14.5 billion exceeded guidance because certain Q4 deliveries were pulled forward. Construction in progress fell to 9.3 billion from 11.9 billion as infrastructure was deployed against power available late in Q2. The spending overshoot reflects accelerated deliveries and conversion of unfinished infrastructure into deployed capacity.

$CRWV | Earnings call Deliveries expected to begin by year-end for contracts covering over half of backlog Management said more than 50% of backlog is attached to contracts where customer delivery is expected to commence by the end of this year. This indicates the timing of initial deliveries for a substantial portion of backlog, without specifying how much revenue will be recognized this year.

$CRWV | Earnings call Management expects existing contracts to cover secured and parent-level debt Management expects cash flow from existing customer commitments to service and repay both outstanding debt and additional debt needed to deliver those contracts. It explicitly included asset-level secured debt and parent-level unsecured debt, before counting renewals. This extends management's stated debt coverage expectation to parent-level unsecured obligations without relying on renewals.

$CRWV | Earnings call Two Q3 price increases; more capacity allocated to short-term contracts Management said it increased pricing across SKUs by approximately 25% in July and another 10% later in the quarter. Pricing continued to rise in Q4, more than offsetting cost increases, and management is allocating more capacity to short-term opportunities where pricing justifies it. Higher pricing and a shift toward short-term contracts affect revenue potential and the duration of contracted revenue visibility.

$CRWV | Earnings call Large customer prepayments expected; limited parent-level funding through 2027 Management expects to collect approximately nine billion in prepayments and deposits from signed contracts over the next three months, including more than three billion from re-contracting existing infrastructure. It also expects EBITDA less interest, excluding prepayments, to more than double in 2027, and expects its funding sources to cover the significant majority of capital needs through 2027, with only limited funding raised at the parent company. These expected receipts and internally generated funds materially reduce the anticipated need for parent-level financing.

$CRWV Reject unsecured financing with the current rate of return, mainly relying on contract supporting loans The management said that there is no need to finance through the unsecured debt market with the current rate of return; it is expected that most of the remaining capital expenditure required for the delivery of orders on hand will be financed through delayed withdrawal of fixed-term loans that match specific customer contracts. By the end of the third quarter, the company’s overall weighted average debt cost was 7.9%, down 250 basis points year-on-year. By 2027, it has been promised that more than 75% of capital expenditure corresponds to investment-level and enterprise customers, and the rest corresponds to the world‘s leading AI laboratories and AI native companies. The clarification of the main financing path of subsequent capital expenditure and its customer base is an important basis for judging the availability of expansion funds and the cost of debt.

$CRWV The real alpha is buried in their Q3 project & infrastructure slides: 1/ First-to-Market Vera Rubin Benchmarks: Brought online multi-rack NVIDIA Vera Rubin NVL72 in production for Cognition — delivering 4.8x greater token throughput than GB200 NVL72. Also announced support for NVIDIA Vera CPUs built for AI agents. 2/ Massive Power & Global Footprint: • 52 active data centers (9 added in 2026) • 1.7 GW+ active power (+200 MW in Q3, doubled YTD; >1.85 GW year-end target) • 4.6 GW+ contracted power (+900 MW in Q3), including 1.3 GW+ overseas across India, Indonesia & Spain 3/ Software Stack & Defense Wins: • Launched CoreWeave Forge + Agent Lens (end-to-end agent SFT/RL/eval stack) • Partnered with Leidos for defense & intelligence sovereign AI workloads • Selected to run managed AI cloud for DARPA biological research • Expanded deals with OpenAI, Jane Street, Perplexity & Cohere + new wins with Canva & Wayve The AI buildout isn't slowing down — it's going vertical.

$CRWV CoreWeave's adjusted operating income tripled in one quarter to $379M, beating its own $200-260M guide by about $150M. Backlog jumped to $132.5B. The bill that still hasn't moved is interest: $846M this quarter, and guided to $1.05-1.15B for Q4. The quarter Revenue $3.66B, up 168% YoY, above the top of the $3.45-3.60B guide. Consensus was $3.54B. Adjusted EBITDA $2.26B, a 62% margin (Q2: 59%). GAAP operating income $207M, up from a $49M loss in Q2. EPS -$1.27 vs -$1.55 consensus. Guidance is up Full-year 2026 revenue now $13.0-13.3B, from $12.4-13.2B. After the first three quarters, that puts Q4 at roughly $4.69-4.99B. Consensus was $4.70B, which is the bottom of that range. Full-year adjusted operating income $1.275-1.375B, 26% higher at the midpoint than the old guide. Capex $37-39B for the year, low end raised from $35B. Exit-2026 annualized run-rate revenue $20-21B, from $18.5-19.5B. The interest bill Adjusted operating income covered 45% of net interest expense in Q3, up from 20% in Q2. Using the new full-year guide, Q4 adjusted operating income comes out at roughly $750-850M, still below the $1.05-1.15B interest guide. Capacity Active power over 1.7 GW, up more than 200 MW in the quarter. Year-end target stays at more than 1.85 GW. Contracted power over 4.6 GW, up more than 900 MW. Backlog $132.5B, up 27% from $104B at the end of Q2. Stock is $51.71 after hours, up 6% from the $48.77 close (4:08pm ET). For the 5pm ET call: C1 Q4 revenue: why does the implied $4.69-4.99B start right at consensus, and what decides where in the range it lands? C2 Power: only about 150 MW more by year-end after 200 MW+ in Q3. Is that caution, or are sites slipping? C3 Interest: what rate is the new $4.2B of convertibles and the $2.6B DDTL costing, and when does operating income pass interest? C4 Backlog: how much of the $28B added in Q3 came from OpenAI and the other top customers? C5 Capex: how much of the $37-39B is already funded?

$RGTI $497.1m in cash and investments buys development time; commercial traction still has to justify the roadmap. 1. Results Q3 2026 revenue was $1.411m versus $1.947m a year earlier. Nine-month revenue rose to $10.9m from $5.2m, but that growth masks a weaker quarter. Operating loss widened to $33.2m from $20.5m. GAAP net loss was $3.1m, versus adjusted non-GAAP loss of $20.5m; diluted losses per share were $0.08 and $0.06, respectively. A $25.3m noncash warrant fair-value gain helped GAAP earnings; it does not show operating improvement. September 30 cash and equivalents were $45.799m; the $497.1m total includes short- and long-term available-for-sale investments, down from $541.3m at June 30. Debt was zero. Outstanding shares increased to 342.7m from 331.3m at December 2025: funding capacity does not eliminate dilution. 2. Our reading Resources support continued development, but sales do not yet validate the scaling plan. The 108-qubit system reached 99.5% median two-qubit gate fidelity, up from about 99.1% in Q2. That is better fidelity at the same qubit count, not evidence of customer delivery. Commerce support advanced from a letter of intent to a definitive award of up to $100m, with a minority equity stake granted to the department. The award ceiling is not cash received or product revenue. Management now targets 300-plus qubits, about 99.7% median two-qubit fidelity and gates below 60ns by mid-2028. It retains the roughly three-year target of about 1,000 qubits, about 99.9% two-qubit fidelity and gates below 50ns. These remain forecasts. The approximately $20m 2026 revenue outlook depends on Q4 C-DAC acceptance and recognition. Confirmed acceptance with recognized revenue would strengthen our view of commercialization. 3. Task questions Q1 — Partly answered: revenue is reported; new order totals and completed customer deliveries are not disclosed. Q2 — Partly answered: losses, liquidity and share growth are clear; the Commerce stake percentage is not disclosed. Q3 — Answered: fidelity improved, a mid-2028 scaling target was specified, and conditional annual revenue guidance was provided. 4. Call questions C1: What remains before C-DAC acceptance? C2: What are Commerce's funding conditions, payment schedule and equity percentage?

$TSM TSMC booked NT$586.9bn in October, its biggest month ever by 14%. That's already running above the top of its own fourth-quarter guide, which works out to NT$576bn a month at TSMC's 31.8 exchange-rate assumption. Revenue rose 14.7% from September and 59.7% from a year ago. January to October is NT$4.49tn, up 43.3%. The quarter is guided at US$52.3–54.3bn, or NT$1.66–1.73tn at 31.8. October alone covers 34.6% of the midpoint. November and December only need to average NT$554bn to land in the middle, 5.6% below October. The top end needs about NT$570bn a month, 2.9% below October. October tends to be TSMC's strongest month of the fourth quarter, though. Last year it was 35.1% of the quarter, and revenue slipped 6.5% in November and another 2.5% in December. Run this year through that same path and the quarter lands around US$52.5bn, just above the low end. So October takes the low end off the table. Whether TSMC clears the top again, like its 2.0% beat in the third quarter, is a November question. On Oct. 15 we said we'd turn negative if October and November ran below the low end, about NT$554bn a month. October cleared that by 5.9%. Taipei had already closed when the filing hit at 1:45 p.m., with 2330 up 2.2% at NT$2,285 on the day. The first market read on this number is the ADR session. Other Taiwanese AI supply chain manufacturers will bring you updates later

$NVDA $TSM $SKHY Seoul and Taipei both took questions on Jensen Huang in parliament this week. Seoul answered about chips: Korea's contracted Nvidia GPUs will arrive on schedule. Taipei's premier answered about security, citing export controls, the Trade Act and the woman's entry records, and said nothing about orders. Korea's vice science minister Ryu Je-myung told the budget committee on Nov 9 that the 10,000 GPUs bought with last year's extra budget are already in use, this year's batch is contracted and arriving, and 2027 pricing and delivery are being worked out with Nvidia Korea. The "Nvidia may cut TSMC orders" line traces to a Commercial Times report this morning citing unnamed supply-chain sources. No company named, no wafer count. Same day, Taiwanese chassis, rack and cooling suppliers told UDN that shipments are normal and orders show no anomalies. The market is reading it the same way. Since the last close before Huang's Nov 8 letter, Nvidia is down 4.3%. TSMC is up 5.2% in Taipei and SK hynix is up 5.8% in Seoul. If wafer orders do move, export screening looks like the likelier cause than the affair. Commercial Times itself ties the risk to tighter export reviews and H200s piling up in Hong Kong. Next checkpoints: Nvidia's earnings on Nov 17, and the closed-door briefing House committee chairs have asked Huang to attend on Nov 18.