The stock market stopped arguing about demand this week. Nobody doubts that every board on earth wants more artificial intelligence models, faster response times, and warehouses full of servers. What investors suddenly started arguing about is something far more grounded: who is actually writing the check? You could see the anxiety spreading across the tape all week. A company can print record revenue, post the highest gross margins in semiconductor history, and watch its stock drop double digits before lunch because the bill is coming due. On Thursday, just 26 chip and software giants accounted for nine-tenths of a $1.32 trillion single-day wipeout in the Nasdaq 100.
The tell was in the breadth. Two-thirds of individual stocks in the S&P 500 actually marched higher this week while the overall index slid 0.47%. The only sector that managed to finish in the green was regulated utilities, up 0.83%. Think about that contrast for a second.
The high-flying tech names that spent two years dominating your portfolio traded like distressed paper, while boring power lines quietly attracted capital. With the 10-year Treasury yield hovering right near 4.99%, money is not free anymore. The cheap debt that funded boundless server farms is drying up fast.
Demand was never the real problem. Nvidia’s Jensen Huang put it plainly at an industry summit: compute appetite is off the charts, but the missing piece in the middle is the working capital required to bankroll that buildout. Credit desks are already pricing in real friction. High-yield spreads are twitching, and private credit is stepping into gaps that public debt markets refuse to touch.
The four companies on our list today sit at four distinct points along that massive, tangled invoice. One designs custom chips and backstops customer leases. One holds a near-monopoly on high-speed memory. One operates the server park while burning through cash. And one owns the nuclear reactors powering the entire machine.
Scorecard: Grading last week's calls
Before diving into this week’s lineup, let us review our calls from last week. Measuring performance strictly from Friday close to Friday close, the results delivered a clean split.
Amazon dropped roughly 0.8% against a broader market dip of 0.5%, leaving our Accumulate stance essentially untested ahead of its upcoming earnings report. SoFi hovered flat with a 0.4% gain, keeping our cautious entry call idling in neutral until it reports its own numbers.
Applied Materials handed us a win on patience, sliding about 7.8% after equipment peer Lam Research spooked the market with soft sequential guidance. We told you to wait for a better entry, and the group gave you one, even if the selloff came from sector contagion rather than the company's own quarterly release.
Constellation Energy, on the other hand, was an outright miss. We called an Accumulate, and the stock tumbled about 6.6% as policy caps crushed merchant power valuations across the board. The contract cash flow thesis held, but regulatory headwinds caught us off guard and cost you six points in five days. In total, two calls remain untested, one wait was right, and one accumulation call was early and wrong.
1. Broadcom (AVGO) — The Bespoke Tailor
Nvidia sells suits off the rack; Broadcom is the bespoke tailor. It measures your shoulders, drafts a custom pattern, and delivers a unique accelerator designed exclusively for your internal software. The company does this for an ultra-exclusive club of roughly six massive tech giants on earth.
In plain English, Broadcom helps cloud giants like Google and Meta design their own internal chips, known as XPUs. These hyperscalers build their own silicon because they want to dodge Nvidia’s hefty profit margins. Broadcom pairs these processors with high-speed networking switches that stitch thousands of chips into a single, cohesive supercomputer. Alongside this semiconductor engine sits a massive enterprise software portfolio anchored by VMware. The strategy is working: custom AI chips accounted for 56% of Broadcom’s total corporate revenue last quarter, climbing rapidly from 49% three months earlier.
The financial performance proves the engineering prowess. In its quarter ended August 2, Broadcom hauled in $29.591B in revenue, up 86% year over year and 33% sequentially. Operating profit hit $15.955B, translating to a non-GAAP operating margin of 67.9%. Free cash flow reached $13.665B, converting 46.2 cents of every single revenue dollar into cold cash.
Because Taiwan Semiconductor manufactures the physical silicon, Broadcom runs lean. Capital expenditures were a modest $532M, roughly 1.4% of sales over the past year. Looking ahead, management guided fiscal 2026 AI semiconductor revenue to $58B, up from earlier targets, with expectations of roughly $115B in 2027.
Yet the stock remains deeply hated. It sits about a third below its June peak, trading near 16.6x forward earnings. That multiple is roughly half what investors gladly pay for AMD or Marvell. Why the discount?
The market is terrified of customer financing games. Broadcom recently partnered with private-credit funds to help lease its expensive server racks to a frontier AI laboratory. Under that arrangement, Broadcom wrote an explicit backstop: if the customer defaults, Broadcom covers the shortfall between 85% of the unpaid balance and whatever the hardware fetches on resale.
That creates an undiscounted liability of up to $29B that Broadcom currently marks as not material. The customer behind this deal is burning cash, holding roughly $518B in future compute obligations against only about $20.3B in cash. If that client stumbles, those custom server racks will have virtually zero second-hand value on the open market. Furthermore, Google is already qualifying a second supplier to build its next-generation chips.
What we're watching: The December quarter report is due Dec 9, as management said on the last call. It must hit guided AI semiconductor sales of roughly $21.7B within $34.8B of total revenue. More importantly, we are tracking whether that $29B financing backstop sees any draws or gets revalued off its current not material accounting tag in the quarterly filings.
Our take: Accumulate. Paying roughly 16x forward earnings for the preeminent custom chip designer provides a comfortable margin of safety, and the financing liabilities are fully visible. We would change our tune if that leasing guarantee triggers actual cash outlays or if December AI revenue misses the $21.7B mark.
2. Micron (MU) — The Memory Merchant
Every advanced AI accelerator requires a high-speed scratchpad sitting directly next to it. Micron sells that scratchpad in a global oligopoly controlled by just three major suppliers, and right now the buyers are stuck waiting in line.
The business model is straightforward: Micron manufactures dynamic random-access memory, known as DRAM, and solid-state NAND flash storage. DRAM holds active operational data for immediate access, while NAND preserves your files after the power cuts out. AI clusters rely heavily on high-bandwidth memory, which stacks multiple DRAM dies like pancakes alongside the primary processor. Micron is the only American player among the big three, sharing the space with Korean rivals SK Hynix and Samsung.
The recent numbers look completely wild. In the quarter that closed its fiscal year on September 3, Micron posted sales of $54.229B, up 379% year over year. GAAP gross margin reached an eye-popping 86.8%, driving full-year adjusted free cash flow to $62.31B. This was not a modest volume gain; it was pure pricing leverage.
DRAM pricing climbed in the high teens sequentially while shipment volumes grew only mid-single digits. Crucially, Micron locked down 26 multi-year take-or-pay agreements, guaranteeing about $150B in forward obligations. That effectively sells out its high-bandwidth memory output through calendar 2027 at premium pricing.
Despite this windfall, the stock sits roughly 30% below its summer high, trading at an absurd 4.2x forward earnings. The market is not mispricing the stock; it is treating these profits as a fleeting cyclical spike. Gross margins near 87% are historically unsustainable in commodity manufacturing. They signal an exit point, not an entry.
Micron is already pouring its windfall right back into the ground, projecting $25B in capital spending over the first half of fiscal 2027 alone. Cleanrooms take years to build, and this fresh capacity will arrive in late 2028 just as competitive supply from China's CXMT and Samsung comes online. If memory pricing rolls over while multi-billion-dollar depreciation charges ramp up, operating margins can drop like a stone. The contracts protect 2027, but 2028 remains totally exposed.
What we're watching: First-quarter earnings expected in mid-to-late December, where guidance calls for $61.5B in sales and non-GAAP gross margins around 86.25%. The key metric will be forward capital expenditure guidance and whether customers are willing to extend take-or-pay agreements into calendar 2028.
Our take: Wait for a better entry. The long-term supply agreements soften the downside, but buying a memory maker at historical peak margins right as it doubles factory spending carries severe cyclical risk. We would shift to Accumulate if gross margins stay above 80% while 2028 contracts begin filling the backlog.
3. Microsoft (MSFT) — The Utility
Microsoft built the premier digital office park, installed the plumbing, and now leases office space to everyone else, including a premier tenant in which it owns a 27% stake. The rent roll is swelling fast, but the maintenance bills are compounding even faster.
At its core, Microsoft rents computing power and model access through Azure, while collecting recurring monthly software fees for Windows, Office, and its Copilot workspace assistant. In fiscal 2026, Azure officially crossed $100B in annual revenue while expanding at a 41% clip. To lower its massive server costs and reduce its reliance on third-party silicon, Microsoft introduced its custom Maia 300 chip. The company aims to deploy over one million units next year, claiming the design delivers 40% more compute per megawatt than competing GPU setups.
The top-line performance remains solid. Fiscal 2026 revenue rose 18% to $331.8B, generating $155.2B in operating profit at a 46.8% margin. Commercial remaining performance obligations expanded 84% to reach $678B, while paid enterprise Copilot seats doubled to 30 million. The company continues to dominate corporate technology budgets across every enterprise sector.
However, the stock trades near 22.1x forward earnings and faces real friction around cash efficiency. Free cash flow dropped 6.5% to $66.99B even as top-line revenue expanded. The company now converts only 20 cents of each incoming dollar into free cash, down from 25 cents. The reason is a massive surge in capital expenditures, which jumped nearly 80% to $115.95B. Bolting capital-heavy server infrastructure onto a 90%-margin software empire has pulled cloud gross margins down from 68% to 65%.
The real vulnerability sits off the balance sheet. Microsoft carries nearly $637.6B in lease and purchase commitments, with $169B coming due in fiscal 2027 alone. Under take-or-pay server terms, Microsoft pays landlords whether clients consume the compute or not. Meanwhile, nearly 45% of its commercial cloud backlog depends on OpenAI, which is currently responding to a broad FTC information-gathering probe into the artificial intelligence sector. Excluding OpenAI, underlying growth tracks closer to 25%.
What we're watching: Fiscal first-quarter earnings on Wednesday, October 28. We are watching for Azure constant-currency growth above 45%, capital expenditures landing around the expected $50B mark, and margin details inside the newly segmented financial reporting that breaks out core cloud consumption.
Our take: Wait for a better entry. Paying a rich multiple for declining cash conversion days before an earnings report offers very little upside room. We would look to Accumulate if Azure growth comfortably beats the mid-40s range and cloud gross margins successfully stabilize.
4. Constellation Energy (CEG) — The Baseload Baron
If every tech titan wants to build massive kitchens on the same city block, Constellation owns the only high-pressure water main. Its nuclear fleet delivers uninterrupted power around the clock, producing a steady baseline that cannot be easily matched by solar or wind farms.
Constellation operates the largest nuclear fleet in the country, managing 22 gigawatts across 14 stations. Following its Calpine transaction earlier this year, it expanded into natural gas and geothermal, controlling 55 gigawatts of total generating capacity while serving 2.5 million retail utility accounts. The business model centers on converting uncontracted wholesale nuclear output into 15- to 20-year power purchase deals with power-hungry cloud operators. It has already signed long-term deals with Microsoft, Meta, Amazon, and Google, aiming to lock up roughly 30% of its nuclear capacity under private contracts by 2032.
The corporate financials clearly reflect this growing structural power demand. Adjusted operating earnings reached $2.55 per share in the second quarter on $7.504B in revenue. Management lifted its full-year earnings expectations to a range of $11.50 to $12.50 per share, projecting combined free cash flow of $8.4B across 2026 and 2027. Backed by multi-billion-dollar corporate power commitments, Constellation repurchased roughly $2.2B of its own stock this year under a broader $5.0B authorization program.
Yet the stock dropped 37% from its peak, sliding 28% year to date and compressing its forward multiple to roughly 19x earnings. The selloff was triggered by administrative caps rather than customer demand. In the regional PJM grid auction, capacity pricing hit a strict administrative ceiling of $325 per megawatt-day. Without that cap, market clearing prices would have landed near $555, wiping out an estimated $13B in potential merchant revenues to protect retail electric bills.
The short argument focuses entirely on regulation and debt. Constellation's net debt jumped from $5.4B to $24.0B over the past six months following acquisitions and capital spending. Real trailing free cash flow after growth expenditures dwindled to just $309M. With residential power bills becoming a heated political topic, regulators are pushing back against high prices. PJM proposed an emergency backstop allowing up to 6.8 gigawatts of competing new generation to bypass existing caps, creating new above-cap supply that could dilute pricing power across the grid.
What we're watching: The revised PJM backstop proposal due October 29, which will clarify whether data center owners can contract around existing generating assets. We are also tracking third-quarter earnings on November 9, watching for consensus earnings of $3.72 per share and further updates on contracted pricing premiums.
Our take: Accumulate. A forward multiple of roughly 19x accounts for regulatory friction while offering exposure to irreplaceable clean power assets that would take a decade to construct from scratch. We would downgrade our view if regulatory updates permit hyperscalers to bypass merchant nuclear fleets entirely.
The bottom line
The investment environment changed this week. The trade is no longer about who can write the most impressive software algorithm; it is about who can finance the staggering real-world hardware bill without breaking their balance sheet. Broadcom cuts custom chips to order while guaranteeing customer leases. Micron captures record-wide memory margins ahead of an impending capital spending flood.
Microsoft leases server space at scale while watching its operational cash conversion slip. And Constellation sells scarce, round-the-clock electrons beneath the thumb of government price ceilings. We will track every link in this financing chain as corporate earnings unfold.
Analysis, not advice. We may own these names and we may sell them at any time.
