On Monday in Singapore, Broadcom head Hock Tan took the stage and punctured the market's favorite assumption. Capital, he warned, might be the single biggest constraint on the entire buildout, adding that many investors set "very extreme expectations" only to find "it's almost like they are in a casino." He estimated the world spends roughly $200 billion annually generating computing tokens while clawing back only $150 billion in direct revenue.
Across five punishing sessions, the twenty largest semiconductor and artificial intelligence firms shed over $1 trillion in market value. The Nasdaq Composite slid roughly 4.5% and the S&P 500 lost nearly 2%. Credit markets showed real strain: rented-compute pioneer CoreWeave saw its five-year default probability surge to roughly 63%.
The argument running through your portfolio has fundamentally shifted. For two years, equities rallied on the assumption that customer appetite for compute was endless. Today, nobody doubts that the demand exists.
Investors are suddenly obsessing over who writes the checks, what terms they accept, and how much debt gets piled onto balance sheets. Money itself has become the bottleneck. That reality surfaced everywhere from a hot consumer price index print to the ten-year Treasury yield swinging past 5.30% before settling near 5.10%.
The four stocks on our list this week sit right at the collision point between physical ambition and financial math.
Our scorecard from last week illustrates how quickly the ground is shifting under individual theses. Alphabet slipped about 3% as it heads into its October 28 earnings print, where we need to see 60% cloud growth alongside margins above 35%. Micron dropped nearly 15% after competitor advancements in China and semiconductor executive warnings rattled memory pricing, though contract rates still hold positive territory.
Vistra saw its catalyst calendar formally broken when regulators suspended a key grid reliability backstop into early 2027, forcing us to defer our procurement test entirely. RTX was the lone bright spot, up more than 4% on fresh defense contracts. Our cautious stance on that entry looks wrong heading into earnings.
1. Amazon (AMZN) — The Check Writer
Amazon is the world's biggest retail storefront by day and one of its biggest computing landlords by night. Most shoppers know the firm through cardboard boxes on their doorstep and Prime video streaming on their televisions. Beneath that consumer surface sits Amazon Web Services, an infrastructure juggernaut that rents processing power, data storage, and machine-learning capacity to global enterprises by the hour.
AWS delivers outsized operating muscle, generating $16.6 billion in operating profit on $42.2 billion of revenue in its second quarter, translating to a staggering 39.4% operating margin. Across the enterprise, net sales hit $200.6 billion, up 20% year over year, while advertising expanded 26%.
Yet the stock trades roughly 13% beneath its 52-week peak, changing hands at roughly 24 times forward earnings. The market is hesitating because Amazon has the largest capital budget in its history, and the spending has drained its cash flow. Andy Jassy raised projected 2026 cash capital expenditures to approximately $220 billion, noting that even this massive sum fails to satisfy inbound compute demand.
That spending produced negative $8.8 billion of free cash flow in the second quarter alone, and long-term debt roughly doubled to $119.1 billion over the past year. Wall Street is uneasy about reported plans to shift roughly $8 billion of processors into off-balance-sheet leasing vehicles to preserve capital.
The short case is straightforward: Amazon is outspending its cash generation to purchase depreciating server racks, bridging the deficit with debt. Free cash flow turned negative over the trailing year, and setting up special financing vehicles to lease back graphics cards signals deep anxiety over residual equipment value. If these chips become obsolete in three years, Amazon is effectively offloading depreciation risk to yield-seeking lenders while saddling itself with inflexible fixed obligations.
If enterprise cloud spending cools, Amazon retail cannot bridge the cash drain. The bears win if AWS growth slips beneath 30% while management refuses to trim its $220 billion capital budget. Operating income under the guided $22.5 billion floor would do it too.
What we're watching: Third-quarter financial results arrive on October 29. Management guided sales between $197 billion and $202 billion alongside operating income between $22.5 billion and $26.5 billion. We need AWS expansion at or above 35% with operating profit securely within range.
Another capital expenditure hike paired with negative free cash flow kills the setup.
Our take: Accumulate. Backlog sits at $496 billion, and enterprise scale cushions the aggressive buildout.
2. Applied Materials (AMAT) — The Machine Shop
Applied Materials is the master toolmaker standing behind the global silicon economy. If chip designers like Nvidia sketch blueprints for digital engines, Applied manufactures the physical, precision-heavy equipment required to print those designs onto microscopic silicon wafers. It sells complex deposition, etching, and inspection tools directly to foundries and memory makers, pairing equipment sales with a lucrative, recurring maintenance services arm.
In its latest reported quarter, revenue jumped 25% to a record $9.1 billion. Gross margins held at 50.3%, and the firm converted $2.3 billion of free cash flow, with a systems mix dominated by logic and high-performance memory.
Despite guidance targeting $10.25 billion in revenue for the October quarter, the stock has dropped roughly 19% across five sessions, sitting about 45% under its summer peak. Investors panicked after an investor day where management outlined ambitious 2030 market-share and margin targets. It then explicitly refused to forecast the size of the wafer fab equipment market.
The stock trades near 32 times current-year earnings and 22 times forward estimates, meaning investors are already paying up for an assumed 44% surge in near-term profits. When a cyclical supplier asks shareholders to bet on a multi-year deck without quantifying end-market demand, institutions head for the exits.
The short argument centers on paying a cycle-peak multiple for a business tethered to volatile equipment buying waves. Memory accounts for a quarter of Applied's systems revenue, and China's CXMT is now advancing a denser memory cell, exactly the kind of transition that reshapes equipment orders. China generated 30% of sales last fiscal year as well, and the company remains under active scrutiny from multiple federal agencies examining export-control compliance.
When two top customers represent about a third of your business, any pause in foundry expansion plans crushes margins. If total chipmaking equipment spend flattens heading into 2027, this multiple lacks any valuation floor.
What we're watching: Fiscal fourth-quarter numbers, projected around November 12. We want revenue at or above the $9.75 billion low end, gross margin near 50.4%, and a January outlook that avoids sequential contraction.
Our take: Wait for a better entry. A cyclical equipment vendor demanding a premium multiple while refusing to size its market offers negative risk asymmetry.
3. Constellation Energy (CEG) — The Landlord
Constellation Energy owns more reactors than anyone else in the country, which makes it the closest thing the AI boom has to a landlord. A traditional regulated utility has to ask state commissions for permission to raise rates. Constellation sells into competitive wholesale markets instead, across a 55-gigawatt fleet that includes 22 gigawatts of nuclear.
Its fleet runs around the clock with zero carbon output, and most of it sits inside PJM. That is the mid-Atlantic grid where data centers keep adding load. In the second quarter, adjusted operating earnings climbed 33% to $920 million, and management lifted its full-year earnings target to a range of $11.50 to $12.50 a share.
The equity sits about a third below its 52-week ceiling, trading around 20 times forward earnings despite landing back-to-back agreements with technology titans. In early October it locked in a 20-year pact with Google for 890 megawatts of uprated nuclear capacity across six sites, plus 2,700 megawatts of existing output. That came days after a 690-megawatt deal with Amazon.
Yet shares dropped 9% across the week. Wall Street is starting to discount the immediate earnings power of these deals. Delivering the extra nuclear output takes billions of upfront dollars and years of construction.
The bear case focuses on the massive gulf between current capital outlays and distant cash returns. Out of the headline Google transaction, 2,700 megawatts represent power that is already running, not new capacity. The 890 megawatts of fresh output takes $4.3 billion of outlays and will not fully arrive until 2032.
Borrowing at yields near 5% to fund mechanical upgrades that arrive next decade is real financing drag. Statutory GAAP net income also dropped 39% last quarter, revealing how heavily adjusted metrics rely on acquisition accounting. If local grid auctions cap prices politically rather than allowing power rates to clear at market scarcity levels, Constellation gets stuck holding expensive development assets.
What we're watching: Third-quarter financial updates, anticipated around November 9. We need operating earnings pacing toward the elevated full-year guide, paired with clear engineering milestones for the initial Limerick reactor uprate. The trade breaks if regional capacity mechanisms cap pricing administratively without contract carve-outs.
Our take: Accumulate. Long-term commercial contracts with balance sheets like Google permanently underwrite cash flow, insulating baseload nuclear power from wholesale commodity volatility.
4. SoFi Technologies (SOFI) — The Bank in Your Pocket
SoFi is a fully chartered modern bank wrapped inside an intuitive smartphone application. Instead of operating physical brick-and-mortar branches, it delivers high-yield savings, stock trading, credit cards, and automated budgeting features straight to 15.8 million mobile members. Most of its revenue is the gap between what consumer and student loans earn and what deposits cost.
It also licenses its banking software to other fintechs. In its second quarter, net revenue surged 42.5% to $1.22 billion and net income rose 61% to $156.6 million. Deposits climbed to $45.5 billion, and management guides full-year revenue to $4.75 billion to $4.85 billion.
Yet the stock recently touched a 52-week low, down roughly 44% year to date and 56% below its 52-week high. At roughly 24 times this year's guided earnings, SoFi is priced like a plain bank rather than a growth story, because rates refuse to cooperate. When borrowing costs stay elevated, consumer lenders face two headwinds.
They must pay competitive yields to keep deposits, and their borrowers feel more repayment pressure. Wall Street is split wide open: KBW reiterated Underperform and cut its target, Goldman cut its target while staying Neutral, and Benchmark upgraded to Buy the same day.
The short thesis targets the sensitive accounting assumptions embedded across the bank's personal loan portfolio. SoFi marked its $26.1 billion of unpaid personal loan principal at 104.7% of par value in June, down 71 basis points sequentially. For every 100 basis points of fair-value premium that burns off due to macroeconomic stress, the bank takes roughly $261 million in pre-tax markdowns.
The funding tells the story: loan balances grew faster than deposits last quarter, and operating cash flow ran negative $3.9 billion. If deposits stop growing faster than loans, funding shifts to wholesale markets at a spread the company itself puts near 156 basis points. The 5.98% net interest margin compresses from there.
What we're watching: Third-quarter results scheduled for October 27. We want earnings per share pacing toward the $0.60 annual mark, net interest margin holding 5%, and deposits growing faster than loans with no fresh markdowns on personal credit. A margin compression below 5% invalidates the thesis.
Our take: Wait for a better entry. Balance-sheet expansion running ahead of deposit gains in a stubborn interest-rate climate threatens book value before credit metrics stabilize.
The bottom line
The market spent this week learning that artificial intelligence is an industrial supply chain wired directly into the cost of capital. That transition rewards discipline and punishes premature valuation, separating the companies managing liquidity from those burning it. Amazon writes the infrastructure checks.
Applied Materials builds the machines, Constellation owns the reactors, and SoFi carries the consumer loans that pay for all of it. We will track each of these four setups against their concrete earnings hurdles over the coming weeks.
Analysis, not advice. We may own these names and we may sell them at any time.
