August 17, 2026, before US open
Written before entry.
Position
Long GOOGL at a reference price of $346.04, taken pre-market on August 17, 2026 ahead of the opening auction, with the bid at 345.95 and the ask at 346.21. The fifty-two week range is $196.15 to $408.61 and the thirteen week low is $314.90. The stock is up 10.63% year to date. Annualized historical volatility is 37.8%. A quarterly dividend of $0.22 carries an ex-date of September 4. I intend to hold six to twelve months.
Thesis
Alphabet is the only hyperscaler that designs the accelerator it buys, so a dollar of capital expenditure lands in its own gross margin instead of in Nvidia’s. The second quarter is the first place that shows up cleanly.
Google Cloud revenue grew 82% to $24.8 billion against consensus near $22.4 billion. The line worth reading sits below that one. Cloud operating income more than tripled to $8.8 billion and segment margin went from 20.7% to 35.6%. Infrastructure businesses compounding at 82% normally give margin away, because capacity is expensive and the marginal customer is won on price. Alphabet added roughly fifteen points of margin while nearly doubling revenue. Tensor Processing Units are a cost line for Alphabet and a revenue line for Nvidia, and the gap between those two treatments is the entire position.
Backlog supports the case rather than carrying it. Google Cloud backlog reached $514 billion, up more than $50 billion sequentially from $460 billion in the first quarter and against $106 billion a year earlier. Backlog is contracted, which puts it in a different evidentiary category from management commentary about demand.
Advertising has also not broken, which is the quieter half of this. Search and other advertising grew 17% to $63.3 billion and YouTube grew 13% to $11.1 billion, inside consolidated revenue of $119.8 billion, up 24% and the twelfth consecutive quarter of double digit growth. The structural bear case for three years has been that generative search would erode monetization. It may still be right. It has not reached the number.
Valuation is where I have to be careful. TradingKey put the trailing multiple at 17.87 times on August 3 with the stock near $360 and called it the lowest since 2019, which works out near 17 times here. Two different second quarter earnings figures are in circulation, though: $2.85 adjusted in CNBC’s coverage against a $2.89 estimate, and $9.11 in Motley Fool’s, described as triple expectations. A gap that size is almost certainly unrealized gains on equity investments rather than anything operating. Until I reconcile it against the 10-Q, the multiple in this paragraph is decoration.
Catalyst
Third quarter results in late October, though I have not confirmed the date against Alphabet’s investor relations calendar. Two lines settle more than the rest. The first is whether Cloud margin holds after Anat Ashkenazi warned of third quarter pressure from leasing outside capacity as a bridge. The second is where 2027 capital expenditure guidance lands: Truist’s Youssef Squali models $270 to $280 billion, revised up from $250 to $255 billion, and a print at or above that without a matching move in backlog is the adverse outcome.
The Morgan Stanley led financing for the Nexus Data Centers campus in Hubbard, Texas has not closed. When it does, the disclosed maturity, interest cost and guarantee structure will show how much contingent exposure Alphabet has actually taken. Until then the terms are reported rather than filed.
Risks
Berkshire Hathaway’s second quarter 13F, filed after Friday’s close, showed roughly 106 million Alphabet shares worth $37.9 billion at June 30, an 83% increase in the quarter and now its third largest United States listed holding behind Apple and American Express. The market read that as a verdict on value. The mechanism interests me more: about $10 billion of it came from a private stock purchase in early June, when Alphabet was raising fresh capital to fund the buildout, with the balance bought on the open market. A company that places stock privately to pay for capital expenditure is telling me operating cash flow was not sufficient on its own. That is a fact about funding, and it argues against the self-financing version of the thesis above rather than for it.
Free cash flow turned negative at $5.9 billion in the quarter, against $39.1 billion of operating cash flow and $44.9 billion of capital expenditure, itself double the prior year. Morgan Stanley calculates trailing twelve month free cash flow still positive at $53.3 billion and lists the funding sources as operating cash flow, equity and debt financing, leasing, custom chips and infrastructure efficiency. So the quarter is a crossover and not a collapse. Contractual commitments now exceed $800 billion, and the floor that a decade of reliable free cash flow used to put under the stock is gone for as long as this lasts.
Then there are the guarantees. Google has covered billions of dollars of Anthropic’s lease and power payment obligations across four data center leases and the related power purchase agreements, which is how Nexus raised a $14 billion bridge loan and a revolving facility for a Hubbard campus with its own 1.6 gigawatt gas plant. Google takes roughly 20% of the project equity in return, and Anthropic fills the site with Tensor Processing Units co-designed with Broadcom on separate vendor financing. Alphabet therefore competes with Anthropic through Gemini, sells it chips, holds equity in it, and stands behind its landlord’s debt. This is the structure the market repriced in July when the names were Nvidia, OpenAI and SoftBank, and Alphabet is inside it now. None of it appears as capital expenditure.
The spending trajectory is the older problem and it has not improved. Full year 2026 guidance went from $175 to $185 billion in February, to $180 to $190 billion in April, to $195 to $205 billion in July, against models near $188 billion, and the stock fell about 5% after hours on a quarter that beat on revenue, cloud and margin. Depreciation is the same risk arriving late. Assets bought this year run through the income statement in 2027 and beyond, so if revenue lags the expense, earnings compress even where the strategy works.
Memory costs are already reaching the hardware line. Alphabet has raised phone prices citing a severe memory crunch, which is the same crunch the SKHY thesis is long. That is a real cash flow offset between the two positions and it does not extend to the share prices, which have repriced together on every capital expenditure headline this year. Correlated multiples are the exposure the mandate’s sector cap does not measure.
Two risks here are mine rather than the company’s. I have not reconciled the earnings figures above, so I am entering without a valuation I would defend under questioning. And I am carrying the same factor twice, deliberately, having written down in the SKHY memo that I would try not to.
Entry
The stock sits 15% below its fifty-two week high of $408.61 and 76% above the low. The post-earnings low was $315.04 on July 24 and the recovery from there is most of what has happened since. The Berkshire filing landed after Friday’s close, so today is the first session that prices it, and I would be buying into whatever that reaction turns out to be. I do not have a read on it and I am not going to pretend the timing is anything better than neutral.
Limits
As set out in the investment-mandate, the position is capped at 30% of the account with a 25% price stop, which sits at $259.53 from here, roughly 17.6% below the thirteen week low and far enough out that ordinary volatility should not reach it. The thesis is invalidated if Cloud operating margin declines in two consecutive quarters, if Search and other advertising growth falls below 10% year over year, if Cloud backlog declines sequentially, if 2027 capital expenditure guidance exceeds roughly $280 billion without a matching increase in backlog, or if Alphabet announces a second guarantee of a third party’s lease or debt obligations at a scale comparable to Nexus.