Alphabet stock (GOOGL): $482 billion gone. Can Google's clankers pay it back?

0:00 Alphabet bets on every AI horse. What do the odds have to pay?

If you were filthy rich like Alphabet, and could bet on every horse in a race against your peers, what would the odds have to pay when you win? The horses: chip, model, cloud, the way it reaches users, and data centers. We know the stake: $195 to $205 billion this year. Is it enough if only some of them win?

Say the chip goes bust. Does that loss eat the win on the others? And what has to be true for you to make money on the stock? The full research, with links and sources, is at shortcircuitcreations.com. Read it at your own pace. So anyways.

0:38 The 12.64% drop, and the line under Alphabet's $19.16 a share

The drop they mean: twelve point six four percent in one day. Alphabet's biggest since two thousand four.

The report hit after the close. That evening's low came during the call, when the CFO said next year's spending will "increase significantly."

Alphabet's profit per share over twelve months: nineteen dollars and sixteen cents. That's the reported number. It should mean the business is booming. So read the line under it. Eight dollars and twenty-one cents of that is the market writing up shares Alphabet owns in other companies. Mostly SpaceX. A paper gain, on shares Alphabet still holds. The holdings are real, on the books.

But the horses didn't earn it. The horses themselves earned ten dollars and ninety-five cents a share in twelve months.

Who earned it? For that we leave per share and count in billions. The line is operating profit: what a business keeps from its sales after the cost of running it, before tax and before the paper gains. Two businesses report one. Services, that's search and YouTube: thirty-nine point one six billion dollars last quarter. That's ads on search, and every other horse eats from it. The cloud: eleven point one two billion. If the cloud wins, the stake comes back as cloud profit. That's the horse the payout runs through. The other three have no line of their own. The model and the chip report nothing, the chip's loss lands in the cloud's number, and the data centers are the cost every horse runs on.

So the horse that pays for the bet today is search. Now the stake. One hundred ninety-five to two hundred five billion dollars this year, Alphabet's guidance. Next year, "increase significantly."

Now a full year. The last twelve months of operating profit, all five horses together. After the losses in Alphabet's other bets and the costs it carries for the whole group: one hundred fifty-six point nine eight billion. So the stake is one hundred twenty-four point two two to one hundred thirty point five nine percent of a year's profit. More than the stable makes, on every horse at once.

The stake comes back out of what the horses earn. So, per share: the latest quarter against the same quarter a year ago, with last year's EU fine taken out of the base. The horses earned eight point one zero percent more. The whole rise is one step, into the first of the last four quarters. Since then: two sixty-seven to two sixty-seven. For you to make money on the stock, and for the stake to come back sooner, that number has to climb. Keep that number.

First, the horse that carries the income today: the way it reaches users.

3:15 Horse one: Google Search, the default deals and AI Mode

Search brought in under half of Alphabet's sales last quarter. First time ever.

Search last quarter: sixty-three point three five billion dollars, up eleven point nine nine percent. The horse is the road to the users. Search is what the road brings in. Two questions on this horse: does the road hold, and does the pace hold. The road first. "Our year-over-year query growth rate is the highest it's been in the last five years." Pichai, on the call.

What speaks against it: Meta. Its ads brought in sixty-three point oh six billion, up twenty-five point nine two percent against search's eleven point nine nine. Meta sells to people scrolling. Search sells to people looking to buy.

The court last year: "No existing rival has wrested market share from Google." Where I land: the road holds today. More searches, and no rival has taken share. Part of the road is bought: the default spot. Those deals went to court in Washington, and in twenty twenty-four the judge wrote: "Google is a monopolist, and it has acted as one to maintain its monopoly."

The court's finding: the default deals carried roughly fifty percent of US searches.

The court let the payments continue. But every deal now runs one year at a time. If the deals renew, the stable keeps the income behind all five horses. If the iPhone deal goes, Google's forecast in court says sixty to eighty percent of iPhone searches go with it. But holding the road and keeping the pace are two different things. And the stake needs the pace.

The pace. Search grew nineteen point one three percent in the first quarter this year. Then sixteen point seven six. Then eleven point nine nine. Three quarters, three lower numbers. Asked about it on the call, the ads chief Schindler: "the strong growth in AI mode queries had an impact because of the lower monetization we were seeing in AI mode."

So AI Mode brings more searches, and each one earns less. More searches is Pichai's record. Less money per search is what AI Mode does to it. Same thing, two sides.

For the stake, holding isn't the bar. Search is the income under it, so this pace has to turn back up from eleven point nine nine. More searches alone won't do it; what each one earns has to grow too. Now the profit side of the same horse. Services, the search horse's profit: plus five point eight eight percent. Its sales grew ten point six three.

So on this horse, costs grew faster than sales, and the margin fell. Two quarters in a row the profit came in below the quarter before, while sales rose.

Against that: this year's legal costs are still in the five point eight eight. Take them out and Services is plus nine point five percent, the upper case, on a split Alphabet doesn't give. Over twelve months, fine out of the base, Services grew seventeen point six zero. That carries the nine months before this one. Five point eight eight is the strict one. One quarter is one quarter, for search as for the cloud.

Where I land: for the stake, the base holds, and the climb it needs didn't come from this horse. In fifteen years of investing, the quarter where costs outrun sales is the one I circle. The next quarter tells you whether it's a habit. What can go wrong on this horse is two things. The money per search falls faster than the searches grow. Or a default deal goes.

Either one lands in the same place: the income every other horse eats from. Say the chip goes bust; this is the income that would have to cover it. That matters most for one outcome: the one where the cloud doesn't grow, and the stable has to carry the whole stake on its own.

Next horse: the model, Gemini. It's the one changing what a search is worth, and it has no number of its own.

7:08 Horse two: Gemini, a cost line that grew 71%

Gemini. Google ships it like bloatware: inside search, inside the cloud, inside the app. So whatever it earns hides in those lines. What the report does show on its own line is the cost.

Alphabet's company-level costs, "primarily" its "shared AI research and development": seven point six six billion dollars last quarter. Up seventy-one point three six percent on a year earlier. That line sits outside both Services and the cloud. So the cloud's margin and search's margin are both counted before the model is paid for.

Back to the pace, the horses' profit per share, two sixty-seven to two sixty-seven. This line is a weight on it: it grew seventy-one point three six percent in a year, while the per-share profit grew eight point one zero. For the pace to climb, either this line stops growing or the other horses outgrow it.

Against the rivals, Google's own count, in court: Google's clanker, one hundred forty million questions a day. ChatGPT, one point two billion. Eight point five seven times. That count is from March last year.

Against that: Gemini does bring in money, through the cloud. The CFO: "Google Cloud has generated tens of billions of dollars in AI revenues year-to-date." No split for Gemini.

Now who buys there. The rival Alphabet has put money into, Anthropic, has filed to list. Anthropic has promised computing purchases of five hundred ten point six billion dollars, part of them conditional, against one hundred thirty-one point seven billion in cash. Google is one of the sellers. A customer whose promises are bigger than its cash, and the stable has bet on the horse that could beat it.

So the model shows up only in the other horses' numbers. Win, and you see it in search and the cloud. Lose, and the cost line is still here, with nothing across from it. And one question it leaves at the cloud: whether what the cloud sells is worth its price. Next horse: the cloud, where Gemini is sold and where the payout runs.

9:12 Horse three: Google Cloud, up 102%, and what it keeps per dollar

The cloud. The lift the last two horses didn't give came from here: cloud sales up one hundred two point five seven percent, to thirty point seven one billion dollars in the quarter. This is the line it all lands on.

What it keeps. Operating profit: eleven point one two billion. Set that against the sales and you get the operating margin: of every dollar the cloud sold, thirty-six point two one cents stayed as profit.

What speaks for it. The order book. Backlog is contracts signed and not yet delivered, so not yet sales. For the cloud alone: five hundred sixty point eight billion dollars.

The CFO, Ashkenazi, on the call: "We expect to recognize just over 55% of the total backlog as revenues over the next 24 months." Fifty-five percent of five sixty point eight is three hundred eight point four four billion, over eight quarters. Spread flat, the signed work alone is one point two six times last quarter's cloud sales, every quarter, for two years.

So the outcome the search horse pointed at, where the cloud doesn't grow and the stable carries the stake alone, is the one this paper argues against for two years. It's signed.

And the growth rivals held from this size. Google Cloud's last twelve months of sales: ninety-three point one seven billion. AWS, Amazon's cloud, started from the same size in twenty twenty-three and has grown twenty-three point two eight percent a year since. Microsoft's cloud as a whole started there four years ago: twenty-three point seven six a year. The rivals' growth, held from this size, is the bar, not the hundred and two.

What speaks against it. First, what's inside the hundred and two. Ashkenazi again: "TPU system sales were also a meaningful contributor." The fourth horse, the chip, sold as hardware. It's booked here, in the cloud's line, when the systems are delivered. In lumps.

How big a lump? No number. What we can read is the quarter before TPU systems started selling: the first quarter, cloud sales up sixty-three point three six percent. So the hundred and two has hardware in it, and the company won't say how much.

Which brings up the base. Run the rivals' growth forward five years from two bases. From the twelve months: two hundred seventy-three point one billion in twenty thirty-one. From last quarter times four: three hundred sixty-five point eight. Ninety-two point seven billion apart. For two years the signed work alone runs above the low curve. Past that, the quarter has the hardware in it, in a lump. One quarter is one quarter, for the cloud as for search.

Second, the margin. The question the model left here: is what the cloud sells worth its price. Thirty-six point two one cents per dollar, and AWS keeps thirty-eight point six five. They're not the same measure. Amazon's filing: depreciation on shared equipment, the yearly cost of machines already bought, is "allocated to all segments based on usage." So AWS's thirty-eight carries its share of the common cost. Alphabet's shared AI research, the seven point six six billion from the last horse, sits outside the cloud's line. The cloud's thirty-six is counted before it.

Put it in, by share of sales. The cloud is twenty-four point zero seven percent of Alphabet's sales, so it carries that share of the seven point six six. Take that off the cloud's profit and the margin is thirty point two two. The gap to AWS is eight point four three points, not two point four four. The thirty-six is the line the company reports; the thirty is what it's worth next to AWS.

When two companies print almost the same margin, the first thing I check is who left which cost outside the line.

Then where the lift came from. Twenty-three point seven one to thirty-six point two one in a year. The segment note: compensation, the staff cost, fell by thirteen point seven nine points of sales. The other costs, where the machines and the TPU inventory sit, rose one point two nine points. So the lift is people spread over more sales, lumps included.

And the machines are about to cost more. The fifth horse: since the very end of last quarter the cloud runs partly on rented capacity. Ashkenazi: "We expect margin headwinds in the near term as we utilize this capacity." The thirty-six is from before the headwind. The next quarter is the first with the rented capacity in it for its full length.

Rented racks because your own data center isn't built are the most expensive capacity there is. You pay the landlord's margin on top of the power and the gear. Alphabet's word for it is "bridging strategy". The bridge is rented.

Microsoft's cloud segment, as its capacity came in: operating margin fell. Against that: AWS held. Depreciation went from twelve point nine zero to twenty point zero seven percent of AWS's sales in two years, and the margin still rose, thirty-eight point zero six to thirty-eight point six five, because every other cost fell. So it can be done.

Third, who signed the order book. The last horse left a question here: a customer whose promises are bigger than its cash. Anthropic's filing: at least one hundred fourteen point five billion dollars to Google, through twenty thirty-three, "regardless of actual usage". If all of that sits in the cloud's backlog, it's twenty point four two percent of it.

Same company, three lines. One: the order book. Two: the holding. Anthropic's filing says Google holds more than five percent of it and put in ten billion in May. Three: the chip. Kurian on the call: "In addition to our work with Anthropic, both directly and via Broadcom." Customer, holding, chip buyer. One company.

What can go wrong on this horse is two things. The sales come and the margin doesn't: the order book delivers, on rented capacity, at fewer cents per dollar than today. Or the customer on three lines stops paying, and the order book, the holding and the chip's buyer fall on the same day. Either one lands in the same place: this horse's margin. The stake comes back as cloud profit, and cloud profit is sales times margin. Back to the pace, two sixty-seven to two sixty-seven per share: for it to climb, this horse's profit has to grow. The margin is the open number.

That's the horse the headlines saw. The number they saw was the hundred and two.

15:42 Not cheap, then the cheapest since 2012: the headlines

One week before the report. The Motley Fool, October twenty-first: "If Google Cloud can grow at a triple-digit rate (which I think it will), the stock price could skyrocket, making Alphabet a phenomenal stock to buy now." Triple-digit rate: delivered. The hundred and two. Skyrocket: biggest drop since two thousand four, eight days later.

Same article, a few lines up: "I wouldn't consider it cheap either." Nine days later. TipRanks, the morning after the drop: "Alphabet Stock Hits Cheapest Valuation Since 2012." Same stock. Not cheap at the Fool on the twenty-first; cheapest since twenty twelve at TipRanks on the thirtieth. Their word, both times. Willy-nilly recommendations. Haha hahaha!

And the Fool, the day of the drop, asking if it's "a Once-in-a-Decade Buy on the Dip?" Same site, April twenty twenty-five: "A Once-in-a-Decade Opportunity." July twenty twenty-five: "A Once-in-a-Decade Opportunity." Three decades in eighteen months.

TipRanks, further down: what makes it cheapest "is artificially suppressed due to non-cash investment gains over the past 12 months." Non-cash gains: the paper gain on shares, the line under the profit. In the article. Not in the headline. So the cheap headline skips one line. The horses showed it: paper gains on shares, not horses earning.

These clowns are welcome to stay on their own platforms. Less graffiti in public restrooms. And the hundred and two has a line too: hardware, in lumps, amount not given. "Triple-digit" has a chip in it, and whether it carries the win or eats it is still open.

17:31 Horse four: the TPU. Can Google beat Nvidia?

The chip. I asked it at the start: say the chip goes bust, does that loss eat the win on the others. This is that horse.

The TPU, Alphabet's own chip from the start, pays two ways. Inside Alphabet's own data centers, where Gemini and the cloud run on it. And as systems sold to customers: the lumps from the hundred and two. Both land in the cloud's line. Kurian, asked about margins: "we don't break out margins for individual products."

What speaks for it. The mechanism. Nvidia's gross margin, in its own filing: of every dollar Nvidia sells for, seventy-five cents is left after the product is made. A cloud with its own chip doesn't pay for those cents. It still pays the design and the same factories and memory everyone buys from. How much that saves, Alphabet doesn't say.

And the structure. Count where each rival has a line of its own, across the five layers on screen: own beside bought or rented still counts, a license doesn't. Alphabet has its own line on all five. Microsoft's model is licensed from OpenAI, through twenty thirty-two. Five of five is Alphabet alone. So if the chip delivers, the cents it skips stay in the house; no chip seller's margin sits between the chip and the cloud.

What speaks against it. The cost is in this year's margin and most of the sales are in next year's. The cloud's own filing on its profit: "partially offset by increases in inventory and related costs associated with the sale of TPU systems". That's the other-costs line in the cloud, the one that rose. So the thirty-six cents per dollar carries the chip's cost now, and the "significant majority" of the chip's sales is booked in twenty twenty-seven. Which way the margin moves when the lumps come depends on what the systems are sold at, and that's the number nobody gets.

And the systems can come back. The risk section, with words new this quarter: "contingent repurchase obligations for certain hardware and embedded software at prices below original cost". A number for one part only: six point six billion dollars, for systems a customer hands on to someone else. The rest has no total. The company's answer, same filing: "we do not believe counterparties have an economic incentive to exercise their option." Any system that comes back is a cost on the margin.

A chip you design yourself is cheaper per unit only when it's busy. In a data center you pay for the floor you lease and the power you reserved; an idle rack bills like a busy one. The filing's words for chips that wait are "excess capacity or inventory". So the chip's edge is a question of how full the racks are, and that number is one more Alphabet doesn't give.

So, the question from the start: say the chip goes bust. Going bust here is the advantage never showing up. The systems sell thin, at a margin nobody outside can see; or they come back below cost; or the cloud runs on bought GPUs and rented racks at the rented price. Then the cents per dollar the cloud keeps are the fifth horse's cents.

Does the loss eat the win on the others? The chip has no line to lose in. Its loss and its win are both margin points on the cloud's thirty-six cents. It eats the win if the systems sell below the cloud's own margin: each point comes off the thirty-six. It carries the win if they sell above it and the data centers skip Nvidia's cents: points on top. Either way it lands on the margin the whole stake stands on, and nowhere else.

The direction shows in the cloud's line once the lumps land, from twenty twenty-seven.

And the chip in Alphabet's own data centers only saves anything if the data centers are built and full. The data centers are the fifth horse.

21:24 Horse five: the data centers, and who pays for them

The data centers. The horse the other four run on, and the stake from the start, this year's, stands on it. What it buys, from the CFO: servers and network gear, and buildings, each written off over its own number of years.

Who earns the money today. Operating profit last quarter: Services, search and YouTube, against the cloud. The shared AI research and the moonshots cost nine point seven zero billion that quarter, eighty-seven point two three percent of the cloud's profit. So today the data centers stand on search's profit, and the cloud's pays for the lab.

What speaks for it: who paid so far. Nine months of this year: cash from operations one hundred twenty-seven point one five billion, capital spending one hundred thirty-one point nine five. So operations covered ninety-six point three six percent of the build; the rest, four point eight zero billion, is all the build needed from outside.

What speaks against it. First, the bill that hasn't started. Depreciation, the yearly cost of machines already bought, starts the day a thing goes into service. On the balance sheet: "assets not yet in service", one hundred forty-five point two seven billion dollars. Thirty-eight point nine six percent of all the buildings and machines Alphabet owns, with no depreciation on it yet.

What the pile costs when it starts. Take one year of this build, two hundred billion, split the way the CFO split last quarter: sixty-five percent machines over six years, thirty-five percent buildings over seven to forty. That's twenty-three point four two to thirty-one point six seven billion dollars a year in new depreciation. Alphabet's whole depreciation over the last twelve months was twenty-eight point zero billion. So each year of this build adds a second bill the size of today's. And next year's build is "significantly" bigger.

Against the stable's profit. A year's operating profit, all five horses: one hundred fifty-six point nine eight billion. The new bill is fourteen point nine two to twenty point one seven percent of it, every year, before the new data centers earn a dollar. That's the weight the two sixty-seven per share has to lift.

And the clock starts before the data center earns. I said it about the chip: an idle rack bills like a busy one. The data center has a second clock. A data center is never full the day the power comes on; the racks arrive for months after, and the customers after them. The depreciation runs from day one at the full rate. So the first year of every data center is a bill against a floor that's still filling, and the only thing that shortens that year is customers signed before the concrete. The order book says two years of them are.

Second, the paper. Alphabet used to buy back its shares. This year it sold paper instead. Nine months: debt, common shares and preferred stock raised one hundred forty-seven point one five billion dollars, against a build that needed four point eight zero from outside. The shares, in the filing: "primarily intended to be used to meet tax obligations". Buybacks: zero. Last year's annual report: "We engage in share repurchases". This quarter: "We have historically engaged in share repurchases".

The debt went from forty-nine point zero nine billion at the start of the year to one hundred twenty-seven point four nine at the end of September. In the new paper: a bond in sterling that matures a hundred years out, and preferred shares. And the share count, which hadn't grown since twenty eighteen, grew one point seven zero percent in nine months. Interest and a preferred dividend come out before the two sixty-seven, and more shares split it.

Against that, straight away. Cash and securities without the listed shares: one hundred seventy-four point seven one billion, against the one twenty-seven point four nine of debt. Forty-seven point two two billion more cash than debt, still. So the lenders aren't the risk. The owners' share is: new shares sold by a company with the cash not to.

And the cash flow that paid ninety-six point three six percent, read closer. Free cash flow, cash from operations less the build, last quarter: minus nine point zero seven billion, with a fine inside it. And the fourth quarter: the guidance leaves at least sixty-three billion of spending for it, and the best quarter of cash from operations Alphabet has had was fifty-two point four zero. So the fourth quarter's build is bigger than any quarter's cash. It comes out of the cash, or out of more paper.

Then what it looks like when one of these companies builds too much, from the one case with a receipt. Meta, twenty twenty-one to twenty twenty-two. Purchases of plant went from eighteen point five seven to thirty-one point one nine billion. Earnings per share fell thirty-seven point six two percent. The share fell seventy-six point seven four percent, from the top in September twenty twenty-one to the bottom in November twenty twenty-two. Meta's fall had more in it than data centers. Spending still went up while profit per share went down.

So what can go wrong on this horse. Not a crash in the news. A year of data centers that finds no customer at the price in the plan, and gets depreciated for six years anyway, at the full rate, year after year. The chip's cents never show up because the racks are empty, the rented bridge bills on top, and the cloud's margin takes it. What one year of build with nothing coming back does to the two sixty-seven is a number this stake turns on.

So the data centers are the cost every other horse runs on. Alphabet can afford them: net cash, and a cloud with two years signed. What it risks isn't the company. It's the margin and the owners' share: the new depreciation against a year's profit, per year of build, and the paper sold beside the build this year.

The build comes back as profit or it doesn't come back at all. Five horses, one stake, and the build has to come back on the two lines we've read all along: what the cloud keeps per dollar, and how fast Services grows. Those two decide what the two sixty-seven becomes.

27:44 Google stock prediction: what a share is worth in 2031 against the index

Those two lines, in one table. Today the stock is two seventy-nine eighty-four. The bar it has to clear is the index: the S&P 500's price over the last five years, put on this stock from today, is four hundred eighty-five dollars in twenty thirty-one. Every price from here on stands next to that one.

The assumptions, said out loud. The cloud sells three hundred sixty-five point eight billion in twenty thirty-one, the high base from the cloud horse. The shared AI research and the moonshots cost what they cost today. And the share count keeps growing one point seven zero percent a year, the rate of this year's nine months, no buybacks. The shares Alphabet holds sit outside the table, both directions.

Rows: what the cloud keeps of every dollar, the margin the cloud horse left open: today's thirty-six cents, forty above it, thirty and twenty below. Columns: how fast Services grows: the quarter's five point eight eight and nine point five from the search horse, and eight between them. Twelve boxes. Each one is what the share earns in twenty thirty-one, and what that's worth. Read the middle column, Services at eight, from the bottom. The number we kept was two sixty-seven a quarter. In twenty thirty-one, a year's worth: cloud at twenty cents, the share earns sixteen seventy-three. Four hundred eighteen dollars. At thirty: eighteen ninety-seven. Four seventy-four. At thirty-six, last quarter's margin: twenty dollars thirty-six. Five oh nine.

Every box is above today's two seventy-nine eighty-four. Beating the index is the harder bar: six of the twelve boxes clear it, and the line between them runs through the margin. On Services at eight the cloud has to keep thirty-one point nine cents. Two dials, on the high base.

Whether it's enough if only some of them win depends on which ones. Take the cloud out of the race: it keeps selling what it sells today, and the stable carries the stake alone. Services at five point eight eight: three forty-one. Above today's price. At the twelve months' seventeen point six zero: five sixty-five. To reach the index with the cloud standing still, Services has to grow fourteen percent a year for five years.

We don't think it does. Search's pace: two falls in a row, and last quarter Services' profit grew slower than its sales. Against that: over twelve months it did seventeen point six zero, above the fourteen. On the growth we believe, some horses winning is enough only if the cloud is one of them.

If the chip goes bust, the table has a row for it: the bottom one. The cloud sells the same and keeps twenty cents instead of thirty-six, because it buys Nvidia's chips and rents the racks at the rented price. Anchor: the cloud's margin a year ago was twenty-three point seven one. On Services at eight: above today's price, under the index. The row counts the margin only; the chip's sales gone, and the systems coming back, aren't in it, and those make it worse.

Whether that loss eats the win on the others: it eats the index. Ten points of margin, the thirty-cent row to twenty, are fifty-six dollars a share, four seventy-four to four eighteen, and every point between is five dollars and sixty cents. That's what the fourth horse is in this race: five sixty a share per margin point, either direction.

Where we land. The cloud wins the sales. It keeps thirty cents of every dollar, down from thirty-six. Services grows eight: the quarter's five point eight eight with the legal costs in, nine point five with them out, and eight sits between. Eighteen ninety-seven a share in twenty thirty-one, and the share is worth four hundred seventy-four dollars. Right around the index.

We think the sales come, because "just over 55%" of the order book is due inside twenty-four months, and that alone runs above today's quarter for two years. So we take the quarter as the base, one hundred twenty-two point eight billion: for two years it's signed. From there the cloud has to grow twenty-four point four zero percent a year, above what the rivals held, from a base none of them has grown from. Run the rivals' growth from the twelve months instead, and the cloud sells two seventy-three point one, and the share is worth four thirty-two. The base is the third dial.

We think the margin slides to thirty, because the company itself says headwind, and the headwind has four parts we've read. The rented racks bill from this quarter on, at the landlord's margin. The depreciation on this year's build hasn't started, and when it starts it runs at the full rate from day one. The lift to thirty-six came out of the staff share, with TPU hardware in the sales line in lumps. And Microsoft's cloud margin fell when its capacity came in. Burning myself on hype taught me to read one good quarter as one good quarter. I don't carry a thirty-six cent margin for five years.

Against that. AWS held its margin while its depreciation climbed seven point one seven points of sales, because every other cost fell, and Google Cloud has done it once already, twenty-three point seven one to thirty-six point two one in a year. The cloud is at thirty-six today, and the line it has to hold for the index is thirty-one point nine. We're calling it under a line it stands above right now. We'd like to be wrong.

It goes both ways, and the table says how far. If it keeps going the way it is, thirty-six cents: five oh nine to five thirty-five, above the index, unless Services brakes to the quarter's five point eight eight: four seventy-five. On our thirty, Services sets it: four forty-one to five hundred, and at eight point six a year the stock meets the index, inside the quarter's span. And the fifth horse: one year of data centers with nothing coming back costs twenty-two to fifty-six cents a share a year for six years. On our row, four sixty to four sixty-nine.

And what the odds have to pay, when you're rich enough to back every horse: the stake comes back as cloud profit, nowhere else. Today the cloud earns forty-four point five billion a year. On the thirty-cent row in twenty thirty-one it earns one hundred nine point seven. That's the payout on our row, next to the stake: one ninety-five to two hundred five billion this year.

One thing under every box. All of these prices assume the market keeps paying for profit close to what it pays today. Meta, twenty twenty-one to twenty twenty-two: profit per share down thirty-seven point six two percent. The stock, down seventy-six point seven four.

So, is the strategy sound? For the company, yes. Five horses, the build paid out of operations, net cash, two years of cloud signed. Alphabet can afford to back every horse in this race, and it's the only one with a line on all five.

The number to watch. The fourth quarter report, by last year's calendar: profit per share without the paper gains, against two sixty-seven. Above it, the pace turned up. At it, no climb in a year. Below, two quarters down in a row. One quarter, first reading, nothing more. What shows whether we were wrong takes longer: the cloud's margin against thirty-one point nine, and Services' growth against eight, report after report.

The lesson this stock taught me: the number that looked like growth was a different number every time. Nineteen sixteen a share had paper gains in it. The step a year ago had fines in the base. The hundred and two has hardware in it, booked in lumps. Read the line under the number before you believe the number.

35:07 The Hype Fund buys Alphabet

Our money is already tied to Alphabet from two sides. The Hype Fund owns SpaceX, and Alphabet's reported profit moves with that stock. And it owns Broadcom, which designs the chip with Alphabet.

After this video, the Hype Fund buys Alphabet. Same money as every other pick, measured against the S&P 500 on the same days, and you see the loss as plainly as the gain. We think four seventy-four, right around the index. The fund buys it anyway. That's the experiment. Margin above thirty-one point nine, Services above eight, and the stock beats the index: the experiment earns. Both under, it trails the index, and that shows too.

The whole research, every link, is on the site; the link is in the description. Ask me again when the fourth quarter lands.

36:08 To the moon

To the moon! Haha hahaha!