Even Google Had To Ask

Google does not sell stock. That has been the rule for twenty-one years. The August 2004 IPO at $85 a share, the September 2005 follow-on at $295, and then silence. Every dollar of growth since — Android, YouTube, AdSense, Cloud, the entire data-center build that runs the internet most people use — was paid for out of profits. No equity raises. No begging. The cash machine funded itself. That was the whole story.

This week the cash machine asked for money. $84.75 billion of it, after oversubscription pushed the deal up from $80 billion. Class A and Class C stock, a depositary share series, an at-the-market program lined up for the third quarter, and a $10 billion private placement anchored by Berkshire Hathaway — the first major artificial intelligence investment Greg Abel has signed since taking over from Warren Buffett on January 1. The shares priced at a 6 percent discount to where they traded the Friday before. GOOGL opened down 3.9 percent on Tuesday and stayed there.

The first read in the financial press was dilution panic. The right read is harder and more interesting. Alphabet is the single most profitable software franchise ever built. It generated $73 billion in free cash flow last year. It has $98 billion in cash on the balance sheet. And it still needed eighty-four and three quarters of a billion in fresh equity to keep up with the bill it is about to write for AI infrastructure. The capex curve bent past organic capacity. Even at Google.

The Brief

• Alphabet announced an $80B equity raise on June 1, then upsized to $84.75B by June 2 after demand outran the book. Structure: $30B in underwritten Class A, Class C and depositary share offerings; $40B in at-the-market sales starting Q3 2026; $10B private placement with Berkshire Hathaway. ✓ Cleary Gottlieb · 2026-06-02
• Berkshire bought $5B of Class A at $351.81 and $5B of Class C at $348.20 — a roughly 6% discount to the prior close. Bringing Berkshire's total Alphabet position to about $26.6B, or 9% of its $325B equity portfolio. ✓ CNBC · 2026-06-01
• This is Alphabet's first follow-on equity offering since September 2005, when it raised capital at $295 per share. For 21 years between then and this week, every dollar of Google's growth was funded out of operating cash. ✓ TechCrunch · 2026-06-01
• Alphabet's 2025 capex was $91.4B (74% higher than 2024). Sundar Pichai guided 2026 capex to $180–190B and said 2027 will be "significantly higher." Semianalysis CEO Dylan Patel warned earlier this year Google's free cash flow could touch zero in 2026 under this trajectory. ✓ 24/7 Wall St. · 2026-04-02
• The Big-4 hyperscalers — Alphabet, Microsoft, Meta, Amazon — now plan $725B in 2026 capex, up 77% from $410B in 2025. Total global compute capex including Oracle, neoclouds, and sovereigns clears $1.04 trillion in 2026 — the first trillion-dollar year of compute capex in human history. ✓ Tom's Hardware · 2026-04-30

The 21-Year Wall Just Came Down

The number that matters here is not $84.75 billion. It is 21.

For twenty-one years Alphabet refused to issue stock. That was not stylistic. It was a deliberate signal — to employees holding RSUs, to founders who structured the dual-class system to keep voting control, to the market that priced Google as a self-funding compounder. Every other Mag-7 name has dipped into capital markets at some point in the last decade for one reason or another. Apple borrowed massively to fund buybacks. Microsoft did debt offerings. Meta issued debt for the first time in 2024. Amazon has been a regular debt issuer for years. Alphabet alone stayed pure. It was the one company where the free-market test — can you fund your own growth from your own cash? — got a perfect score year after year.

So when Alphabet's CFO sat down to negotiate the equity raise that closed Monday night, what got broken was not a balance sheet rule. It was an identity. The company that had spent two decades not needing the capital markets was now bidding for the capital markets' attention. And the deal got oversubscribed within hours. $80 billion announced, $84.75 billion priced. Berkshire wired in $10 billion at a 6 percent discount and Greg Abel's first AI position in his five months as CEO is now Alphabet.

That last detail is doing real work. Abel inherited Berkshire's $325 billion equity portfolio and roughly $400 billion in cash. He took over from a man who spent six decades famously skeptical of technology — Buffett told shareholders in 1999 he didn't understand it and would not invest. Abel's first material public move with that pile of capital was to write a $10 billion check directly to the company most exposed to the AI capex cycle, at a discount to market, in a primary offering. That is not a value purchase. It is a vote that the AI infrastructure trade is the trade.

The 3.9 percent drop in GOOGL on Tuesday was the dilution math. The underwriters sold $84.75 billion of stock into a market that had not priced the supply. Mechanically, that has to push the share price down a few percent. But the same trading day, Microsoft, Meta and Amazon all opened green. The market did not read Alphabet's raise as "Google is in trouble." It read it as "the AI capex curve is real, and the company at the top of the leaderboard just confirmed it by paying for the right to keep building."

What The Capex Curve Actually Looks Like

Pull the numbers in order and the pattern is brutal.

In 2023 Alphabet spent $32.3 billion on capex. In 2024 it spent $52.5 billion. In 2025 it spent $91.4 billion — a 74 percent year-over-year jump. Sundar Pichai is now guiding to $180 to $190 billion in 2026. He has already told the Street 2027 will be "significantly higher." The number triples in three years. The number doubles again in one year. There is no historical precedent for a single mature company increasing capex on this slope without either issuing stock, issuing debt, or both. Alphabet has now done both. The 100-year sterling bond it sold in February for £1 billion — one of only a handful of century bonds ever issued in that market — was the first signal that the financing math was changing. This week's equity raise was the second signal, and it is louder.

Look across the street and the picture is the same shape. Microsoft is at $190 billion in 2026 capex. Amazon is at $200 billion. Meta is at $135 to $145 billion. The Big-4 total is $725 billion in 2026, up from $410 billion in 2025, up from $230 billion in 2024. Roughly 75 percent of that is AI infrastructure — GPUs, TPUs, data centers, substations, fiber. Add Oracle, neoclouds like CoreWeave, Apple's quiet build, and sovereign AI programs, and 2026 becomes the first trillion-dollar year of compute capex in human history. For comparison, the entire global telecom capex peak during the dot-com bubble was around $325 billion in 2000. AI capex in 2026 is more than three times that, in a single year, concentrated in roughly ten companies.

The aggregate revenue from this capex is not yet visible at the same scale. Alphabet's cloud business is growing fast, but it has not yet doubled to match the doubling of capex. Microsoft's Azure AI revenue is real but is one line item inside a much larger business. Meta is monetizing AI through better ads. Amazon is selling AWS capacity. Across the Big-4, total AI-attributable revenue in 2025 was perhaps $150 to $200 billion. The capex-to-revenue gap is the widest it has been in any technology cycle. Goldman Sachs flagged a 46 percent growth gap between capex and revenue right now — wider than the 32 percent gap that opened up during the 2001 telecom bust.

That comparison is not meant to be a bubble call. It is meant to be precise about the math. Telecom companies in 2000 borrowed money to build fiber that no one used. The AI hyperscalers in 2026 are deploying capital — increasingly through new equity, not just debt — to build compute that does have customers, but whose long-run unit economics are still being discovered. The 100-year bond and the $84.75 billion equity raise are the financial market's way of saying: keep building, but the price is rising, and you will share more of the upside with us.

Why Berkshire Wrote The Anchor Check

The Berkshire piece deserves its own paragraph because of what it signals about how the smartest concentrated investor in America is reading this cycle.

Berkshire had already built an Alphabet position in early 2026 — Greg Abel added 36.4 million shares of GOOGL in his first three months as CEO. By late May, the Alphabet stake was already 28 percent of Berkshire's equity portfolio when combined with Apple. The $10 billion private placement on top of that does two things simultaneously. It deepens the position by a third in one transaction. And it does so by writing a check directly to Alphabet rather than buying on the open market — meaning Berkshire is putting capital onto Alphabet's balance sheet for AI capex, not just owning more secondary shares.

Read what that is actually saying. Berkshire under Abel just decided the highest-return use of $10 billion sitting in T-bills earning 4.5 percent was to fund Google's data center buildout at a 6 percent discount to current market. The implicit return assumption is that the AI infrastructure Google deploys with that money will earn more than the 4.5 percent risk-free rate by enough margin to justify the equity risk. Berkshire is not famous for being early on technology themes. It is famous for being right after the dust settles. If Abel is writing his first AI check now, the dust is settling.

The framing matters because Berkshire's job is to allocate capital where it earns the highest risk-adjusted return. That used to mean See's Candies and Geico. It now means an anchor stake in the AI capex cycle. The free market did not need a White House meeting or a congressional hearing to figure out where the next decade of returns lives. It needed Greg Abel to look at a spreadsheet and decide $10 billion belongs in Google's AI compute build rather than in short-term treasuries. He did. The market priced the rest.

The Free Market vs. The Central Planner

This is also a story about which system can actually fund this kind of capex without breaking.

In the United States, Alphabet announced an $80 billion raise on Monday afternoon, the book was oversubscribed by Tuesday morning, and the deal priced at $84.75 billion by Tuesday night. Berkshire put up $10 billion of it. Underwriters distributed the rest into institutional accounts globally. The whole thing took about thirty hours. No regulator approved the price. No ministry directed the allocation. No state-owned bank was leaned on to participate. Capital found the deal because the deal offered an expected return that beat the alternatives, and the market processed that judgment at the speed of an electronic order book.

Now compare. The CCP's national AI champions — Huawei, SMIC, Cambricon, the rest — fund their compute build through a combination of state subsidies, directed lending from policy banks, and forced equity participation by state-owned enterprises. The "Big Fund" alone has committed hundreds of billions of yuan to semiconductor and AI infrastructure since 2014. None of that capital came from a market test. It came from a planner deciding which technology should win and ordering the banks to fund it. Capacity gets built. Whether it is the right capacity, at the right scale, at the right cost — that question gets answered years later, often badly, because no one along the chain had to pay a market-clearing price to be there. The CCP can throw a lot of money at AI. It cannot throw market-tested money at AI.

The contrast was visible in real time this week. American capital priced Google's build at $84.75 billion in thirty hours. The CCP cannot do that. It can only declare a five-year plan and command its banks to participate. The first system selects winners by reading the prices the second system has to manufacture from the top. That is the long-run reason American AI infrastructure is going to be more efficient than Chinese AI infrastructure even if the dollar totals look similar — because each dollar that goes into the American build had to clear a hurdle rate set by someone risking their own capital, and each yuan that goes into the Chinese build cleared a hurdle rate set by a planner who is not.

The Trump administration's posture through all of this has been to keep export controls tight on the high-end accelerators and to keep capital markets open for the American buildout. The fingerprints of that policy mix were on yesterday's tape. The American hyperscalers can raise nine-figure capital in a day. They have access to TSMC's 3nm output. They have a customer base that pays for the inference. The CCP got a worse map again. Not because anyone in Washington made an announcement this week — but because the free market did its job and the planner cannot match the speed of it.

What This Means For Your Money

If you own GOOGL, the dilution is real but small. $84.75 billion of new shares is about 1.6 percent of the $5.3 trillion market cap before the raise. The earnings-per-share drag in 2026 is maybe 1.5 to 2 percent. The trade is whether the capital deployed earns more than that. If Cloud keeps growing at 30 percent, if Gemini's 900 million monthly users keep compounding, if the $155 billion contracted backlog converts at decent margins — the math works easily. If demand softens in late 2027 and the capex has to keep rising anyway because the build is committed — the math gets ugly. The honest read is the trade is no longer a "wonderful business at a fair price." It is now a leveraged bet on the AI capex cycle.

If you own NVDA or AVGO, this week was confirmation, not news. Every dollar Alphabet raised this week ends up flowing toward GPU and ASIC purchases over the next twelve to twenty-four months. Broadcom's selloff on Tuesday despite 143 percent AI revenue growth was about guidance, not demand — there is no demand problem. The demand problem is the other direction. There is more demand than the silicon supply chain can fulfill, and the financing to meet that demand is now coming straight from public equity markets. That is the bullish underpin for the chip stack.

If you own BRK, the news is that your manager's first big move post-Buffett was to put $10 billion into an AI infrastructure trade at a discount to market. Whether that is the right move depends on how the next three years go, but it is a meaningful signal about how Greg Abel sees the world. The Berkshire that used to avoid technology now has 28 percent of its portfolio in two AI names. The Apple-and-Alphabet concentration is not an accident. It is a thesis.

If you do not own any of this, the practical implication is that the cost of capital for AI infrastructure just got priced — and the price is high enough that even Google had to bid for it. Returns on capital are going to be wider than people expect for the names that placed bets early, and narrower than people expect for marginal entrants who arrive late with debt-funded data center plans. The capex cycle's winners get separated from its losers somewhere in 2027 or 2028. The dispersion is going to be ugly.

The Read

The right way to read this week is that the AI capex cycle just crossed a threshold no software company has crossed before. Google does not sell stock. Until this week. The most cash-rich, highest-margin software business in history needed to issue $84.75 billion of equity, at a discount, with Berkshire as the anchor, to keep up with the bill for the next eighteen months of compute. That is not a sign of weakness. It is a sign that the build is bigger than anything the equity culture of Silicon Valley has had to fund before.

The deeper read is about which system can fund this kind of build well. The United States can. The mechanism is boring: a press release on Monday, a book on Tuesday, $84.75 billion priced by Tuesday night, Berkshire writing $10 billion of it because Greg Abel decided the return beat T-bills. No regulator was consulted. No central planner directed the allocation. The price discovered itself. The CCP cannot do this. It can only command its banks to fund whatever Beijing has decided is strategically important, at whatever price the plan dictates. The first system finds winners by listening. The second system manufactures winners by ordering. Over a decade, the difference compounds into the gap between American AI and CCP AI.

For twenty-one years Google funded itself. This week it asked the capital markets for money and the capital markets said yes — at a 6 percent discount, with Berkshire's $10 billion check on the front of the book, and a 3.9 percent dilution haircut on the stock. None of that means Google is breaking. All of it means the AI buildout is bigger than any single balance sheet can carry on its own, and the free market is doing its job by routing capital into it from every direction that has capital to route. The cash cow just sold stock. That is not the end of the cash cow. It is the start of the next cycle, where the question is not which company has the best balance sheet, but which company can deploy outside capital fastest into compute that customers will pay for. Google just bought itself eighteen more months at the front of that line. The rest of the field has to find its own $84.75 billion. ~ Framework

Market Truths · 財經真言 · Published Tuesday, Thursday, Saturday · markettruthspod.com