After missing out for twenty years, Berkshire finally takes action, betting on Google to become the “BNSF Railway” of the AI era.
```
After witnessing the excellence of Google’s business model with his own eyes, Warren Buffett’s Berkshire Hathaway waited for a full twenty years. Now, his successor is replicating Buffett’s classic capital allocation strategy with a $10 billion investment—only this time, it is Berkshire itself playing the role of “See’s Candies”, and Google Cloud might be becoming the BNSF Railway of the AI era.
According to Bloomberg, Google’s parent company Alphabet is raising $80 billion through a bundle of equity financing options, including a $10 billion directed share issuance deal with Berkshire Hathaway, a $30 billion underwritten offering, and a $40 billion ongoing at-the-market issuance plan. This is one of the largest equity transactions in history, and Berkshire’s participation has surprised the market.
Ben Thompson, founder of tech strategy analysis firm Stratechery, believes the core logic of this investment is: Google is not only one of the best business models today, but its cloud business could become a new engine of absolute scale far surpassing advertising, as AI demand explodes.
Buffett’s successor and new Berkshire CEO Greg Abel’s timing on this move is quite meaningful.
Berkshire currently holds $373 billion in cash and will have $25 billion in free cash flow in 2025. In an environment where the AI infrastructure race is in full swing and compute demand consistently exceeds market expectations, Google—thanks to the cost advantages of its self-developed TPU chips, as well as multi-layered bets on models, services, and cloud capacity sales—is seen as one of the few options able to absorb Berkshire’s massive funds while generating high returns.
Buffett’s Regret: The Business “Never Seen Before”
Google is the one business Buffett has publicly admitted missing out on.
At Berkshire Hathaway’s 2017 annual meeting, he said that Berkshire-owned GEICO was an early Google ad customer, paying $10–11 per click even then. "Anytime you’re paying someone $10 to $11 per click where their cost is near zero, you know that’s a good business—unless someone takes it away from you." Buffett said, "You’ve almost never seen a business like that."
Yet, it was this very “asset-light, high-margin” model that made Buffett hesitate.
Ben Thompson’s analysis on Stratechery notes that Google’s role as an aggregator means it maximizes absolute value by sacrificing relative value—content creators (supply), advertisers (demand), and users all provide value in this system, but the per-unit relative return is diluted. This does not naturally fit Buffett’s preferred framework of “buying quality companies at reasonable prices”—he’s more familiar with asset-heavy companies that rely on tangible asset accumulation and have stable, predictable cash flows.
See’s Candies and BNSF: The Two Faces of Capital Compounding
To understand this investment, you must first understand Berkshire’s capital management logic.
In 1972, Berkshire acquired See’s Candies for $25 million; at the time, See’s had $30 million in sales, under $5 million in pre-tax profits, and needed just $8 million in working capital. By 2007, See’s had $383 million in annual sales and $82 million pre-tax profits, with only $32 million in cumulative additional investment, but with $1.35 billion in cumulative pre-tax profit. This is what Buffett meant by “time is the friend of the wonderful business.”
Berkshire took the steady cash from See’s and allocated it to a completely different kind of business—BNSF Railway.
Railroads are classic heavy-asset businesses; BNSF’s capital spending last year reached $3.8 billion, but its absolute returns are also impressive: $23.4 billion in revenue and $5.5 billion in net profit. Ben Thompson points out that all profit See’s has generated for Berkshire (last disclosed as “over $2 billion” in 2019) is possibly less than BNSF’s profit in a single year. Both are the same essential logic: use cash from a “cash machine” to leverage into a much larger asset-heavy business.
Google Cloud: Lower Margins, Bigger Potential
Ben Thompson cites data showing this dynamic is being replayed inside Google.
In Q4 2019, Google services (core ads, etc.) had $43.2 billion revenue and $13.5 billion operating profit; Google Cloud had only $2.6 billion in revenue and lost $1.2 billion. By Q1 2026, Google services grew to $89.6 billion with $40.6 billion operating profit; Google Cloud had $20 billion revenue and $6.6 billion operating profit with a 33% profit margin, while Google services had 45% margins.
Over seven years, Google services more than doubled revenue and nearly tripled profit; Cloud’s share of revenue rose from 6% of services to 22%, and its profit share from near zero to 16%, with much faster growth and margin expansion than the core business.
The more important question: ads hit a ceiling due to their share of the overall economy, but cloud & AI’s potential market in some circumstances could cover the entire economy. As Ben Thompson writes, Google services today perhaps exist to fuel a future business of even greater scale, slightly lower margins but higher absolute profit.
Why Now for Berkshire, Why Equity Issuance for Google
This deal raises two interesting questions. First: why is Google issuing equity, not debt?
Financially, debt is usually cheaper—interest is tax-deductible and doesn’t dilute shareholders. Google currently has $81 billion in debt but $126 billion in cash, so theoretically has plenty of capacity for more borrowing.
Ben Thompson’s view: Google is likely to issue a lot of debt next. The signal in this equity deal is—the scale of compute demand is still systemically underestimated, and Google is willing to use every financing tool to meet supply.
Of course, another reading is that Google is unsure about big capital spending returns and wants to spread risk. If no major debt issuance follows, this view gets more credible.
Second: what is Berkshire’s logic?
Ben Thompson says Greg Abel is basically replaying the Buffett playbook: Berkshire as See’s Candies, Google as BNSF.
Given $373 billion cash and $25 billion of annual free cash flow, it’s rare to find targets that can absorb this much while maintaining high returns. Google’s advantage is having multiple real options: its ad business benefits directly from AI investments, Gemini is competitive at the model layer, and with the cost benefits of TPUs, Google Cloud can maintain above-normal profits even as compute becomes a commodity, and sell capacity to outside customers.
Cash Is King: The Underlying Logic of the AI Race
Ben Thompson previously wrote on Stratechery: The key to the AI race isn’t just who acquires the most compute earliest, but who can sustain cash flow to keep buying compute.
He uses Anthropic as an example: although OpenAI claimed to lead in compute resources, Anthropic’s steady commercial income and fundraising enabled it to sign a big compute purchase with SpaceX, showing that strong cash flow lets you keep buying compute in the market.
This logic extends to the ultimate competition: As compute supply tightens, companies with the greatest cash generation will secure the most lasting compute advantage, forming a flywheel of dominance. In Ben Thompson’s view, right now, the company best situated for this is Google. Berkshire has given its answer with $10 billion.
The original column follows:
Google Capital Company
What is the most beautiful business model in history?
First, imagine your supply is free. Second, your customers volunteer to outbid each other for higher prices. Third, your users decide which customer gets the privilege of paying you. All you have to do is build a bit of infrastructure to make it happen, pay a minimal depreciation on that infrastructure, and rake in billions at some of the highest profit margins in business history.
Of course, I’m describing Google—a company so good that legendary investor Warren Buffett never managed to convince himself to invest in it. As he explained at Berkshire Hathaway’s 2017 AGM:
“For example, through GEICO, we were an early customer, and we saw—these numbers are way out of date—but I remember we were paying them about $10 or $11 a click. Anytime someone clicks on something where you have basically no cost, and you collect $10 or $11, you know it’s a good business, unless someone takes it from you. So we saw the power up close... But you know, you almost never see a business like that.”
One feature of “aggregators” like Google is how they maximize absolute value by sacrificing relative value. For the supply side—web content—Google dramatically increases audience even though any visitor from Google is worth less than a direct visitor; for advertisers, a click makes up for thousands of worthless ad impressions; for users, Google helps them find what they want among an overwhelming glut of free content. In every case, the aggregator trades off relative quality for quantity, confident that absolute quality grows more over time.
Ironically, from an investor’s perspective, this is exactly the opposite of why these companies are valued so highly. The best tech companies are “asset-light”, optimized for near-zero marginal cost. Yes, they spend heavily on R&D and infrastructure, but don’t actually participate in those markets; simply taking a cut—keeping most of it—excites Wall Street. In market terms, relative profit margins usually matter more than the absolute dollar sum earned.
Berkshire Hathaway & Productive Capital
Before Buffett, Berkshire Hathaway was a failed textile firm; he first invested because its shares traded below liquidation value, ultimately taking it over after a dispute with management. He later regretted it. As the 1989 letter to shareholders said:
“If you buy shares cheaply enough, business ups and downs usually give you a chance to sell at a profit, even if the long-term future is poor... Time is the friend of the wonderful business, the enemy of the mediocre one...
I could give you more examples of my ‘buying cheap’ mistakes, but you get the point: It’s far better to buy a wonderful company at a fair price than a fair company at a wonderful price. Charlie figured this out long ago; I learned slowly. Now when buying companies or stocks, we look for first-class businesses and first-class management.”
One first-class acquisition was See’s Candies, bought in 1972. As Buffett wrote in the 2007 letter:
“We paid $25 million for See’s, when it did $30 million of sales and under $5 million in pre-tax profit. The business capital required was $8 million (plus a few months’ seasonal debt each year). The pre-tax return on invested capital was 60%...
Last year, See’s did $383 million in sales, $82 million pre-tax profits. It takes $40 million in capital to run it now. Since 1972, we’ve needed to invest only $32 million more to handle modest real growth—and tepid financial growth. Meanwhile, pre-tax profits have totaled $1.35 billion. Apart from the $32 million, all that money flowed to Berkshire (or in early years, to Blue Chip Stamps).”
The "problem" with See’s was, there was nowhere to put all this profit; a private owner would have too much idle cash, a public company would struggle to return it via buybacks & dividends. Berkshire used this cash for growth:
“After paying corporate tax, we use what’s left to buy other attractive businesses. Just as Adam and Eve started the chain of 6 billion humans, See’s begat multiple new cash flows for us. (‘Be fruitful and multiply’ is a commandment we take seriously at Berkshire.)”
One business funded by See’s profits was at the opposite end of capital intensity: BNSF Railway. Railroads require massive capital; BNSF spent $3.8 billion last year, but also made a ton: $5.5 billion profit on $23.4 billion revenue. Berkshire’s total profits from See’s may be under $3 billion (latest disclosure: "over $2 billion" in 2019)—so See’s decades of profits are less than BNSF earns in one year.
So which business is better?
The Scope for Google Cloud
In Q4 2019—when Alphabet first disclosed cloud revenue—Google Services (the super-high-margin business) had $43.2 billion revenue, $13.5 billion operating profit; Google Cloud had $2.6 billion revenue and lost $1.2 billion, just 6% the size.
In Q1 2023, Google Cloud was first profitable: $7.5 billion revenue, $200 million profit; Cloud’s income was 12% that of services, but only 1% as much profit.
By Q1 2026: Google Services $89.6 billion revenue, $40.6 billion profit. Google Cloud $20 billion revenue, $6.6 billion profit. Cloud is 22% of services’ revenue, 16% of profits.
Google Services is more scalable than See’s Candies—doubling revenue, tripling profit in seven years is amazing. But Cloud is growing even faster—its 33% margin trails Services’ 45%, but its expansion is more rapid.
The big question is how far this can go. Ads are very profitable but, by definition, only a small slice of the economy. Meanwhile, Google Cloud’s growth is powered by AI—something many think or fear could reshape the whole economy. Will we someday look back and see that Google Services only existed to fund a lower-margin, but higher-dollar, business—just as See’s funded BNSF?
Berkshire Hathaway & Google Equity
The backdrop is this Bloomberg story:
“Google parent Alphabet Inc. is raising $80 billion through a package of equity offerings, including a deal with Berkshire Hathaway, as it races to fund ambitious AI spending plans. According to a Monday statement, the move includes a $40 billion ‘at-the-market’ (ATM) plan, occasionally selling stock from Q3 onwards. The company will also offer $30 billion in underwritten common and mandatory convertible preferreds and a $10 billion deal with Berkshire. These combine for one of the largest equity deals ever—bringing a surprise twist to this blockbuster IPO year.”
A sizeable chunk of the ATM plan, starting in the fall, will likely go to pay the tax bill on Google’s stock compensation (which is huge due to soaring prices).
That leaves the equity currently being issued—especially the Berkshire $10 billion—which is intriguing for several reasons. The first question is: why issue stock, not debt? In general, debt is preferred: its costs can be covered from income, and current equity holders keep all the upside. Equity reduces risk but gives up some future profit.
So far, Google has funded its massive AI capital spend with free cash flow. It has around $81 billion in debt—balanced by $126 billion in cash. In short, its capacity to issue more debt—and the resulting tax savings—is huge.
This points to an Occam’s razor explanation: Google will also issue much more debt soon; that is, everyone is still underestimating demand for compute. Of course, there’s a bearish read: Google isn’t sure these investments will pay off, and prefers to share risk (and reward). If no big bonds follow, this could be correct.
The second question: why does Berkshire care now, at a price just off all-time highs? Is it just because Buffett’s no longer making the calls, and his CEO successor, Greg Abel, is? In fact, Abel may just be reenacting Buffett’s strategy, except this time Berkshire is See’s, and Google is BNSF. At quarter-end, Berkshire had $373 billion in cash, and will generate $25 billion in free cash flow in 2025. How many companies can deploy such sums at a high rate of return?
It’s hard to think of a better answer than Google. The company isn’t just betting on AI, but has optionality: its Services business benefits directly, Gemini is competitive at the model layer, and it can sell compute supply to frontier labs. Thanks to the TPU, its compute is cheaper, so in a world where compute is commoditized—however hard to imagine now—Google is best-placed to be the most profitable hyperscale cloud provider.
Notably, $10 billion is small for both companies. Its main use may be as a signaling mechanism. For Google, it marks that expected demand is far above anyone’s imagination, and the company is ready to use all available means (including equity) for supply; Berkshire’s investment is an endorsement. Conversely, if the signal is right, Berkshire will have made a good bet, putting its cash machine to work building the future.
Cash: The Ultimate Commodity
A few months ago, when Anthropic was clearly on the rise, OpenAI supporters tried to argue OpenAI was in fact dominant, since it had locked up more compute; as I wrote in “Mythos, Muse, and the Opportunity Cost of Compute”, this is hardly decisive:
“OpenAI bet that compute constraints—and their deals to overcome them—would matter more than Anthropic’s momentum among users... I’m not sure that’s decisive. With AI, distribution and transaction costs are still free—the two prerequisites for an aggregator—so the winner should be whoever has the best products. These products attract the most users, producing the cash to source more compute; considering Anthropic’s deal to lock up a big chunk of TPU supply, given TSMC’s capacity limits, this was taking Google’s supply. I suspect Anthropic could take more, including the hyperscaler and neocloud capacities already built. Yes, the compute will be more expensive, but if demand is high enough, the cash flow will be there.”
Just weeks later, when SpaceX supplied Anthropic (at a high price), this was confirmed:
“This is a terrific illustration of basic economics. If demand outpaces supply, prices should rise. Simultaneously, prices are elastic: lower prices mean more demand; higher prices mean less. Here, demand for compute is broad, but Anthropic probably wants it most badly—and they’re willing to pay, not just because they’ve got serious revenue but because they can raise the funds to win in AI.”
The analysis assumes plentiful compute supply globally; but what if, someday, there isn’t? What if the ultimate battle—the war for compute—becomes a matter of who can commit the most cash? What if this advantage feeds on itself so that the company best at generating cash ends up with most compute (and can sell the rest), building an even bigger cash engine? In that world, where would you bet?
We now know where Berkshire is making its bet.
Risk Disclosure and DisclaimerMarkets are risky; investments need caution. Nothing in this article constitutes personal investment advice, nor does it take into account any individual’s specific investment objectives, financial situation, or needs. Readers should assess whether the opinions, views, or conclusions in this article apply to their own circumstances. Invest accordingly and at your own risk. ```