Berkshire Hathaway just bought $10 billion worth of Alphabet stock. The market sold $45 billion in a single day. That 9% haircut erased more value than most companies will ever generate. Volatility is just liquidity leaving the room. But this isn't a crypto crash. This is Google.
For twenty years, Alphabet ran a cash machine. No debt. No preferred stock. Just search ads printing money. That habit died on July 22, 2026. The Q2 earnings release showed something unprecedented: free cash flow halved to $12 billion, while capital expenditure doubled to $36 billion. The company that never needed financing just issued $25 billion in corporate bonds. The market didn't celebrate.
Context: The 20-Year Habit Breaks
Alphabet's balance sheet was a fortress. No long-term debt until 2014, and even then only for tax optimization. The company generated enough cash to fund every moonshot—self-driving cars, balloon internet, life sciences—without tapping capital markets. That discipline was a signal: we don't need your money.
Now they do. The 2026 CapEx guidance: $190 billion for the year. That's more than Google's entire revenue in 2018. The breakdown reveals the obsession: data centers, TPU fabrication, and Nvidia GPU clusters. The Gemini 3.5 Pro delay—a model that was supposed to launch in April but slipped to Q3—triggered the first wave of doubt. When a product delays, the capital behind it doesn't. The spending continues. The depreciation clock starts ticking.
Core: The Unit Economics of a $190B Bet
Let me dissect the numbers because code doesn't lie. People do.
Revenue growth: 14% year-over-year. Cloud revenue: $20 billion, up 63%. That looks strong. But look at the cost structure. The $36 billion in quarterly CapEx will be capitalized and depreciated over five to six years. That means roughly $6-7 billion in quarterly depreciation charges starting 2027. The question: will AI revenue grow faster than those depreciation charges?
In my audit experience, I've seen projects raise $50 million on a whitepaper and burn through it in six months. Google raised $25 billion in debt. Same pattern, different scale. The unit economics are deteriorating. The cost to acquire a cloud customer is rising because the sales team has to explain why TPU is better than Nvidia. The customer lifetime value is uncertain because AI revenue is still experimental. The ratio is inverted.
Take the TPU strategy. Google's custom tensor processing units are cheaper and more efficient than Nvidia's H100. By some estimates, 40% cheaper per teraflop. Yet Nebius, a cloud provider, said "99% of our demand still points to Nvidia." Why? Because switching costs exist—developers trained on CUDA, models optimized for CUDA, frameworks built on CUDA. Google's TPU ecosystem is a ghost town. They spent billions on a chip that only their internal teams use.
The Gemini delay is a symptom, not the disease. The disease is that Google's AI model is no longer the default mental reference. When developers think "best LLM," they think OpenAI or Anthropic. When enterprises think "cloud AI," they think AWS or Azure. Google is fighting for third place in a market they invented.
And the financial strain is real. Free cash flow margin dropped from 30% to 15%. The $25 billion bond issuance came with a coupon that signals stress: 5.2% for a 10-year, when Apple pays 4.3%. The market is pricing in risk. Trust is a variable I refuse to define.
Contrarian: What the Bulls Got Right
It would be lazy to only list the risks. The contrarian case has technical merit.
Berkshire Hathaway didn't buy $10 billion for charity. Buffett's team spent months analyzing the AI infrastructure play. Their thesis: Google's data moat is irreplaceable. Every search, every YouTube watch, every Gmail interaction feeds the AI training pipeline. OpenAI scrapes the public web. Google owns the private data. That's a moat no competitor can cross.
Google Cloud's 63% growth is real. It's not just reselling Nvidia GPUs. The revenue mix includes Vertex AI, BigQuery, and security products. The customer base is diversifying: healthcare, finance, retail. The $20 billion quarterly run rate means Google Cloud is now larger than Oracle's cloud business. That's not hype. That's traction.
The AI revenue bubble concerns are overblown for one reason: Google has the distribution. They can embed Gemini into Search, YouTube, Workspace, Android. Two billion users see AI-generated summaries daily. Even if only 5% convert to paid cloud subscribers, that's 100 million enterprises. The funnel is unmatched.
Buffett's signal matters because he's a net buyer when everyone else is a seller. He bought during the 2008 crash. He bought during COVID. He buys when fear peaks. The $10 billion purchase says: this dip is overdone.
Takeaway: The Next Earnings Call Decides the Narrative
The clock is ticking. Every quarter that AI revenue grows slower than depreciation costs erodes the bull case. Every month Gemini 3.5 Pro stays in development gives OpenAI more time to embed their models into enterprise workflows.
But execution swings both ways. If Google delivers Gemini 3.5 Pro with benchmark-leading performance, and if Q3 cloud revenue shows accelerated AI adoption, the narrative flips overnight. The $190 billion CapEx becomes a moat, not a burden. The slump becomes a buying opportunity.
I've audited projects that collapsed under their own weight—overpromised, underdelivered, then tried to raise more. Google is not a startup. It has $100 billion in cash reserves. It has the talent, the data, and the infrastructure. But talent and data don't matter if the product misses the window. Code doesn't lie. The earnings call will.