Hook
$180 to $190 billion in capital expenditure by 2026. That’s the number Alphabet is now betting the farm on. For context, that’s roughly the entire market cap of most Fortune 500 companies being sunk into data centers and AI chips. The market is waiting for Q2 earnings with one question: is this a land grab or a burn pit?
Context
Alphabet reports its Q2 2025 earnings next week. The narrative has shifted from "AI growth story" to "profit conversion efficiency." Google Cloud revenue surged 63% year-over-year last quarter, with a $460 billion backlog in cloud contracts. But the cost of that growth is staggering. The company already broke its tradition of self-financing by issuing new shares to fund capital expenditure. Wall Street is rotating: some institutional money has moved from Meta to Google, betting that Google’s infrastructure-heavy approach (own chips, own data centers) will yield a more durable moat than Meta’s social-dependent AI layer. But the bears point to Gemini’s repeated delays and the looming risk that AI search overviews cannibalize traditional search ad revenue.
Core: The Order-Flow Analysis
Let me walk through the numbers like I would a DeFi liquidity pool audit.
First, the CapEx. $190 billion is not a typo. That’s a 10x increase from 2022 levels. Alphabet is essentially pre-paying for a decade of compute. But here’s the catch: the majority of that spend is on TPU—Google’s own AI chips. They are now selling TPUs externally. That’s a direct challenge to NVIDIA’s CUDA monopoly. My 2017 ICO audit experience told me to never trust a narrative without code verification. Here, the code is the hardware ecosystem. TPU has zero developer mindshare compared to CUDA. Google needs to build software stack compatibility fast, or this $190B becomes a stranded asset.
Second, the cloud backlog of $460 billion. That sounds like a fortress, but lock-in is only sticky if the switching cost is high. AWS and Azure are not going to let Google steal share easily. The cloud margins are “nearly doubled” according to management, but absolute margin is still below 10%—compared to AWS’s 30%+. This is a capital-intensive growth phase. The unit economics depend on whether those backlog contracts have positive net present value after accounting for hardware depreciation.
Third, the search ad risk. AI overviews are already live in beta. If users stop clicking on ads because the answer is in the snippet, ad revenue drops. Google’s entire machine is built on the attention-auction model. A reduction in click-through rates will cascade into lower quality scores for advertisers, raising their cost per acquisition. That’s a negative flywheel. I modeled this in my 2020 arbitrage scripts: any friction in the matching engine reduces total surplus. The market is pricing Alphabet as if the ad business is immortal. It’s not.
Fourth, the capital allocation signal. Issuing new shares to fund CapEx is a red flag. In crypto, we call that “dilution without lockup.” It signals that internal cash flow cannot keep pace with spending. This is the equivalent of a DeFi protocol printing tokens to pay for yield farming subsidies. It works in a bull market. It kills you in a bear market.
Contrarian: The Silent Value in the Friction
Now the counter-intuitive angle that most retail analysts miss. The market is obsessed with the short-term profit conversion of AI. But the real alpha is in the friction—specifically, the friction between Google’s legacy ad monopoly and its nascent AI infrastructure monopoly. Here’s the contrarian thesis: Alphabet is building a new moat that is harder to copy than a software product.
Software moats (like social networks) are subject to herd behavior. Hardware moats (like ASICs and data center interconnects) take years to replicate. If TPU achieves even 10% of NVIDIA’s ecosystem within three years, Google becomes the default provider for cost-sensitive AI inference workloads. The $460 billion cloud backlog is not just for virtual machines—it includes AI model training contracts that lock customers into Google’s infrastructure. That’s a multi-year lock-in with high switching costs because migrating petabytes of training data is painful.
The market is undercounting the option value of that infrastructure. They see the CapEx as a cost. I see it as a call option on the next wave of enterprise AI adoption. The bear case is that Gemini fails and the TPU ecosystem never takes off. But the bull case is that Google replicates what AWS did: turn a cost center (infrastructure) into a profit center. AWS started as internal capacity. Google is doing the same with AI compute.
Also overlooked: Alphabet’s balance sheet can sustain this without bankruptcy. Even with $190 billion in CapEx, free cash flow from the ad business remains positive. This is not a gamble; it’s a leveraged bet. Levered bets can blow up, but the upside asymmetry is significant if the bet works.
Takeaway
Watch for three signals in the Q2 call: 1. Cloud revenue growth rate—if it drops below 50%, the growth narrative is breaking. 2. Cloud margin percentage—if it crosses 12%, the scale is kicking in. 3. Any mention of TPU customer wins outside of Google—if they name one Fortune 100, the hardware flywheel starts.
My bias? I lean bearish in the short term because the market is pricing in perfection on ad revenue stability. But I am long the infrastructure thesis. The yield is not the prize, the exit is. Right now, Alphabet is building the offramp to a new platform. Whether they reach it depends on execution, not marketing. Ledgers do not forgive, they only record. Q2 will add a few lines.
Alpha is found in the friction, not the flow. Due diligence is the only hedge you control. Profit is the receipt, not the purpose.