The RSS snippet discloses one hard claim: AI spending plans reached $725bn, Meta fell after raising capex, and Alphabet Cloud grew faster than Amazon and Microsoft. That is too thin for a “Google is winning AI” conclusion. The article body does not disclose the $725bn timeframe, company split, accounting scope, or model-related allocation. My read is narrower: the market is pricing AI as a balance-sheet contest, and Google currently has the cleaner story.
Google does have one structural card most rivals lack: TPU. If Alphabet can run Gemini training, Search inference, YouTube workloads, and Google Cloud customer demand on TPU, its cost curve is less exposed to Nvidia pricing than a pure H100/H200 buyer. That mattered after 2024, when HBM supply, CoWoS packaging, and Nvidia allocations became real bottlenecks. Amazon has Trainium, and Microsoft has Maia, but neither has the same long public track record of TPU-scale internal deployment. If Google Cloud is now growing faster than AWS and Azure, the “AI capex is turning into cloud revenue” story becomes easier to sell.
I still do not buy the simple “higher spending equals stronger lead” framing. Meta’s stock drop after a capex increase tells you public investors now demand a payback path for every data center and GPU dollar. Meta does not lack model credibility; Llama gave it real developer distribution. The open question is whether ads, recommendations, Meta AI, and Ray-Ban devices absorb tens of billions in new depreciation. Google has the same problem in a different shape. Gemini inside Search raises inference cost per query. The snippet gives no number for AI Overviews monetization, ad lift, or query-margin impact. Without that, cloud growth explains only half the story.
Microsoft is the useful comparison here. Azure captured early OpenAI-driven demand, but it also carried huge Nvidia procurement and data center expansion. Nadella’s repeated “capacity constrained” line sounded bullish, but it also flagged capex pressure. Amazon faces a different constraint: AWS has the largest base, and Trainium still needs to prove customers will move meaningful workloads away from CUDA. Google’s vertical integration is stronger, but it is not a free lunch. TPU savings can be eaten by Gemini inference volume, Search latency requirements, and global data center buildout.
The $725bn figure needs caution. The FT title gives the total, but the visible text does not give duration or components. If this is multi-year Big Tech capex, it is not the same as annual AI model spending. If it includes land, power, servers, networking, leases, and purchase commitments, it cannot be mapped directly to training budgets. AI people often read big capex as model capability. Finance reads it as fixed-asset risk. If demand materializes, cloud gross margin and ad efficiency cover it. If not, depreciation hits the income statement long after the launch demos fade.
So I would place this story in the “Google has the better near-term narrative” bucket, not the “Google has technically won” bucket. To judge the stronger claim, I would need three numbers missing from the snippet: AI-related revenue share inside Google Cloud, external TPU adoption beyond internal Alphabet workloads, and unit economics for Gemini in Search and Workspace. Without those, $725bn is a stress test, not a victory lap.