AI Capex Cycle Enters Payoff Phase as Alphabet Lifts Spend to $205 Billion, Cloud Backlog Swells to $514 Billion

The dominant macro-equity arc of the week is the maturation of the AI capital expenditure cycle from speculative build to contracted demand. Alphabet management's decision to raise full-year AI infrastructure budgets to roughly $205 billion was published in the same quarter Google Cloud disclosed a $514 billion remaining performance obligation backlog — a ratio that materially reframes the bear case. Where 2025 debate centered on depreciation drag and free-cash-flow compression, the backlog figure implies multi-year contracted revenue coverage against the spend, and desk commentary this session leaned heavily toward the backlog as the more informative disclosure. The read-through is broadening across the supply chain. SK Hynix shares posted a monster week on tightening high-bandwidth memory supply and firming DRAM pricing, while Applied Materials printed records on nearly every line item and guided higher — yet trades more than 25% below its high, a valuation gap that implies the market is still pricing the cycle as peaking rather than extending. Coverage of Nvidia's expanding role as de facto financier to the AI ecosystem — the 'Bank of Nvidia' framing — introduces a second-order concern: vendor financing and circular capital flows can flatter near-term order books even as underlying end demand remains genuinely robust. Peter Thiel's disclosed $118 million position in a surging large-cap technology name adds a notable smart-money signal to the tape. Strategists remain constructive but selective. Morgan Stanley's positioning framework favors quality balance sheets, AI adopters rather than pure AI suppliers, and financials as earnings growth broadens beyond the megacap complex. Peripheral beneficiaries are being repriced aggressively: TTM Technologies, a historically low-multiple PCB manufacturer, is now being underwritten as an AI growth asset, while Bloom Energy's 150% year-to-date advance reflects the power-constraint bottleneck emerging as the binding limit on datacenter buildout. The risk to the thesis is not demand but reflexivity — capex guidance increases now require backlog validation to sustain multiples.