AI infrastructure capex has crossed from strategic investment into arms race. Amazon's FY 2025 10-K records $128.3 billion in capital expenditures — up 65% in a single year. Alphabet spent $91.4 billion, up 74%. Meta committed $115-135 billion for 2026 before a dollar has been spent. Reading all five filings together reveals three structural patterns investors are still underestimating: the spending is accelerating, not plateauing; Microsoft's reported margins already show the cost compression that peers only discuss in the future tense; and NVIDIA's own 10-K names the same data centers and power grids that hyperscalers are racing to build as the binding constraint on its revenue growth — completing a self-reinforcing cycle that no single filing captures alone.
NVIDIA generated $216 billion in revenue with 75% gross margins — the most dominant operating performance in tech history. But the FY2026 10-K reveals $165 billion in total obligations growing 2-5x faster than revenue, a $17 billion Groq license that created $14.4 billion in goodwill on zero revenue, and a $40 billion equity portfolio entangling NVIDIA with its own customer ecosystem. These aren't reasons to sell — they're the specific conditions investors should monitor at 25x forward earnings.
Arista Networks reported $9 billion in revenue growing 29% with 43% operating margins, zero debt, and a $10.7 billion cash fortress. But the FY2025 10-K reveals something the earnings call doesn't say: 42% of that revenue comes from two customers (Microsoft surging to 26%), $6.8 billion in non-cancellable purchase commitments just jumped 42% in a single quarter, and the headline cash flow quality flatters the income statement. ANET trades at nearly the same P/E as NVIDIA. The filing shows why that comparison deserves scrutiny.
AVGO's top 5 customers account for 40% of revenue. NVDA's largest customer is 22%. ANET depends on Microsoft (20%) and Meta (15%). Standard AI stock screens show none of this. Here's our 2-signal framework for stress-testing AI beneficiaries.
Morgan Stanley estimates 99% of software FCF goes to stock-based compensation. But the impact varies dramatically: Snowflake's SBC is 14x higher than Amazon's as a percentage of revenue. Our Filing Intelligence analysis reveals which companies offset dilution - and which leave shareholders holding the bill.
The Magnificent 7 make up 35% of the S&P 500. But comparing them all using P/E ratios is methodologically flawed—they span 5 distinct business models. Our sector-adjusted scorecard reveals GOOGL as best value (17.4x P/E), TSLA as most overvalued (85% optionality premium), and which stocks actually EARN their premiums.