Zacks Optimistic About AI Cloud Giants' Valuation Recovery, Capital Expenditure Is Not a Cost, Idleness Is the Cost
Zacks strategist Ethan Feller believes that after several months of no significant increase in stock prices, AI hyperscale cloud vendors have seen a resonance of revenue, profit growth, and valuation, which may attract capital inflows again.
Feller identifies Amazon, Alphabet, and Meta as core observation targets, stating that the forward P/E ratio for this sector is currently in the range of a little over ten to twenty times, while sales and profits continue to grow at a compound rate; his judgment is that valuations have not fully reflected profit improvements.
Amazon's Q2 revenue grew 20% year-on-year to $200.6 billion, and operating profit increased 43% to $27.5 billion; AWS revenue grew 37% year-on-year to $42.2 billion, marking the fastest growth rate in 18 quarters, with AWS operating profit rising 64% to $16.6 billion.
AWS operating profit margin rose from 32.9% in the same period last year to 39.4%, with an annualized revenue run rate of approximately $169 billion. Amazon also disclosed that its AI and self-developed chip businesses each have annualized revenues exceeding $25 billion, both maintaining triple-digit growth rates.
Feller believes that the accelerated growth of AWS, margin expansion, and the independent revenue scale of AI and chip businesses provide direct evidence of the returns on capital expenditure for hyperscale cloud vendors; however, Amazon's free cash flow has turned negative due to investments in data centers and AI infrastructure, with ROI verification still primarily coming from the operational level of cloud business.
In market mechanics, the buyers of hyperscale cloud vendors are developers and enterprise customers needing GPUs, model inference, storage, networking, and enterprise-level AI services, while the sellers are companies like Amazon, Alphabet, Microsoft, and Meta that control cloud, models, distribution, and data center assets. If AI revenue and cloud profit margins continue to expand, funds that previously avoided this sector due to massive capital expenditures may shift to pursue profit upgrades; if enterprise AI demand is insufficient to cover depreciation, electricity, and computing expansion costs, cash flow pressure will first suppress valuations.
Source: Public Information
ABAB AI Insight
The previous narrative of AI capital expenditure by hyperscale cloud vendors has long been viewed by the market as "invest first, validate later." Amazon, Microsoft, Alphabet, and Meta are continuously expanding data centers, GPU, networking, and self-developed chip procurement from 2023 to 2025, with the market initially focusing on depreciation, free cash flow, and payback periods; recent performance from Amazon has shifted the discussion to specific recovery mechanisms: analysts estimate the payback period for servers and network assets to be less than three years, while the lifespan of data center buildings can reach over 30 years, allowing for reuse across multiple generations of servers.
The capital path is shifting from benefiting single-point chip suppliers to secondary monetization of cloud platforms. Nvidia confirms revenue through GPU sales; AWS packages GPUs, Trainium, Graviton, networking, and storage as on-demand cloud services, converting model training and inference demand into long-term enterprise consumption. Amazon disclosed that its AI and self-developed chip businesses each have annualized revenues exceeding $25 billion, indicating that capital expenditure is not only being absorbed as depreciation but is also beginning to generate measurable business line revenue.
A historical analogy is the replacement of in-house data centers by cloud computing in the 2000s and the revaluation of advertising and subscription revenues driven by mobile internet in the 2010s. AWS initially had to bear the upfront investments in servers, networking, and data centers, subsequently increasing profit margins through scale utilization and shared infrastructure; the current difference with generative AI compared to past cloud migrations is that the unit cost of computing is higher, and power and chip supply are more constrained, but the business logic remains the same: fixed costs are initially borne by infrastructure providers, and then enterprise customers shift their IT budgets to usage-based services.
Essentially, this represents a transfer of pricing power. When enterprises no longer directly purchase servers but continuously pay for GPU usage, model calls, hosted inference, and data services, AI costs transform from one-time capital expenditures to operational expenses that cloud vendors can price. Cloud giants possess regional data centers, customer relationships, self-developed chips, model products, and distribution channels, allowing them to maintain high unit economics during computing shortages; however, only when utilization, pricing, and enterprise renewals grow in sync can massive data center investments truly convert into free cash flow for shareholders.
ABAB News · Law of Cognition
Capital expenditure is not a cost; idleness is the cost.
Chips are sold once, but cloud services are charged many times.
Valuation re-evaluation begins when returns are quantified.