Search Engine to Steam Engine

Alphabet’s AI spending spree is weakening the economics of search and cloud, while Tencent’s more disciplined, commodity-oriented approach may create better long-term shareholder returns.

• Alphabet expects capital expenditure to rise from $90bn in 2025 to around $200bn in 2026, with capex increasing from about 10% to almost 50% of sales; Tencent spent RMB79bn in 2025 on RMB752bn of revenue, roughly 10% of sales.
• AI summaries reduce Google search users’ click-through rate on traditional links from 15% to 8%, forcing Google to increase the number and price of sponsored links while network effects weaken and computing becomes more centralised.
• Frontier labs have encouraged hyperscaler investment through large compute contracts: Oracle’s backlog reached $664bn, nearly ten times its annual revenue, while Google Cloud’s backlog reached $514bn; Hosking warns that these backlogs depend on the solvency of counterparties such as OpenAI and Anthropic.
• China’s export restrictions on Nvidia GPUs, whose share of the Chinese market fell from 95% to zero according to Nvidia’s CEO in October 2025, pushed Chinese firms to develop near-frontier models with much leaner capital budgets.
• Tencent integrated DeepSeek into Weixin when its own model lagged in early 2025, later rebuilt its own model, increased marketing revenue by more than 20% year on year and lifted operating margins excluding new AI products above 40%.

Takeaway: Investors should favour pockets of rational capital allocation rather than predict the AI winner; Hosking is reducing Alphabet because rising capex and competition threaten incremental returns, while continuing to own and add to Tencent for its business-model consistency and lower capital intensity.