AI must now deliver

AI has entered a delivery phase in which investment will be judged by orders, deployed capacity and cash flow rather than by announced spending alone.

• Alphabet's share price fell by around 6% in July after data-centre investment pushed free cash flow negative for the first time, while Meta's free cash flow fell 91% and its share price dropped almost 10% at times; Microsoft and Amazon benefited from cloud contracts and booked capacity.
• Alphabet, Amazon, Meta and Microsoft plan to invest well over $700 billion combined by 2026, shifting the focus from budgets to order books, utilisation and the risks created by heavy capital requirements.
• Of roughly 12 gigawatts of US data-centre capacity announced for 2026, only about one third was under construction in spring; 2.06 terawatts of generation and storage projects awaited grid connection, while large transformers had three- to four-year lead times.
• Nvidia reported $96.2 billion of revenue on 26 August, up 106% year on year, including $89 billion from data centres, and forecast around $108 billion for the following quarter; the opportunity now extends to infrastructure suppliers, power generators, cooling and grid-equipment companies.
• Chinese open-source models such as Qwen, DeepSeek and Kimi are increasing price pressure despite export controls and chip shortages, while defence, healthcare and space are creating applications beyond chatbots; SpaceX's Starlink generated $4.3 billion of revenue and $1.7 billion of operating profit in the second quarter after its IPO.

Takeaway: Investors should favour AI businesses whose expansion is supported by visible orders and revenue, while testing whether valuation, financing needs and infrastructure constraints leave enough cash-flow upside; the same discipline applies to potential Anthropic and OpenAI IPOs and to choosing SpaceX bonds over equity exposure.