Hyperscalers Set to Spend Seven Hundred Billion Dollars on AI Data Centres in Twenty Twenty-Six

The Biggest Infrastructure Bet in History

The world's largest cloud computing companies are pouring nearly seven hundred billion dollars into AI data centre infrastructure this year, according to Moody's Ratings. The figure represents an eighty-one percent jump from twenty twenty-five and is nearly six times what these companies spent in twenty twenty-two, the year ChatGPT launched.

Amazon leads the pack with a projected two hundred billion dollars in capital expenditure, followed by Alphabet at up to one hundred and eighty-five billion, Microsoft tracking around one hundred and fifty billion, Meta at up to one hundred and thirty-five billion, and Oracle committing fifty billion.

New Chips, New Partnerships

The spending surge is reshaping the semiconductor supply chain. AWS and Cerebras Systems recently announced a collaboration to deploy Cerebras CS-3 chips inside AWS data centres, creating a disaggregated inference architecture that separates AI processing into two specialised stages. The partnership promises dramatically faster AI inference by pairing AWS Trainium chips for prompt processing with Cerebras hardware for output generation.

Questions About Returns

Not everyone is convinced the bet will pay off quickly. Moody's has flagged six hundred and sixty-two billion dollars in off-balance-sheet data centre lease commitments among the top five hyperscalers, obligations that will eventually hit their financial statements. Free cash flow projections for several companies have turned negative in future forecasts, and analysts warn that infrastructure built today may take eighteen to thirty-six months to generate proportional returns.

Executives Remain Bullish

Despite the risks, tech leaders are doubling down. Nvidia's chief executive has estimated that between three and four trillion dollars will be spent on AI infrastructure by the end of the decade, while Morgan Stanley expects Alphabet's spending alone could reach two hundred and fifty billion by twenty twenty-seven.