Broadcom, the American chip designer behind custom AI processors for Google, Meta and OpenAI, delivered another quarter of results that beat Wall Street's expectations.
Yet its shares slipped 0.8% in after-hours trading, a sign that investors are growing warier of even strong AI-related earnings.
The muted reaction stemmed largely from Broadcom's fourth-quarter revenue forecast of $34.8 billion, which fell short of the $35.03 billion analysts had pencilled in.
For a stock that has already lagged the wider market this year, up just 6% against the S&P 500's 12% gain, that miss carried extra weight.
The scale of the AI build-out
What stood out most was not the historic numbers but the scale of spending still to come.
Chief executive Hock Tan said Broadcom aims to double its AI revenue to $115 billion in the 2027 financial year, then double it again to $230 billion the year after.
Anthropic, the artificial intelligence lab in which Amazon and Google are major investors, is reportedly planning to deploy five gigawatts of Broadcom's TPU 8i chips in 2027, with potential for another ten gigawatts beyond that.
OpenAI, meanwhile, is preparing to tape out a second custom chip with Broadcom and has begun discussions on a third.
Who is really taking the risk
The more striking disclosure came from finance chief Amie Thuener, who said Broadcom may offer "residual value guarantees" to AI labs, effectively underwriting some of the financial risk these companies take on when committing to vast infrastructure spending.
That detail matters because it points to a pattern now emerging across the AI supply chain: chipmakers extending financial backing to customers whose revenues do not yet match their spending commitments.
For UK investors watching the sector through London-listed proxies and index trackers, it is a reminder that the AI capital expenditure cycle increasingly rests on circular financing arrangements between a small number of firms.
