Debates about whether we are the peak of the AI infrastructure cycle or close to the peak are popping up on a regular basis as investors are trying to figure out when the inevitable growth slowdown for companies exposed to it (chips, networking) will happen. Broadcom’s latest earnings and recent press reports about xAI spending ambitions suggest we are far from the peak and that the AI infrastructure market still has significant room to grow.
Broadcom, which is the leading maker of custom AI chips, indeed commented last night that the addressable market at its three AI customers (believed to be Google, Meta and Bytedance) could reach $60-90 billion by end 2027 (~80-85% AI chips, ~15-20% networking) as these customers plan to launch clusters of more than 1 million chips each.
This points to an addressable market expansion of 5x in the next three years compared to a current market around $15 billion. Interestingly, this estimate does not take into account the two additional customers that Broadcom just landed (likely OpenAI and Apple, as reported in the press) and that have a revenue potential similar to the company’s three initial customers.
Even if this massive increase in Broadcom’s addressable market probably factors in market share gains at the expense of standard AI chips (=GPUs), it sends a very positive signal about hyperscalers’ capex ambitions over coming years and demand for data center infrastructure hardware.
A recent Financial Times report about Elon Musk’s xAI confirms the bullish environment, with xAI’s newly built Colossus supercomputer in Memphis expected to install more than 1 million GPUs vs. “only” 100,000 right now as Musk aims to catch up with AI rivals (OpenAI, Anthropic, Google…).
Needless to say, the price tag would be significant. Assuming an average GPU price of $30K, the expansion of Colossus would translate into a close to $30 billion GPU investment, suggesting that xAI will have to raise additional money after its recent $6 billion capital injection.
For Nvidia (and other data center plays), the opportunity would be massive. Depending on the timing of the GPU ramp at Colossus (1 year or 2 years), we estimate that the revenue upside would stand between 8% and 15%, which is significant for a company of this size and shows that the revenue/EPS revision momentum is far from over.
