OpenAI and Broadcom jointly unveil Jalapeño, a custom AI inference chip claiming 50% cost reduction versus alternatives

To own Broadcom, you must believe its AI custom chips and networking will remain central to hyperscale buildouts, while its software business (VMware) provides steadier margins. The Jalapeño launch deepens Broadcom's embedded role within OpenAI's infrastructure stack and supports the near-term AI revenue ramp. However, it does not eliminate key risks: heavy reliance on a handful of hyperscale AI customers and a large debt load.

Among recent announcements, the AI XPV Platform partnership with Apollo and Blackstone stands out alongside Jalapeño. The US$35 billion financing tranche for Anthropic compute, tied to over 20 gigawatts of AI capacity through 2028, reinforces the core catalyst driving the OpenAI deal: Broadcom's XPUs and Ethernet networking are being written directly into large, multi-year AI infrastructure build plans rather than treated as one-off component sales.

Yet behind the OpenAI excitement, investors should weigh how Broadcom's customer concentration and elevated debt could affect returns if AI capex expectations shift. Broadcom's base case projects $213.4 billion in revenue and $93.7 billion in earnings by 2029—requiring 46.2% yearly revenue growth and a $68.7 billion increase in earnings from $25.0 billion today. These projections yield a $476.78 fair value estimate, representing roughly 25% upside to current prices.

Some of the most bullish analysts were already assuming Broadcom could reach approximately US$326.5 billion in revenue and US$162.0 billion in earnings by 2029. If Jalapeño and similar deals accelerate or stall the hyperscaler-led AI pipeline, those exceptionally optimistic forecasts and their associated customer concentration risk could prove either prescient or overly bold compared with today's more measured consensus.