JP Morgan thinks that Broadcom is still way undervalued, here is why

Broadcom designs custom AI accelerators, known internally as XPUs, for a small group of hyperscale customers that cannot use off-the-shelf chips at the scale they need. It also supplies the high-speed networking gear that ties tens of thousands of those chips into a single training cluster.

That combination, custom silicon plus the fabric that connects it, has made Broadcom one of the indispensable suppliers in the artificial intelligence buildout. In the latest quarter, total revenue hit a record $22.2 billion, up 48% year over year, while AI semiconductor revenue reached a record $10.8 billion, up 143%.

Management has guided July-quarter AI chip revenue to roughly $16 billion and reaffirmed a fiscal 2027 target of more than $100 billion in AI revenue. Free cash flow and operating profit also hit records, helped by a 77% gross margin and strong operating leverage.

The bank raised its price target to $580 from $500 and kept an overweight rating, citing better-than-expected second-quarter results and a July-quarter guide that topped consensus. AI bookings have now crossed $30 billion against the $10.8 billion shipped in the quarter, swelling the backlog and giving Broadcom unusual multi-year visibility.

Analysts walked through a customer roster that is no longer hypothetical. A multi-generation chip program with Google. An Anthropic agreement that adds about 5 gigawatts of next-generation capacity from fiscal 2027 on top of an existing 1-gigawatt-plus footprint. OpenAI's 1.3 gigawatts slated for fiscal 2027. A Meta program that totals 3 gigawatts through 2028. Two further customers begin ramping in late 2026 with $6 billion in orders already booked.

On the 20-plus-gigawatt compute platform Broadcom is building with Apollo and Blackstone, the bank estimates only about 5 gigawatts will ship in fiscal 2027, leaving meaningful upside in later years. Folding in Google, Meta and the other programs, JPMorgan modeled that fiscal 2028 AI revenue could approach $300 billion, well above current Street expectations.

BofA Securities lifted its target to $530 and projected AI growth near 180% in fiscal 2026 and roughly 100% in fiscal 2027. Evercore ISI raised its target to $582 with an outperform rating, noting the April quarter beat by 2% and the July guide topped consensus by 3%. UBS trimmed its target slightly to $485 but kept a buy rating, flagging supply constraints on AI guidance. Raymond James lifted its target to $450 with an outperform rating. RBC Capital raised its target to $400 while keeping a sector perform view.

Across 28 analysts covering the stock, 24 carry a buy rating and four a hold, with an average target around $510, implying roughly 29% upside from current levels. Even the more cautious voices remained constructive, and management reiterated the fiscal 2027 plan of more than $100 billion in AI revenue.

As custom AI chips become a larger share of the sales mix, gross margin gets squeezed. Management guided third-quarter gross margin down to roughly 74%, reflecting the thinner margins of bespoke silicon relative to the software business. The chief financial officer urged investors to model semiconductor and software margins separately, framing the compression as a product-mix effect rather than a weaker chip franchise.

Management's broader argument is that operating leverage keeps profit dollars climbing even as margin percentages dip. For investors, the key question is whether hyperscaler capex stays at its current torrid pace. The new $580 target says one of Wall Street's most influential desks believes it will.