Morgan Stanley defends Broadcom's AI networking and custom silicon strategy, arguing Google TPU revenue ramp doesn't diminish Broadcom's margin expansion.
Broadcom shares climbed on July 14, 2026, after Morgan Stanley published one of its most forceful defenses yet of the company's position in artificial intelligence chips. Yet the relief appears fragile. The stock remains down approximately 19.1% from its record closing high of $480.77, set on June 2, 2026, and this year's gains have badly trailed the broader semiconductor rally.
Broadcom Inc (AVGO) shares rose 1.32% on July 14 to close at $389.11, though they had climbed nearly 3% earlier in the session. This modest gain underscores the persistent weakness that has puzzled Wall Street: a company posting triple-digit AI revenue growth should not trail its own sector this badly.
Morgan Stanley analyst Joseph Moore addressed this disconnect directly in a note to clients, expressing surprise at Broadcom's underperformance given the strength of its AI growth trajectory. He traced the weakness to one persistent fear: that Taiwan's MediaTek is quietly taking over Broadcom's work building Google's Tensor Processing Units, the search giant's own AI processors and an alternative to buying chips from Nvidia.
Moore's answer is blunt. He reiterated a buy-equivalent rating and a $502 price target, saying Broadcom should keep roughly 80% of Google's TPU business over time. "Our view is that MediaTek participation is real, but not disruptive," he wrote, calling projections of a drop to 50% share or full replacement premature. Moore also acknowledged that Google has its own incentive to lean on multiple chip partners to gain what analysts called "supplier optionality."
The numbers behind Moore's confidence are not in dispute. Broadcom's fiscal second-quarter revenue hit a record $22.19 billion, up 48% year over year, with AI semiconductor revenue climbing 143% to $10.8 billion. The company told investors it expects AI semiconductor revenue to grow more than 200% in the current quarter.
The skepticism did not start this week. On Broadcom's June 3, 2026, earnings call, the company beat Wall Street's revenue estimate but did not raise its full-year AI chip sales target. Shares fell about 15% the next day. CEO Hock Tan acknowledged on that call that Google would likely draw on multiple chip suppliers going forward, a comment that gave the MediaTek narrative real momentum heading into the summer.
Moore's 80% estimate is optimistic relative to some peers. Macquarie has projected that Broadcom's share of Alphabet's TPU revenue will fall from around 95% in 2026 to 80% in 2027 and 65% in 2028—a materially steeper decline. Bernstein's Stacy Rasgon took a more patient view, telling clients the stock may simply need a couple of quarters to digest the transition before its growth reasserts itself.
Other banks are more bullish than Moore, not less. JPMorgan has reiterated an overweight rating and a $580 price target, arguing the Google TPU program has not been delayed or canceled despite what it called "noise" from the Asia supply chain. The spread between analyst targets shows this is a live argument about pace, not a settled question.
Google is not the only reason to own Broadcom. The company extended its custom silicon relationship with Meta in April and signed a $30 billion-plus chipmaking agreement with Apple earlier this month. That diversification is the real hedge against any single customer, including Google, deciding to spread its orders around.
The broader lesson extends past Broadcom. Hyperscalers increasingly treat supplier concentration as a risk to manage rather than a relationship to reward. Google's TPU roadmap is simply the clearest example playing out in real time. Investors in every custom silicon stock should expect a single supplier's share of any one account to compress over time, even as the total pool of AI chip spending keeps growing fast enough to make a smaller slice worth more in dollars.
Whether Moore's 80% figure holds, or whether Macquarie's steeper decline proves closer to reality, will become clear only as Google's next TPU generations actually ship.