Macquarie upgraded Broadcom to Outperform and called the chipmaker the cleanest listed proxy for Anthropic’s compute buildout, arguing that the risk from Google bringing chip design in-house has largely played out and is already priced in.
Analyst Arthur Lai pointed to Broadcom’s pullback from its 2026 high and to Google’s direct investment in MediaTek as evidence that the insourcing threat was reflected in the share price. His note, published Saturday by Investing.com, lands at a moment when custom AI silicon has become one of the busiest corners of the semiconductor market, with every major hyperscaler now designing or funding its own chips alongside merchant GPUs.
Six customers, not one
“Growth now rests more evenly on six XPU customers,” Lai wrote, referring to Broadcom’s custom accelerator business. Management expects OpenAI to become the second-largest XPU customer by fiscal 2028, and Meta is expected to ship three generations of its MTIA chip, the in-house accelerator it has been developing for years. The customer list spreads the risk that any single hyperscaler changes plans, which was the market’s main worry earlier in the year when Google was widely reported to be deepening its MediaTek partnership.
Broadcom’s custom chip business has grown quietly into one of the most important profit pools in semiconductors. Unlike merchant GPUs sold off the shelf, XPUs are co-designed with a specific customer and lock in multi-year orders at higher margins. Winning design slots with OpenAI and Meta would make the company the default supplier for two of the largest AI compute programs outside Nvidia’s ecosystem. Anthropic’s own ramp, which Lai cites as the core of the bull case, fits the same pattern: frontier labs that once rented cloud GPUs are now securing dedicated silicon to control costs at scale. Anthropic itself has been one of the year’s biggest AI stories, targeting a record IPO with projections of $190-200 billion in revenue by 2028, per its own investor materials reported earlier this week, and compute is the input its plan depends on most.
Nvidia still the top pick elsewhere
Baird, in the same analyst roundup, kept Nvidia among its top large-cap ideas, noting the company’s nearly $50 billion in investments in frontier AI labs and describing those bets as “working capital limited, not customer nor technology limited.” OpenAI has committed 12 gigawatts of cumulative capacity with Nvidia, including a recently announced agreement supporting the chipmaker’s buildout. Nvidia partner Hon Hai reported August sales up 52% on AI server demand, per Bloomberg, a sign the supply chain is running hot across both merchant and custom silicon.
The equipment side confirms the demand. Global semiconductor equipment billings rose 23% year over year in the second quarter to another record, the second consecutive record quarter, driven by AI infrastructure spending, industry group SEMI reported this week. President Ajit Manocha pointed to sustained investment in advanced manufacturing capacity for chip demand, particularly AI infrastructure. Record equipment billings lead the wafer output by a few quarters, so the number doubles as a forward indicator of how much compute will actually ship next year.
The financing side is expanding too. Nscale, an AI cloud infrastructure provider, lined up a $3.5 billion pre-IPO round backed by Nvidia this week, per Yahoo Finance, part of a wave of AI infrastructure funding that includes gas pipeline deals repositioned as power plays for data centers. Roughly $10 billion of energy deals closed in a single week as developers chased electricity rather than chips as the bottleneck.
Apple the odd one out
While the AI chip names got upgrades, Apple got a downgrade of expectations. KeyBanc reiterated an Underweight rating and $250 price target ahead of the September 9 iPhone launch event, with analyst Brandon Nispel arguing the event is likely to be a negative catalyst as investors learn the key unknown, pricing, with neither scenario working in the stock’s favor. The fall keynote is expected to feature the iPhone 18 Pro, Apple’s first foldable iPhone and new Apple Watch hardware.
The split view reflects where the money is going. AI data center spending keeps setting records, while the consumer side of big tech faces a tougher test: buyers hesitant about pricing, and upgrades that feel incremental. Tesla’s Optimus entering production at Fremont last week was one of the few consumer-adjacent robotics stories to move a stock, lifting shares 3.7% on the day.
What it means for the sector
For Broadcom, the upgrade shifts the narrative from defensive to offensive. The stock had been one of the weaker large-cap AI names of 2026 as investors worried Google’s in-house designs would eat into custom orders. If the six-customer spread holds and OpenAI ramps as expected, the story becomes one of a widening moat in custom silicon rather than a shrinking one. The MediaTek investment cuts the other way: Google is hedging its own supply chain, and Broadcom’s hyperscaler customers are not guaranteed accounts.
The broader read is that AI chip demand has entered a phase where growth is broad enough to support multiple winners. Nvidia keeps the merchant GPU market, Broadcom and MediaTek take the custom designs, and equipment makers bill at record levels either way. The IPO market tells a similar story of an industry at full speed: 238 US listings through early September, up 3.5% year over year, with AI infrastructure and consumer hardware names crowding the calendar, per Value Add Pulse.
The catch, as always with buildout stories, is that hyperscaler capex plans can change with little warning, and Broadcom’s stock will reprice faster than its orders can. For now, though, analysts who cover the name have stopped treating custom silicon as a threat and started treating it as the prize.
discussion