5 big analyst AI moves: iPhone launch to be negative for Apple stock

Here are the most significant analyst moves in the artificial intelligence space over the past week.

Macquarie upgrades Broadcom, calling it the 'cleanest listed proxy' for Anthropic's compute ramp

Macquarie raised its rating on Broadcom to Outperform, arguing that the risk of Google bringing chip production in-house has largely been reflected in the market and that Broadcom now stands as the clearest publicly listed way to participate in Anthropic's compute buildout.

Analyst Arthur Lai said the stock's notable decline from its 2026 high, along with Google's direct investment in MediaTek, showed the insourcing threat was already priced in. "Broadcom is the cleanest listed proxy for Anthropic's compute ramp," he wrote, projecting that Anthropic will buy more than $40 billion from Broadcom in fiscal 2028, more than making up for lost Google business.

Broadcom's fiscal third-quarter results came in line with expectations, with revenue up 33% quarter-over-quarter to $29.6 billion. Gross profit margin dipped 2 percentage points to 75.0%, as faster-growing but lower-margin AI semiconductor revenue outpaced software, while net profit surpassed Macquarie's estimate by 4%.

Lai also highlighted a broader customer base as a risk reducer. "Growth now rests more evenly on six XPU customers," he wrote, noting that management expects OpenAI to become the second-largest XPU customer by fiscal 2028 and Meta to ship three generations of its MTIA chip.

Macquarie raised its fiscal 2028 earnings estimate by 12% and increased its price target by 12% to $490, implying a total shareholder return of roughly 35%.

Lai wrote that Broadcom dominates the rapidly expanding AI ASIC market, with its SerDes and intellectual-property moat securing a long-term edge in custom silicon, while its software business adds margin resilience and recurring revenue visibility.

Nvidia remains 'one of top large-cap ideas': Baird

Earlier this week, brokerage firm Baird reaffirmed an Outperform rating on Nvidia, calling it "one of our top large-cap ideas" as demand for agentic AI drives a fresh wave of revenue growth.

Nvidia's fiscal 2028 revenue guidance implies growth of about 70% year-over-year—roughly 12 times fiscal 2024 levels—reflecting a supply-constrained outlook driven largely by enterprises building GPU clusters to power proprietary AI agents. Nvidia expects hyperscaler capital expenditures to rise about 65% in calendar 2027, with the cloud industry's backlog now exceeding $2 trillion and hyperscaler AI spending projected to climb to $1.3 trillion in calendar 2027.

Baird highlighted Nvidia's nearly $50 billion in investments in frontier AI labs, describing those ventures 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 an initial 4.25 gigawatts, with the first site expected online in late 2028 or early 2029.

Agreements that use Nvidia's balance sheet leverage with AI labs are expected to account for 25% of demand next year, and clearer visibility into their profitability could provide "the needed impetus for multiple expansion in Nvidia's stock."

On competition, analysts said rival custom AI chips are typically built for a single cloud or service, while Nvidia's GPUs span the full AI life cycle across clouds. They added that many customers' custom ASIC plans have shown performance well below Nvidia's own.

"Nvidia remains one of our top large-cap ideas owing to market leadership, share gains in inferencing, platform solutions positioning Nvidia as AI leader to regional AI native clouds, overall AI industry CAGR, and the lowest P/E within the AI super-compute space," they concluded.

iPhone launch likely to be negative catalyst for Apple stock, analyst says

KeyBanc reiterated an Underweight rating and $250 price target on Apple, with analyst Brandon Nispel arguing that the September 9 iPhone launch event is likely to be a negative catalyst as investors learn the key unknown—pricing—with neither scenario carrying positive implications.

Nispel outlined two possible outcomes: a larger price increase that eases gross margin pressure but risks reducing unit volumes and causing "sticker shock," or more selective price increases that heighten focus on gross margins and raise the possibility of further hikes down the road.

"We don't think either is a great outcome," he wrote, adding that Apple's September event is typically followed by a modest negative reaction in the shares.

GLJ calls Tesla Cybercab launch a 'sell-the-news' event

GLJ Research reiterated a Sell rating and $24.86 price target on Tesla, with analyst Gordon Johnson dismissing the Cybercab launch as falling well short of what even the most sympathetic sell-side coverage had identified as a good outcome.

Johnson said his rule for Tesla product launches is simple: "count what can be counted." On this occasion, that meant 45 Cybercab registrations with the Texas DMV. He noted there was no deployment target, no pricing, no consumer order button, no livestream and no CEO on stage.

Attendees, whom Johnson described as "a curated group of Tesla-friendly creators/cultists," took rides in a two-seater with no steering wheel or pedals inside a small geofenced area in Austin, then posted about their experiences.

Johnson acknowledged that the engineering progress was real. "Twenty-three months ago, in Burbank, the same car was a prop on a soundstage," he wrote, adding that the gap between then and now "is real engineering and we do not wave it away." But he argued the stock is not priced for a car—it is priced for a network—"and on the network the company told us nothing it had not already said."

The deeper problem, Johnson said, is that Tesla landed below the low end of what its most supportive sell-side coverage had outlined in writing. Morgan Stanley's September 2 note had said five to ten Cybercabs would read as "a handful," while 25 to 50 across Texas would likely push shares higher, with the bull case being paid, unsupervised rides across Austin, Houston and Dallas.

"What we got was one city, an undisclosed subset of 45 registered cars, and an app that as of Thursday night would not let you ask for one," Johnson wrote.

"Sell-the-news," he concluded.

Piper Sandler upgrades Tempus AI to Overweight

Meanwhile, Piper Sandler upgraded Tempus AI to Overweight from Neutral, raising its price target to $76 from $56, saying three growth drivers that were previously difficult to underwrite are now increasingly visible.

Analyst David Westenberg said his prior Neutral rating reflected a valuation driven more by AI sentiment than by the performance of Tempus's diagnostics and data businesses.

He now sees a clearer fundamental case built on three catalysts: the pending acquisition of Personalis, which gives Tempus ownership of a differentiated tumor-informed MRD platform; the positive INTerpath-001 readout, which raises the strategic value of Personalis and Tempus as the tumor sequencing provider for a potentially new therapeutic class; and FDA approval of the tumor-only xT CDx, which clears a key hurdle for Tempus to pursue unified ADLT pricing for xT.