Apple Explores Intel and Samsung for U.S. Chip Production

Mark Gurman said Apple has held early discussions with Intel and Samsung about using new U.S. fabrication plants to produce future A-series and M-series chips, the processors that power the iPhone, Mac, iPad, and other devices. The talks are exploratory, but Gurman framed them as part of a larger shift: Apple wants more manufacturing capacity closer to home and more redundancy around the single most important component in its supply chain.

Apple has designed its main processors in-house for more than a decade while relying on TSMC to manufacture them, primarily in Taiwan. Gurman said some production has been shifting to TSMC’s fab in Phoenix, and Apple has said it expects 100 million chips from the area by the end of the year. But additional fabs have been slow to come online, and Apple is looking for secondary suppliers.

The logic, in Gurman’s telling, is not simply political. It is about supplier concentration, geographic concentration, and the risk of a disruption in Taiwan. He described the dependence on Taiwan as “probably the biggest risk factor for Apple right now,” because if Apple cannot get the chips it needs from Taiwan, “they're not gonna have a very good year.”

Politics are still part of the calculus. Gurman said working with Intel could improve Apple’s relationship with the Trump administration, especially given the U.S. government’s stake in Intel and Donald Trump’s recent public attention to Intel’s stock price and investment growth. Tariffs had also been a concern months earlier, giving Apple another reason to consider U.S.-based chip production.

But Gurman emphasized the operating logic. Apple routinely uses multiple suppliers for other components, such as displays and speakers, to reduce supply-chain risk. The same logic is now being applied to the company’s own silicon. Hyde added that the dependence is not only Apple’s problem, saying on air that “the entire world relies upon Taiwan” and referring to “about 90% of all chips made out of there.”

The market reaction showed how consequential investors considered the possibility of Apple diversifying its chip manufacturing. Intel was shown up 13.06% intraday, Samsung Electronics up 8.70%, and Apple up 1.42%. Hyde said Apple executives had started to look at Samsung production out of Texas, while Gurman’s later description was broader: early discussions with Intel and Samsung about new U.S. fabs that could potentially produce Apple chips down the road.

The Apple question is whether a company whose products depend on custom silicon can preserve the advantages of its current chip strategy while reducing reliance on a single supplier geography.

Alphabet, Microsoft, and xAI have agreed to give the U.S. government early access to their AI models before public release, joining OpenAI and Anthropic, which have had similar agreements since 2024. Maggie Eastland said the agreements are with the Center for AI Standards and Innovation, an office within the Commerce Department that evaluates models.

Eastland said the new agreements signal “at the very least appetite from U.S. officials” to understand what advanced models can do before they reach a wider audience. But she drew a distinction between evaluation and regulation. The center, she said, “doesn’t do any sort of regulation.”

That distinction matters because, as Eastland put it, The New York Times and The Wall Street Journal have separately reported that the White House is weighing a cybersecurity executive order that could include an oversight mechanism. She said the new model-access agreements could be part of the evaluation side of that policy environment, but are unlikely to be the oversight mechanism itself. More evaluation could still become important for enforcement, including enforcement of existing laws.

Hyde noted that AI companies’ relationships with the U.S. government are becoming more formal, including updated relationships involving Alphabet, xAI’s work within government, and AI deployment in departments connected to war or defense. Eastland’s emphasis was on scope: more firms are allowing the Commerce Department’s AI evaluation center to review models early, but the arrangement does not itself create a regulator.

Seth Boro said the latest AI model releases have accelerated the cybersecurity threat landscape. He referred to “Mythos” as the model drawing current attention for cyber risk, but said the important point is not one model: “there’s going to be a lot of them.”

Boro, managing partner at Thoma Bravo, said the firm’s cybersecurity portfolio companies collectively produce about $8 billion of revenue and cover multiple layers of cyber defense. He argued that enterprises now need a layered security approach because threats will emerge faster than companies have previously had to handle.

When Dani Burger pressed him on models that could find zero-day vulnerabilities in minutes, Boro pointed to network effects inside cyber platforms. Proofpoint, he said, has 14,000 customers and sees malicious inbound emails and employee interactions with them every day. That breadth of visibility helps detect and respond to zero-day threats quickly. Burger challenged whether that is as quick as a model like Mythos; Boro replied that Mythos has not hit the market yet, but that organizations should act as if such capabilities are out there or already being used.

Boro treated the publicity around model cyber risks as useful, not merely promotional. He said it gives enterprises and consumers a heads-up about what is coming. As agents are deployed more widely, governance becomes critical: what agents are doing, what information they have, where the data comes from, and what actions they take after receiving it. He said portfolio companies including SailPoint, Ping, Proofpoint, and Darktrace monitor those environments and respond to malicious activity.

The economics of AI deployment were a separate constraint. Boro said many people do not yet understand the cost of rolling out AI solutions. For many higher-functioning roles, he said, current research suggests it is more expensive to bring tokens into an organization and perform work agentically than to use a human, though he added that this is not true for everything and will not always remain true.

Inside enterprises, Boro said budgeting, process re-engineering, and daily inference costs all take longer and cost more than many expect. The response he is seeing is the use of specific models for specific use cases rather than relying on general-purpose models for every function. He expects efficiency and power consumption to become major areas of focus.

Lauren Webster made a related point from the software-investing side. AI adoption and implementation, especially inside enterprises, will require work around deployment, security, process design, and integration. Webster said she is bullish on the services sector that will emerge around those implementation projects.

That services opportunity sits alongside labor pressure. Webster said investors are pushing companies harder on profitability, and that the balance between growth and profitability has shifted over the past year. Businesses would be looking for cuts regardless of AI, she said. But she also said AI will displace some types of tech workers.

Technology stocks were broadly higher, with the Nasdaq 100 shown up 1.27% intraday and the Philadelphia Semiconductor Index up 3.66%. Hyde described both as reaching another record high. Lauren Webster described the move as a turnaround from the start of the year, when software stocks sold off amid uncertainty about how AI would affect traditional software companies.

Webster said April produced the first “constructive tape” for software indices, even though software remained below the broader market. She characterized the investor shift as moving from “AI panic and uncertainty” toward “AI execution” and greater discernment about which companies will benefit and which may struggle.

For software moats, Webster highlighted enterprise adoption. Enterprise software is harder to “rip and replace,” she said, because projects are budgeted years in advance, implementation timelines are long, and services are wrapped around deployments. Easier-to-deploy tools may be more vulnerable to replacement by newer technology from foundational labs.

Defense technology is another area attracting capital. Webster pointed to a rotation into defense tech, particularly companies with a software angle. She linked that demand to geopolitics and the flow of money into national security, along with “a new way of doing business in defense technology.”

Palantir was a test case for that investor scrutiny. Its shares were shown down about 5.5% intraday even after the company posted results that beat expectations on most lines. Lizette Chapman said the issue was not that the quarter was weak, but that the stock’s valuation demanded more. Hyde noted that Palantir traded at more than 80 times future earnings. Chapman said Palantir beat even its raised revenue guidance and raised guidance again, but not by as much as in prior quarters.

The competitive question is whether frontier AI companies such as Anthropic and OpenAI, which are increasingly using forward-deployed engineers, can encroach on Palantir’s terrain. Chapman said Palantir executives argue that what separates the company is its ontology: a real-time mapping of an organization’s data across enterprise apps and silos into one operational view. That layer, in Palantir’s framing, sits between the enterprise and large language models.

Chapman gave defense and industrial examples: using data to identify where improvised explosive devices are in wartime, or to locate supply-chain disruptions at companies such as Airbus. The concern for investors is that Palantir’s U.S. business remains the bright spot and that international growth may not be keeping pace. Chapman said U.S. concentration was a worry this quarter, especially for commercial sales, and that a lot of growth is already baked into the stock.

A Milken Conference exchange reinforced Palantir’s own account of its moat. Josh Harris said the company spent years being misunderstood because it did not fit a familiar software category. Concepts such as forward-deployed engineers are now common in tech vocabulary, he said, but Palantir faced years of skepticism that it was consulting rather than software.

Harris argued that Palantir’s long-standing focus on data integration and Foundry is what enables agentic AI and other applications on top of it. The prerequisite, he said, is a reliable data asset and “a single source of truth.” What was once hard to explain, he suggested, is now becoming more obviously foundational.

Pinterest shares were shown up about 11.5% intraday after first-quarter sales topped estimates and the company forecast revenue above Wall Street expectations. Hyde described it as Pinterest’s best stock day in a year. Bill Ready said Pinterest has delivered 11 straight quarters of record-high users and 10 straight quarters of double-digit user growth.

Ready’s central claim was that Pinterest has become an “AI-powered shopping assistant.” Gen Z is now more than half the platform and its fastest-growing demographic, he said, adding that “Pinterest is where Gen Z goes to shop.”

The company has more than 80 billion searches per month, and Ready said more than half are commercial. That gives Pinterest, in his view, a different intent profile from most chatbots and from historical general-purpose search. The platform’s AI strategy is built around visual search and shopping discovery rather than broad conversational intelligence.

Ready said Pinterest primarily runs its own compact, fit-for-purpose models, also using open source and retraining models on its own dataset. He argued that this approach can deliver comparable or better results at often less than 10% of the cost of broader proprietary models. For shopping recommendations, he said Pinterest has achieved 30% better relevancy than leading off-the-shelf proprietary models.

The reason, he argued, is not the model alone. Pinterest’s signal comes from users expressing taste and style through what they save, search, and assemble. “The AI doesn’t have taste or style,” Ready said. “Humans have taste and style.” Pinterest learns from those taste signals and uses them to produce recommendations that users describe as “Pinterest just gets me.”

Ready also defended Pinterest’s move into connected-TV advertising through its acquisition of TV Scientific. At more than 630 million users and 80 billion monthly searches, he said Pinterest has “one of the highest commercial intent platforms anywhere in the world.” Connected TV is one of the fastest-growing areas of ad supply and demand and is projected to surpass linear TV in 2028, according to Ready. He said adding Pinterest audience data to TV Scientific’s capabilities produced a 65% improvement in purchases resulting from shown ads.

On youth safety, Ready said he came to Pinterest nearly four years ago wanting to prove that “a more positive business model” was possible in social media. Pinterest made accounts private-only for users under 16 about three years ago and turned off social features for that group. Ready said he has publicly argued that social media “as currently configured is not safe for users under 16,” and said it is encouraging that regulators globally are paying attention.

Grab’s first quarter showed 24% growth in gross merchandise value, 24% year-over-year revenue growth to $955 million, net income of $136 million, and operating income of $22 million, according to the on-screen results. Peter Oey said the quarter was unusually strong despite Q1 normally being softer because of Ramadan and Lunar New Year.

Oey attributed the growth to strength across food, mobility, and financial services. Grab’s financial services business crossed more than $1 billion in loan disbursal. He said affordability has been central to expanding the company’s user funnel and increasing monthly transacting users. Grab now has 52 million monthly transacting users, which Oey described as “1 in 13 Southeast Asians” using the product.

Fuel prices are a pressure point, but Oey said the impact varies sharply by market. Grab operates in eight Southeast Asian markets. The Philippines has been hit hardest, with fuel prices up two to three times in some cases, while increases in Malaysia and Indonesia have been nominal because of different fuel-subsidy structures. Grab has responded with fuel vouchers, pump discounts through oil-company partnerships, and lending programs for drivers. Oey also said higher fuel prices create an opportunity to accelerate EV adoption across two-wheel and four-wheel vehicles.

Indonesia’s regulatory changes present a more direct business-model challenge. Hyde asked about a new 8% maximum commission in Indonesia. Oey said the affected business — two-wheel mobility in Indonesia — is less than 6% of Grab’s overall mobility GMV. He called the change unexpected but said Grab understands where it is coming from and is working with the government and industry peers on mechanics.

Oey said Grab has levers to neutralize some of the impact, including existing driver benefits such as medical coverage, insurance, and incentives. The company may need to recalibrate the two-wheel business model in Indonesia or the fare structure, but he said passing the change directly to consumers is “the last thing” Grab wants to do. The goal, he said, is to keep the marketplace healthy while ensuring drivers benefit.

On credit, Oey said first-quarter loan quality remained strong and that stress testing had not shown anything detrimental to the loan book. Grab is incorporating variables including fuel prices, inflation, and potential unemployment into its models. In some countries, he said, the company may tighten the funnel to keep risk appetite balanced.

Alphabet’s debt issuance underscored the scale of financing around large technology companies as AI spending rises. Hyde said Alphabet was issuing a record €9 billion in euro-denominated debt and seeking $3 billion to $5 billion in Canadian-dollar debt. She described the proceeds as for general corporate purposes, then added: “but it’s AI.” Alphabet, she said, has planned capital expenditure of up to $190 billion. Its shares were shown up about 1.4%, and Hyde said Alphabet was moving toward Nvidia’s market capitalization, with Alphabet at $4.69 trillion and Nvidia at $4.79 trillion.

Semiconductors remained the center of market enthusiasm. Ian King said AMD’s expected numbers would likely be good, but the question was whether they would be good enough after a major stock run. He compared the setup to three months earlier, when AMD was bullish but “just not bullish enough” for investors.

King described AMD CEO Lisa Su as “innately conservative,” more inclined to execute and show results than to promote multibillion-dollar agreements or extremely bullish forecasts. The challenge, he said, is that the current market rewards companies speaking in that more promotional register. Investors are tuned to very large AI opportunity claims, and companies that do not make them risk appearing to be missing out, even if they are executing.

Intel’s stock strength fit the same market environment. King said Intel had materially missed out on the broader semiconductor “cavalry charge,” but even signs of promise and improvement were being rewarded.

He also addressed concerns about circular financing in AI infrastructure. Jensen Huang, Nvidia’s CEO, had said at Milken that he had put a lot of money to work to help get the ecosystem going and hoped enough had been done. King interpreted that as a possible signal that fewer such deals may be needed if companies improve gross margins and become more self-sustaining.

Paramount Skydance shares were shown lower by about 4.6% even after the company beat some expectations in its fiscal first quarter. Hannah Miller said Paramount was pleased with the momentum since the Skydance merger, including an earnings beat and a strong outlook for the year.

The pressure, in Hyde’s framing, was that investors were looking at how the company executes its acquisition of Warner Brothers. Miller also noted weakness in streaming tied to the end of an international streaming agreement, which led to subscriber losses, though she said Paramount expected that to be a blip.

Hyde said Paramount had seen a gain in average revenue per user and domestic subscribers, and then connected the streaming picture to Disney. For Disney, Miller said attention would be on parks, cruises, streaming, and the company’s new CEO, after Disney had previously described the February quarter as tepid.