Apple’s Reported Intel Deal Shows Compute Bottlenecks Driving Industrial Policy

John Coogan framed Intel’s reported preliminary chip-making agreement with Apple as part of a larger attempt to rebuild American semiconductor capacity. Intel’s stock was up sharply on the news, and Coogan treated the move less as an isolated commercial announcement than as the next piece in a months-long effort to get major U.S. technology companies to commit business to Intel.

The on-screen source was a post from Andrew Curran saying Intel had reached an agreement to manufacture chips for Apple devices, with the Trump administration pushing for the deal. Coogan said the Wall Street Journal report described “intensive talks” between Apple and Intel that had been going on for more than a year, with a formal deal hammered out in recent months. The specific Apple products Intel might manufacture chips for were still unclear.

The significance, in Coogan’s telling, is that Intel is both a chip designer and a manufacturer, and both sides of that business had been underperforming before Lip-Bu Tan took over as CEO. Intel’s foundry ambitions depend on convincing external customers that its next advanced manufacturing process is worth betting on. Coogan pointed to Intel’s 14A process as the node Intel hopes will attract a coalition of customers: “if you build it, we will come.”

The reported Apple agreement sits alongside prior Intel commitments involving Nvidia and Elon Musk’s companies. Coogan said Commerce Secretary Howard Lutnick had met repeatedly with Apple executives including Tim Cook, as well as Elon Musk and Jensen Huang, in an effort to push them toward Intel. Nvidia invested $5 billion in Intel and announced a partnership under which Intel would build custom data-center CPUs for Nvidia. Musk had announced a Terafab project, and Coogan noted that Musk had long been “in and around the fab world,” including with Samsung for Tesla chips and rumors involving GlobalFoundries.

Jordi Hays emphasized the financial upside for early Intel backers. Nvidia, he said, got in at about $23 a share on its $5 billion investment, and Intel was trading around five times that level. Coogan added that Jensen Huang’s position likely produced a one- or two-billion-dollar mark-to-market gain just from the day’s move.

The federal government’s role was central to the discussion. Coogan said the Trump administration had converted nearly $9 billion in federal grants into Intel stock, giving the U.S. government a 10% stake. He also described the administration as having pressured “both sides” in the Apple-Intel talks by invoking American manufacturing resilience and the need to reduce dependence on Taiwanese supply chains. In Coogan’s words, the government was trying to give Intel “a real shot at underwriting the next big fab.”

The political relationship with Tan had also shifted. Coogan recalled that Trump had initially raised concerns about Tan’s ties to China and had called for his ouster after Intel hired him in March 2025. Tan, Coogan said, then won Trump over through what the Journal called a “charm offensive,” after which the government announced its 10% investment. Coogan also cited Trump saying, “As soon as we went in, Apple went in, Nvidia went in. A lot of smart people went in.”

Apple’s motivation was not described purely as patriotism. Coogan stressed that Apple remains heavily dependent on TSMC for chips across iPhones, iPads, Macs, and other devices, and that TSMC’s capacity is increasingly contested by Nvidia and other AI-chip designers. Cook had blamed shortages of advanced chips on Apple’s inability to meet demand for iPhones, and said some Mac models — including Mac Mini and Mac Studio — could take months to reach supply-demand balance. Coogan noted that Apple had raised the Mac Mini starting price after the earnings call.

The constraint changes Apple’s leverage. Coogan said Apple has long been TSMC’s top customer, but the surge in AI-chip demand means Apple no longer has the same ability to secure the supply it needs. Intel, then, is a geopolitical hedge, a capacity hedge, and a potential beneficiary of Apple’s need to dual-source production.

Coogan connected Intel’s move to a broader market pattern: a handful of AI-adjacent companies are driving a disproportionate share of stock-market returns. He cited a chart shown from Steve Hou comparing peak concentration levels in historic bubbles. The chart showed an “AI Big 10” at 40% of the S&P 500, compared with 63% for railroads as a share of the U.S. stock market in the 1835–1910 period, 44% for Japan as a share of MSCI ACWI, 41% for the Nifty Fifty, and 40% for tech and telecom during the 2000 period.

Coogan identified the AI Big 10 as the Mag 7 plus AMD, Broadcom, and Micron. The label mattered because it widened the lens beyond the best-known platform companies into the firms providing compute, networking, and memory for the AI buildout.

Hays said there had recently been “three sort of beautiful weeks” in which many market participants argued it was not a bubble because revenue growth was too strong. Coogan did not dismiss valuation concerns, but he repeatedly returned to the point that revenue growth in the AI economy remains substantial. The tension was not presented as a simple bubble-or-not-bubble call. The market can be highly concentrated and expensive while the underlying companies are still reporting extraordinary growth.

Coogan cited Greg Ip’s Wall Street Journal analysis as a way to separate the AI economy from the rest of the American economy. According to Coogan’s summary, Ip estimated that the AI economy grew 31%, while the non-AI economy grew just 0.1%. The split showed up in investment categories: personal consumption grew a muted 1.6%, investment fell in housing, business structures, factories, and transportation equipment, while investment rose 43% in tech equipment, 23% in software, and 22% in data.

Hays pushed back slightly on the idea that software is merely weak outside AI, pointing to a meaningful rebound in SaaS revenue growth at companies such as Datadog and Atlassian. Coogan agreed that software had shown more resilience than the market had expected. He said the pressure had been widespread — even companies with strong network effects, where software itself was not the moat, had to “answer to the market” — but some recovered quickly, some took a quarter or two, and some remain beaten down.

The result is a market with multiple layers of divergence. AI infrastructure is pulling indexes higher. Some software companies are reaccelerating. Other parts of the economy are barely growing. Coogan called this “K shapes within the K shapes,” meaning the familiar split between winners and losers is itself subdividing into additional winners and losers.

The April jobs report complicated the view that AI alone is holding up the economy. Coogan said the U.S. added 115,000 jobs in April, far above the 55,000 expected by analysts polled by the Wall Street Journal. The visual shown from the Labor Department charted nonfarm payroll changes and highlighted April’s 115,000 increase.

The gain was down from a net 185,000 jobs in March, but Coogan treated the upside surprise as material because it “breaks the narrative” of a real economy merely being carried by Nvidia earnings. Retail, transportation, warehousing, and healthcare all contributed to the job gains. Those are not the sectors most directly associated with AI capex, which made the report more important to his point.

Coogan made a scale argument. A layoff of 4,000 people at a technology company may matter a great deal inside the tech industry, but the U.S. employs more than 100 million people. Those cuts do not necessarily move the national labor-market needle. The labor market can absorb visible tech weakness while still adding jobs in healthcare, logistics, retail, and other large employment categories.

Hays was more cautious about declaring victory. He said he was waiting “to hit the gong” until revisions came in, noting that revisions can arrive months later. Coogan acknowledged that revisions had not yet appeared for March in the data being discussed.

Coogan quoted KPMG chief economist Diane Swonk describing the labor market as “still a high anxiety job market.” In that framing, people who have jobs are holding onto them, while job seekers feel frozen out. That distinction helped reconcile the strong payroll number with widespread anxiety: the economy can be hiring, but mobility and confidence can still feel poor.

The point was not that the non-AI economy is booming. Coogan’s claim was narrower: outside AI, there is more resilience than the dominant market narrative suggests. The real economy is not generating the same explosive growth as AI infrastructure, but neither is it collapsing in the way a pure “AI is holding everything up” story would imply.

Coogan used Whirlpool and Six Flags as a concrete test of what is happening outside AI. The contrast was intentionally counterintuitive. Whirlpool sells appliances — refrigerators, washers, dryers — that look like necessities. Six Flags sells theme-park visits, which look discretionary. Yet Whirlpool had cut its dividend after paying one consistently since the 1950s, while Six Flags reported higher first-quarter revenue, growing attendance, and rising customer spending.

For Whirlpool, the dividend cut carried symbolic weight. Coogan said the company had maintained its dividend through the Great Recession and the dot-com crash. Cutting it now signaled serious pressure for a business long treated as a dividend stock. The stock had traded down on the news and was down 80% over five years. Coogan also pointed to debt, heavy competition, and weaker existing-home sales as pressure points.

For Six Flags, the screen showed a news article titled “Six Flags Revenue Rises on Attendance, Spending Growth.” Hays added that Six Flags trades under the ticker FUN, and that Travis Kelce and a group had invested $200 million. He said the stock had fallen after that investment but had since recovered to roughly where it had been around November. Coogan noted that Six Flags was still down significantly over a longer period — about 50% over two years, and formerly a roughly $5 billion company versus about $2.3 billion — but the operating result was still notable.

Coogan’s explanation rested on deferrability. A refrigerator is necessary, but buying a new refrigerator can be delayed. Households can repair old appliances or postpone upgrades if money is tight. Theme parks are discretionary, but children are only “roller coaster age” for a limited time. Families may still prioritize that experience even in an uncertain economy.

The competitive structure also differs. Whirlpool faces brutal global competition from LG, Samsung, and other international appliance makers. It is also exposed to housing turnover: fewer existing-home sales mean fewer appliance upgrades. Six Flags, by contrast, benefits from a form of “anti-slop” demand — experiences that cannot be replicated by software or generative AI.

Coogan connected this to what he called the “barbell thesis” of the AI future. One end is AI infrastructure and synthetic content. The other end is scarce, physical, human, or legacy experiences. He used the Ellison family as an example: long AI infrastructure through Oracle, and long “anti-slop” through legacy media properties such as Batman and Superman via Warner Brothers. He also used Josh Kushner as an example, with exposure to OpenAI on one side and the San Francisco Giants on the other.

In Coogan’s shorthand, “you can’t vibe code Space Mountain.” The line captured why some real-world leisure and experience businesses may prove more resilient than their discretionary label suggests.

A Financial Times chart shared on screen by Ryan Petersen framed the possible AI futures in extreme terms: AI could end scarcity, end humanity, or simply boost trend growth by 0.2 percentage points. The chart plotted U.S. real GDP per capita on a log scale from 1870 to 2024 and projected paths for “Tech singularity: end of scarcity,” “AI-boosted growth path,” and “Tech singularity: human extinction.”

Coogan said the bull case takes real GDP per capita north of $1 million, while the human-extinction scenario goes to zero. But the more plausible near-term economic question in the discussion was the middle path: a steady upward trend from AI-boosted growth. Coogan described that as the goal people should be working toward, with the end-of-scarcity outcome as another possible upside case.

The chart helped Coogan explain why AI produces “dual anxiety.” If AI becomes too capable, people worry about mass unemployment. If AI proves to be a bubble and collapses, people worry about recession and job losses. In both cases, workers and investors can arrive at similar economic anxiety from opposite premises.

That, Coogan said, is where many bubble concerns come from. The worry is not only valuation. It is that the economy and the market have become dependent on an AI investment cycle that seems unstable whether one believes in its transformative potential or in its overextension.

Coogan’s position was not that there are no bubble characteristics. He acknowledged that valuations are high. But he argued that there were not many obvious signs of a bubble because the revenue-growth charts remain strong. In other words, the skepticism has to account for the operating data, not just the price action.

The discussion briefly veered into a joke about “back of the envelope” versus “napkin math,” but the distinction echoed the larger point. Hays said he is spiritually a napkin-math person: if an opportunity cannot be exciting based on what fits on a napkin, he will not get excited about it. Coogan said that for something serious, he would “pull out the envelope.” Around AI markets, the implication was that the napkin case is obvious — huge demand, huge capex, huge potential — but the envelope math is now required because the stakes include employment, industrial policy, and market concentration.

Tyler Gold attended the OpenAI–Elon trial in Oakland and described the courtroom scene in practical terms. Because it was a federal case, some seats were available to the public, but only around 20 to 30 by his estimate, with part of that capacity reserved for media. Tyler arrived at roughly 5:30 a.m. to secure a place in line. The trial began around 8:30 a.m. and ran until about 2 p.m., with two 20-minute breaks.

The day he described was heavy on video depositions and documents. The first major segment was a video deposition of Mira Murati, focused mostly on the timeline of Sam Altman’s ouster. Tyler said that was where many of the text messages being discussed publicly had surfaced. The court then went through documents again, including text messages.

A post shown on screen from Yash Bhardwaj turned Sam Altman’s texts to Murati into an AI-generated “2011 style emo teenage heartthrob anthem.” The visible lyrics included Altman asking, “can you indicate directionally good or bad? satya and others anxious,” and Murati replying, “Directionally very bad.” Another visible line included “can you wrap up soon? lots of pressure from microsoft for an update.” Coogan admired the motion graphics and asked whether the song was made with Suno. Tyler said the creator had described the process: OCR the images of the texts into plain text, remove names from the dialogue, paste into Suno, and iterate 20 to 30 times until a catchy version emerged.

Hays said he would like to play the AI-generated song for an AI skeptic because, in his view, “how do you see this and not want to build?” The remark was partly comic, but it also fit the episode’s broader interest in how AI tools are becoming culturally legible through casual, fast production.

Tyler also saw Shivon Zilis testify in person. He described her as an OpenAI board member from 2020 to 2023 who stepped down when Elon started xAI. Coogan initially assumed she had been on the board during the November ouster, but Tyler corrected the timeline: Zilis left in February 2023, before the ouster. Coogan noted that the trial moves between two main periods of conflict — earlier battles involving Sam Altman, Greg Brockman, Ilya Sutskever, and Elon Musk, and the later board ouster.

Tyler said the testimony involved a great deal of jargon, though he did not find the lawyers’ use of it especially sophisticated. Zilis, he said, used more relevant AI language when discussing why she got into the field, including acceleration-oriented language. The day also included about an hour of Helen Toner’s deposition, focused on the ouster.

The public section of the courtroom, Tyler said, was effectively full of media-adjacent observers. Many people were typing notes on laptops. Tyler said he mostly sat there and enjoyed it, describing the material as “good stuff.”

Near the end, Coogan turned to a post from Ejaaz claiming DeepSeek was raising $7 billion at a $50 billion valuation, described in the post as China’s largest-ever AI raise. The post said founder Liang Wenfeng was personally contributing 40% of the round — $3 billion — and owned 90% of the company. It also said DeepSeek was founded inside Liang’s hedge fund, described as one of China’s most successful funds.

Coogan focused on what the money would be used for: acquiring compute to push out new DeepSeek models more often and turning the company revenue-positive through enterprise products, in the same broad pattern as OpenAI and Anthropic. He said DeepSeek appeared to be trying to get back onto the frontier curve after prior charts had shown Chinese open-source models falling behind on a different performance trajectory.

The DeepSeek item led into another compute-allocation story. A Zach Brock post shown on screen said: “congrats to anthropic for defeating grok in the market and feasting upon the compute of their fallen enemy.” Hays identified the reference as the xAI or SpaceX–Anthropic deal that had broken while the show was off.

Coogan said many people had predicted such a deal, but he had not. He thought it was rational for the parties, yet doubted they could get past the cultural tension between Elon Musk and Anthropic, especially given that Musk had recently been insulting the Anthropic team. His conclusion was simple: “demand for compute finds a way.”

Hays said he and others had previously identified the possibility that Elon’s infrastructure operation could become a neo-cloud — building compute and power quickly, then renting that capacity to others. Coogan said he had discussed that idea often the prior year because Elon’s companies are unusually good at building infrastructure fast and bringing power online. But as events evolved, he thought the Cursor deal might be the long-term use for the compute. The Anthropic arrangement suggested there may be enough capacity — across multiple clusters and Colossus data centers — for more than one major buyer or use case.

Across Intel, Apple, Nvidia, DeepSeek, Anthropic, and xAI, the same constraint kept reappearing. The companies differ in geography, politics, business model, and culture. But the market is rewarding whoever can secure advanced manufacturing, power, chips, and usable compute at scale.