Intel and the US Semi Future, Will Any Big Tech Incumbents Lose in an AI World?, Questions on Tea, Grok, and The Ringer
Summary
Intel’s 14A disclosure turns a long-running strategic problem into an explicit capital constraint: internal products cannot generate acceptable returns without a meaningful external customer. Ben Thompson traces that back to missing mobile volume, but warns that industrial foresight is not the same as a trade: “If I had shorted Intel, I would’ve been both right and bankrupt.” The cleaner long-duration winner was TSMC.
Intel’s obstacle is less process technology than a vertically integrated culture that never learned to serve outside customers. Its historic obsession with performance tolerated poor yields, inefficient manufacturing and extensive product binning; Thompson now believes “Intel’s culture is fundamentally incapable of becoming a foundry.” A viable US foundry may therefore require a split and direct government support rather than another corporate turnaround.
Backside power captures Intel’s predicament: 18A and 14A are technologically ambitious but insufficiently matched to what customers actually want. Apple’s roughly 7–8-watt phone chips do not need the feature, while Nvidia can pay TSMC’s higher-cost, bonded-chip approach for processors drawing 700–800 watts and selling for $30,000–$60,000. Intel remains “more technologically interesting and creative,” but “the problem is it’s not meeting anyone’s needs.”
Samsung’s $16.5 billion Tesla deal could establish the second US leading-edge supplier before Intel finds its anchor customer. Tesla’s AI6 chip is returning to Samsung at two nanometers and anchoring its Texas fab; Apple and AMD appear locked into TSMC, Nvidia historically distrusts Intel, and Qualcomm has kicked the tires most seriously but has said no. If Samsung becomes viable, Thompson asks whether there is room for three suppliers at all.
US semiconductor resilience looks materially better even as Intel deteriorates. TSMC now refers to Arizona as a “GigaFab,” with its own R&D and a figure Thompson recalled as roughly 30–40% of its advanced chips eventually being made in the US; Samsung’s Texas investment could acquire similar gravity. Thompson’s defining line: “TSMC is becoming Intel, and they’re becoming Intel much faster than Intel was able to become TSMC.”
AI increasingly looks sustaining for Big Tech, whose distribution, infrastructure and free cash flow can overwhelm venture-funded challengers. Google Cloud is exploiting differentiated AI and workload gravity as Google’s CapEx rises from $55 billion in 2024 to $75 billion and then $85 billion, even though success lowers the parent company’s margin profile: “They are investing heavily to get worse margins, which is exactly what they need to be doing.”
OpenAI remains the most credible cross-platform disruptor, but Meta can weaken it simply by making the race prohibitively expensive. ChatGPT has won substantial consumer mindshare, yet OpenAI cannot stop model R&D without falling behind; Zuckerberg’s reported $100 million talent offers either recruit its people or inflate its costs, “like a boxer doing body blows,” financed from Meta’s free cash flow. Google Search and Apple’s position remain genuinely unresolved fronts.
The episode’s broader business-model lesson is that durable financing often matters more than the celebrated product. Grantland vanished when ESPN stopped funding it, while The Ringer’s monetizable podcasts let Spotify and Bill Simmons preserve lower-return writing; text acquires audiences, podcasts monetize them. Thompson’s epitaph is memorable: Grantland is “the Kurt Cobain of websites,” permanently great because it disappeared before it could fade.
Deep dive
1. Intel’s 14A gate makes a volume crisis twelve years in the making explicit
Lip-Bu Tan’s message was unusually direct: Intel products can generate reasonable returns through 18A, but 14A’s higher capital cost requires both internal volume and “a meaningful external customer.” He will invest only when convinced those returns exist.
Thompson’s translation starts with mobile: Intel missed the industry’s largest source of chip volume, leaving fewer units over which to spread the rising fixed cost of each leading-edge node. Twelve years later, the delayed consequence has become a capital-allocation decision.
The timing lesson matters for investors. Thompson could see the structural failure far in advance, yet “if I had shorted Intel, I would’ve been both right and bankrupt”; the more investable expression was owning TSMC, the manufacturer that captured mobile volume.
He preserves some uncertainty around Tan’s statement: it could be a negotiating tactic, a genuine return threshold, or evidence Tan always wanted to exit manufacturing. From a shareholder perspective, becoming fabless and outsourcing like AMD remains a coherent path.
2. Intel’s integrated culture is the foundry problem
Thompson returned to his pre-Pat Gelsinger prescription: split Intel’s products from manufacturing. He gave Gelsinger “a run,” then reversed course last fall—“Nope, my initial take was right”—because an integrated Intel will inevitably prioritize itself over external foundry customers.
The complication is volume. Intel products provide a large share of the load its fabs need; a fully independent foundry would require several outside customers, not merely one. Yet prospective partners will not trust a supplier culturally conditioned to put Intel products first.
His categorical call: “Intel’s culture is fundamentally incapable of becoming a foundry.” If domestic leading-edge capacity is truly a national-security requirement, the US government may have to “suck it up and fund it,” potentially alongside a split.
Sharp’s pushback was less sentimental: some finality might be healthy. Between uncertain commitment, no outside-customer track record and weak service muscles, clarity about an exit could let customers and policymakers plan around the suppliers that will remain.
3. Intel optimized for performance while TSMC optimized the factory
Tan’s criticism that Intel pursued performance without enough attention to yield sounded to Thompson like the same critique Intel has faced for 30 years. In logic, one defective unique component can ruin an entire chip; memory can disable bad repeated cells, making it a different discipline and risk business.
Intel’s integration and historic margins cushioned mediocre yields. Its many processor variants reflected aggressive binning: defective sections could be switched off and the resulting chip sold into a lower tier—such as a cheap netbook processor—rather than discarded entirely.
The 200-to-300-millimeter wafer transition captured the cultural divergence. Canon and Nikon largely adapted existing equipment, an approach Intel accepted; ASML and TSMC re-engineered lithography around the larger wafer’s physics to preserve throughput and earn more chips per pass.
Tony Fadell’s old objection to the story that Intel merely declined the iPhone order was sharper: Intel “had no chance” because even its ARM-based XScale chips prioritized performance over efficiency. That culture was rational when computers were plugged in and chips became dramatically faster every year; mobile inverted the requirements.
4. Backside power showcases technical brilliance without customer fit
Conventional chips place logic below communications and power, forcing electricity through increasingly dense layers. Backside power separates power from communications, reducing interference and easing design constraints—but it changes where costly failures occur during fabrication.
Intel’s technologically forward method lays down power, then logic, then communications as one integrated structure. If the logic layer fails, however, more completed work must be discarded; the design therefore risks worse yield and higher total cost.
TSMC chose flexibility. Its N2 process for Apple retains conventional power placement, while A16 for Nvidia fabricates a separate power structure and bonds it beneath the chip. The latter is clunky and effectively makes two chips, but each customer pays only for what it needs.
Apple’s phone envelope is roughly 7–8 watts, so backside power offers little benefit. Nvidia’s chips draw 700–800 watts and can absorb a premium when selling for $30,000–$60,000. Intel’s solution is “classic Intel”—innovative and forward-looking—but “not meeting anyone’s needs.”
5. Samsung now threatens to close Intel’s last opening
Tesla’s $16.5 billion agreement makes its AI6 chip the anchor customer for Samsung’s delayed Texas fab at two nanometers. Samsung previously produced Tesla’s seven-nanometer chips, lost AI5 to TSMC after struggling with EUV and Gate-All-Around at three nanometers, and now appears to have crossed that technical hump.
Samsung still carries customer concerns resembling Intel’s, including potential IP leakage and competition with its own products; it previously separated its silicon operation amid that distrust. Its memory heritage is also culturally different from logic, although memory boom-cycle cash can fund difficult process transitions.
Intel’s customer map is punishing. Apple helped finance TSMC’s leading edge; AMD remains loyal to the supplier that “saved their bacon”; Nvidia wants a second source but “hates Intel, like, with a passion”; and Qualcomm has examined Intel most seriously yet so far said no because it does not trust Intel to meet its needs.
Thompson entertained a speculative Broadcom acquisition because Broadcom brings meaningful volume, while acknowledging it may not want the fabs. The harder question is structural: if Samsung becomes TSMC’s credible second source, “is there room for three?”
6. Arizona shows US chip resilience no longer depends on Intel
Thompson has revised his skepticism about TSMC Arizona. The first fab may have been intended to mollify the first Trump administration and initially looked destined to remain generations behind Taiwan, but the effort created its own incentive: once the foundation exists, “you build more.”
TSMC now discusses producing a figure Thompson recalled as roughly 30–40% of its advanced chips in the US over time and adding local R&D, making Arizona more self-sufficient if Taiwan becomes unavailable. That is fundamentally different from an isolated satellite fab frozen on old technology.
The “GigaFab” model co-locates multiple generations and designs equipment for reuse across seven, five and three nanometers. TSMC once depreciated fabs and ran them for 30 years, but leading-edge costs and limited demand for upgrading mundane chips led it toward faster payback and flexible conversion.
Thompson’s synthesis: “TSMC is becoming Intel, and they’re becoming Intel much faster than Intel was able to become TSMC.” Arizona is therefore intended as a continually advancing manufacturing organism, not a satellite that merely copies yesterday’s Taiwanese line.
7. Allied capability is more valuable than a perfect national champion
Samsung’s key R&D and pioneering fabs remain in South Korea, so Thompson would press it to build a broader Texas ecosystem around Tesla’s order. The deal raises the odds because “investments have their own gravity”: a two-nanometer anchor makes adjacent capacity and expertise more economical.
The policy question is whether US-located production must belong to a US-headquartered company. Thompson compared chips with shipbuilding, where Korean and Japanese expertise may be more useful than insisting American companies recreate every capability domestically.
Sharp invoked Nippon Steel: if an allied company wants to invest in and upgrade American mills, rejecting it sacrifices real capacity for formal ownership. Thompson acknowledged Asia’s geographic risks but asked whether policy should pursue “the perfect” while abandoning “the good.”
His bottom line is cautiously optimistic. TSMC’s expansion began under Trump and continued through a bipartisan effort; Samsung now has a plausible US anchor. “We’re in a much better spot than we were,” even if Intel itself is worse and may need to be “broken and reconstituted.”
8. Big Tech may have reached the automobile industry’s settled phase
Thompson revisited his January 2020 essay “The End of the Beginning.” Automobiles progressed from hundreds of startups to four or five scaled manufacturers and eventually three; Tesla was, in practical terms, the first genuinely new car company in roughly a century.
Tech may be following the same pattern. The phone and cloud form a durable computing foundation, while each major incumbent dominates a layer and uses scale and distribution to overwhelm direct challengers. Innovation continues, but increasingly “on top of this foundation.”
Google was always the ambiguous case—not because it lacked AI technology, but because answering questions directly could undermine search. OpenAI itself arose from fear Google would run away with AI, yet Google needed that external threat before turning its transformer research into compelling products.
Sharp kept the disagreement alive: betting that the same five companies dominate consumer technology ten years from now still feels uncomfortable. He said Meta has not created a major new hit in years and that Apple looks “legitimately lost”; he would be open to trying genuinely attractive OpenAI hardware.
9. Google Cloud converts AI leadership into a second growth engine
Thomas Kurian supplied the “Oracle discipline” Google Cloud needed: enterprise reliability and structure arrived just before generative AI created a differentiated reason to choose GCP. Google’s inability to build polished end-user products matters less when customers build those products themselves.
GCP combines Google’s models, networking and old dark-fiber investments with lower AI costs. A company may retain Azure or AWS while placing AI workloads on Google; once one workload arrives, operational gravity can pull more of the enterprise estate with it.
Thompson said the evidence now shows margins, growth and market share moving upward, with OpenAI apparently placing some workloads on GCP. This opportunity is independent of whether AI ultimately damages search, making it an unusually compatible second engine.
Google spent $55 billion on CapEx in 2024, then raised the figure to $75 billion and $85 billion. Thompson did not attribute the increases definitively to the new CFO; he said only that investment had clearly been insufficient and that Google was now being much more aggressive.
10. Incumbent cash makes OpenAI’s narrow path more punishing
GCP is structurally lower-margin than search, so success commits Google to a worse consolidated margin profile. Thompson praised management for accepting that trade-off: “They are investing heavily to get worse margins, which is exactly what they need to be doing.”
If AI is sustaining for incumbents, the financing risk shifts toward OpenAI, Anthropic and xAI. They must fight companies whose model spending is funded by existing profits, while challengers rely on venture capital, debt and increasingly large operating losses.
Thompson suspects ChatGPT subscriptions have positive unit economics if OpenAI stopped R&D—but “does that count if you can’t stop?” Falling behind is existential. Zuckerberg’s talent offensive either takes researchers or raises retention costs, delivering “body blows” financed from Meta’s free cash flow.
OpenAI is nevertheless the fascinating exception because it is “basically taking on everyone.” It might have cemented an Apple alliance rather than hiring Jony Ive, but Altman “wants it all,” pursuing a broad consumer, platform and hardware position.
11. Search and entertainment keep the incumbent outcome unresolved
Thompson repeatedly refused to declare Google Search safe. A 20% decline in traditional query volume or share would leave Google extremely profitable but meaningfully less dominant; Sharp’s household example was that questions once sent to people can now be sent directly to ChatGPT.
YouTube gives Google another underappreciated fortress. Thompson called it the real Netflix threat: not rival subscription streamers, but a platform “devouring everything” with a superior cost structure because creators provide the content.
That leaves multiple credible outcomes. Google may accept a lower parent margin as cloud grows, Meta may convert its distribution into consumer AI, and OpenAI hardware may exploit Apple’s weakness. The individual incumbent bull cases are strong, but Sharp still expects the ground could shift.
12. Grok is more useful as a stalking horse than a standalone winner
Both hosts rejected the premise that they had identified Grok as a likely overall AI winner. Thompson sees a possible niche for a model that is less coy, less flattering and more willing to “say it like it is,” but ChatGPT has already won substantial consumer mindshare.
xAI’s clearer role is as a negotiating alternative. A developer such as Cursor can threaten to substitute Grok when relations with Anthropic deteriorate; that makes xAI valuable to the ecosystem, but being the “bridesmaid” or designated backup is not obviously a durable business model.
The overlooked opportunity is entertainment. AI discourse focuses on productivity, yet “most people just wanna be entertained”; Sharp estimated his own phone use at roughly 80% entertainment. Grok’s willingness to be silly might help, though ChatGPT or Meta could just as easily capture that demand.
13. Doomsday rhetoric sells products and distorts policy
Matt Levine’s “business negging” frame resonated: an assistant that does everything but might ruin your life is psychologically more compelling than a safely competent one. Altman’s recurring pre-launch suggestion that a model may be too dangerous to release functions as extremely effective marketing.
Thompson’s larger objection is political. Anthropic and other doom advocates persuaded Washington to treat AI like a one-time superweapon that could deliver permanent global control, influencing chip, Nvidia and China policy in ways he considers disconnected from observed development.
His base case remains slower and larger: near-term effects will be smaller than advertised, long-term productivity and labor effects enormous, and society will acclimate as the change steadily diffuses. By 2035 or 2037, the world may be transformed without ever experiencing a singular discontinuity.
Policy should multiply potential harm by its probability and ask whether intervention can actually prevent it. Instead, Thompson argues, “people with the largest imaginations got the most influence and control”; doom deadlines moved from 2025 toward 2027 without falsifiable metrics.
14. Tea exposed the gap between iOS security and App Store theater
Sharp summarized Tea as an app where women posted identities and experiences involving men they had dated, followed by reported exposure of thousands of users’ names, selfies, identity documents, locations and later direct messages. Both hosts found the underlying premise dystopian before reaching the breach.
Thompson sharply distinguished iOS from App Store review. Device security comes from sandboxed system architecture that prevents apps from interfering with one another; it does not prove Apple has validated an app developer’s cloud implementation.
The reported failure involved a publicly accessible S3 URL, not someone breaking through an iPhone sandbox. Apple could attempt SOC 2-like cloud verification as polished “vibe-code slop” proliferates, but that would be a new obligation—not evidence its current screening ever delivered the protection it markets.
KYC rules compound the risk by conditioning lawful users to continually upload identity documents and verify themselves. Thompson conceded legitimate reasons to identify customers, but stressed the trade-off: pursuing money launderers and other bad actors can force everyone else to create highly sensitive, breachable records.
15. The Ringer survived because podcast economics subsidize the writing
Thompson’s blunt answer was that ESPN stopped funding Grantland while Spotify continues funding The Ringer. The deeper difference is format: Grantland was an extraordinary website with podcasts attached; The Ringer is a podcast-first business whose website supports the network.
Sharp estimated that five or six successful shows can finance ambitious written work because podcasts monetize far better than digital print. Bill Simmons also cares about writing and has enough leverage from his profitable audio portfolio to keep the site alive.
Thompson’s multimodal model assigns each medium a job: text is shareable and effective for customer acquisition but weak at monetization; podcasts capture attention and monetize well but are difficult to spread. Writers who can build through text and transition into audio connect the two.
The governing lesson is that “business models matter even more than product sometimes.” Grantland was a magnificent product that lost money—“the Kurt Cobain of websites,” immortal because it ended early—while The Ringer’s less revered product has durable financing.
16. A $9.99 backup became the purest test of paying for services
Thompson’s practical recommendation began with keys: use a real, strong carabiner, give every key its own ring, and remove individual keys without dismantling the whole system. It was his specimen for having “thought too much” about nearly every mundane choice.
The sharper argument concerned Sharp moving photos to an external drive rather than paying Apple $9.99 monthly for two terabytes. Thompson’s categorical response: “An external hard drive is not a backup”; it merely transfers irreplaceable data into another single device that can fail, disappear, be thrown away or burn.
Thompson framed cloud storage as insurance, not a “tax for living.” Childhood photos cannot be recreated, so automatic redundancy is worth the recurring fee—especially through a family plan. Sharp remained dug in, although his wife already pays for the storage, perfectly illustrating their productive contrarian friction.