The Pan Am–Boeing playbook and what Musk’s Terafab order could mean for Intel Foundry

“We either build the Terafab or we don’t have the chips.” That’s Elon Musk, speaking to Reuters, stating a supply constraint as plainly as anyone has stated one. TSMC is sold out. Samsung is committed. The existing supply chain can’t expand fast enough to meet what his companies will need for AI, robotics, and space. So he went looking for a different kind of supplier — one with capacity, knowhow, and no queue.

Pan Am’s Juan Trippe found himself in the same position in the mid-1960s with a seat shortage. Existing aircraft couldn’t move enough people at a price that made mass air travel possible. Boeing was hemorrhaging money chasing Concorde, consumed by the prestige race for supersonic speed. Trippe didn’t ask for the fastest plane, but one with enough capacity to move the most people, to the destinations they dreamed of but couldn’t afford.

Trippe walked into Boeing with a vision of a world that had never flown before — and left with Boeing committed to building something beyond their own comprehension. Allen built a factory larger than anything Boeing had ever constructed. Trippe ordered planes Pan Am couldn’t fully pay for. The supersonic program died. The 747 reshaped global travel for fifty years.

What made both men effective was their audacity — the alpha instinct to get out in front of the pack and identify where the big kill is before anyone else sees it. Trippe walked into Boeing with a betcha neither company could afford to lose. In his mind the 747 with Pan Am’s blue and white livery was already crossing every ocean, connecting cities worldwide, carrying people who had never flown.

Musk walks into Intel with the same completed picture — a conveyor belt of silicon feeding his robots, his cars, his satellites, a future he can already see in full resolution even if no one else yet can. He walked in as a partner proposing a betcha Intel can’t afford to lose alone. Both pairs of men at the bridge’s edge — cord untested, betting it holds.

In mid-April, Intel CEO Lip-Bu Tan sent a memo to his staff — two days after Intel announced its Terafab involvement via a 60-word post on X with no press release published to its own website. “Musk’s expansive vision across AI, transportation, communications, robotics and space travel relies heavily on an ample and uninterrupted supply of silicon chips,” Tan wrote of Musk. “Intel is thus a natural partner to help him realize his vision.” He noted that he and Musk had held “wide-ranging and deep conversations,” from which “both sides quickly realized that working together would be mutually beneficial.” That’s the Allen moment — the handshake that commits the company before the engineering is fully proven.

The prestige race is with TSMC. And like Boeing’s pursuit of supersonic transport, it has been expensive, consuming, and largely beside the point of what Intel actually needs to survive.

TSMC built its dominance not by inventing new physics but by running sophisticated machines with greater discipline, consistency, and yield than anyone else. Five companies supply those machines and collectively control the world supply of leading-edge chip production. ASML holds a monopoly on EUV lithography — the only process capable of printing patterns that define a leading-edge transistor, with no alternative anywhere on earth. Applied Materials encodes deposition and materials engineering. Lam Research owns plasma etch. KLA dominates inspection and metrology. Tokyo Electron controls thermal processing and the track systems every wafer passes through before lithography. One Dutch, one Japanese, three American. Remove any single link from the chain and leading-edge production stops globally. Not slows. Stops.

These companies represent the accumulated process expertise of an entire generation of engineers, encoded into capital equipment over decades. The lithography knowledge that once lived in the minds of physicists at Philips now lives in a $400 million machine. The etch knowledge that took Lam’s engineers thirty years to develop runs in software on a plasma chamber. The inspection intuition KLA’s founders built from first principles is now a metrology system examining wafers at resolutions the human eye cannot approach.

The encoding worked — up to a point. What it captured was the physics, the measurement, the repeatable process. What it couldn’t capture was the judgment — what to do when you find something unexpected in a combination of circumstances that has never occurred before in exactly that way. That knowledge lives in people. It always has.

ASML and TSMC didn’t arrive at dominance separately. They built each other. TSMC needed a lithography partner willing to develop equipment around a pure-play foundry model. ASML needed a customer whose volume commitments made the economics of EUV development viable — a technology that took twenty years and billions of dollars to become real. TSMC’s process discipline gave ASML a proving ground. ASML’s roadmap gave TSMC the node leadership no integrated device manufacturer could match.

They didn’t plan to build a duopoly. They kept showing up for each other when the technology needed one more generation of commitment to become real. The three American equipment companies — Applied Materials, Lam, KLA — built the tools that made TSMC possible. The Dutch-Taiwanese partnership built the manufacturing culture that knew how to run them. The tools stayed in California. The judgment went to Hsinchu.

Pan Am and Boeing ran the same play a decade earlier. Neither knew they were building something that would reshape global travel. They were solving immediate problems for each other — Pan Am needed seats, Boeing needed a customer large enough to justify the factory. The 747 was the output of that mutual dependency, not a grand vision executed from the top down.

Tan’s memo reads the same way. Not a grand vision — a natural partnership. A customer with an uninterrupted need for silicon, a supplier with allocated capacity and no queue. Tan has assigned his chief of staff and interim CTO, Pushkar Ranade — an 18-year Intel veteran — to manage the engagement directly, with Tan overseeing it personally. “I have asked Pushkar to assemble and engage select technologists across the company to contribute to this project,” Tan wrote. That is not a business development assignment. That is the company’s institutional memory being pointed at the problem.

What makes the situation urgent is that the judgment Hsinchu accumulated is exactly what Hillsboro is now losing.

The process tech laid off at National Semiconductor in the 1980s whose departure preceded an immediate yield collapse is the oldest story in semiconductor manufacturing — management cutting what it couldn’t measure, losing what it couldn’t replace. Intel’s layoffs are running the same script right now, with engineers who ran 14nm and Intel 4 through production ramps walking out the door carrying knowledge no machine has recorded.

The fab managers executing those layoffs made locally rational decisions. So did every executive who offshored assembly, every investor who rewarded the fabless model, every university that defunded process engineering programs because students wanted to study AI. Each decision looked good on its own terms. The aggregate is TSMC’s Arizona operation running identical machines to Taiwan at lower yields — because the machines transferred and the judgment didn’t.

You cannot allocate budget for the intuition that forms when enough people who know what they are doing work in close enough proximity for long enough that the knowledge stops being individual and becomes environmental. The CHIPS Act funded the machines. It didn’t fund the community. AI is making this worse in a way the policy apparatus hasn’t fully reckoned with.

AI is doing to semiconductor process engineering education what offshoring did to semiconductor manufacturing employment — making it invisible as a career path at exactly the moment it becomes strategically critical. A Stanford graduate who could become a process integration engineer at Hillsboro is instead becoming a machine learning researcher at a billion-dollar startup. The pay, the status, and the trajectory all point the same direction, and it isn’t toward a fab.

AI runs on chips. The foundation models, inference engines, training clusters — all of it requires leading-edge silicon that requires the process engineering expertise the AI industry is pulling talent away from. The demand that makes AI possible is eroding the supply chain that makes AI possible. The next generation of potential process engineers is being pulled into AI’s orbit before they ever acquire the knowhow that needed recording. The reservoir isn’t just draining from the top. It is failing to refill from the bottom.

Musk isn’t requesting Intel’s most sophisticated capability. Tesla’s inference chips for Optimus and FSD are demanding but structurally simpler than the hyperscaler XPUs that exposed Intel’s yield limitations with Broadcom. Where Broadcom throws fastballs — face-to-face stacking, “Correct By Construction” yield requirements, zero tolerance for first-pass failure — Musk lobs softballs. Single-die inference chips, EMIB packaging Intel already owns, volumes that ramp gradually. Intel can hit that. Every wafer it runs for Tesla is a yield learning cycle on 18A — batting practice that makes the fastball more hittable the next time Broadcom steps onto the mound.

Intel also has something Terafab couldn’t acquire independently: ASML machines already installed, allocated, and running on a node with no external customer queue. Musk doesn’t need to build a fab. He needs to fill one that already exists and is currently running below its potential. TSMC’s allocation is spoken for years out. Samsung is committed. There is no line to cut at either foundry. Intel is sitting on capacity the rest of the industry can’t offer.

Pan Am’s 747 order didn’t make Boeing immediately competitive with Concorde. It kept Boeing solvent and learning while the prestige race burned money in the background. Musk’s inference chip orders may do the same for Intel Foundry — keeping the organization alive, building process confidence, absorbing displaced Hillsboro talent into a program with real volume and real deadlines before that talent disperses entirely.

The skepticism is warranted. Brad Gastwirth, global head of research at supply chain firm Circular Technology, noted last week that while “the ambition implied is significant,” visibility into execution remains limited. “There is no defined timeline to high volume manufacturing, no disclosure around capital intensity or cost per wafer, and no guidance on yield ramp expectations — which are critical given how sensitive advanced node production remains.” Those gaps are real. Tan’s promise to disclose the scope in coming weeks will either close them or widen them.

The warning the Pan Am analogy carries is the one Boeing learned the hard way. Pan Am pushed faster than the engineering was ready, contributing to early reliability problems that nearly killed the 747 before it found its footing. Musk’s timelines carry the same risk. The customer who saves you can also be the customer who breaks you if the pressure outruns the capability. Tweets never rescued a process node that couldn’t sustain yield.

The machines are extraordinary. The equipment suppliers encoded more process expertise into capital than any industry in history. But yield still lives in people. And the people are leaving. The window is closing —one career change at a time, the way it was lost.

Tan’s memo said Intel is a natural partner. The 747 flew. Whether Intel’s version does is the question the next few years will answer.

Also Read:

Disaggregating LLM Inference: Inside the SambaNova Intel Heterogeneous Compute Blueprint

Who’s Buying America’s Foundry Future?

Intel, Musk, and the Tweet That Launched a 1000 Ships on a Becalmed Sea