Former Intel CEO on What Went Wrong, What's Next + Lovable CEO on the Real Promise of Vibe Coding

Summary

• Pat Gelsinger’s Intel postmortem is that the company stopped being run as a technology company, then reinforced that error through capital allocation. The five or six years before his 2001 return sent $100 billion to shareholders while Intel went a decade without building a factory and failed to buy EUV equipment. His rule: billion-dollar technical choices cannot be made “through a spreadsheet.”
• Apple, Nvidia, and TSMC each beat Intel through patient capability-building rather than one miraculous bet. Steve Jobs quietly kept Apple’s operating system ready for x86 across four releases before integrating silicon and system design; Nvidia compounded CUDA until GPUs escaped graphics; TSMC standardized foundry access until it produced 5× Intel’s wafers in 2001 and roughly 7× now. Apple’s logic was not “you failed as a supplier,” but “I can supply myself better.”
• Semiconductor resilience is improving, but Taiwan’s energy dependence leaves the global economy exposed to a blockade without a shot being fired. Gelsinger put U.S. leading-edge production at roughly 12% when the CHIPS Act began and 18% today, yet said Taiwan holds under three weeks of energy reserves and a shut fab takes 90 days to restart. A Taiwan brownout, he argued, would have an economic impact “greater than the Great Depression.”
• Gelsinger sees AI as a multi-decade buildout whose natural cap is electricity, not demand for intelligence. Energy availability prevents unlimited speculative data-center construction, while the goal should be AI that is 10,000× better, cutting token cost and energy by five orders of magnitude so Jevons’ paradox expands usage. He expects “a couple of decades” of progress—but not a smooth curve.
• High AI multiples may correct repeatedly without invalidating the underlying thesis, because these businesses already have real revenue and margins. Gelsinger welcomed periodic corrections and further “apocalypses” as safeguards against excess, then extended the opportunity into a “trinity of computing”: classical, AI, and quantum. He predicts meaningful quantum results before 2030, with encryption potentially solved around 2032–33.
• Lovable’s numbers suggest vibe coding has crossed from prototyping into production and business operations. After 20 months it reported more than 50 million apps, one million new projects weekly, 700 million monthly application visits, and fastest growth in enterprise; Anton Osika also corrected Jason Calacanis’s $400 million revenue estimate with “We reached 500 in May.” Jason’s internal example compressed a formerly $500,000 intranet into roughly four to eight hours and under $2,000 in a year.
• Lovable’s defensibility is shifting above any single foundation model toward orchestration, operational data, security, and accumulated feedback. It routes work among commercial frontier and open-weight models, post-trains on high-impact failures, and refuses cheaper intelligence when measurably worse for customers. Osika’s limiting factor is increasingly human judgment: models can produce sophisticated software immediately, but deciding “what is the right thing to build” improves more slowly.

Deep dive

1. Intel’s spreadsheet culture starved its technical flywheel

• Gelsinger joined Intel at 18—“I went through puberty at Intel”—under deeply technical leaders including Andy Grove, Gordon Moore, and Bob Noyce. On his first executive staff, probably 15 of 20 attendees held PhDs; that technical density shaped whom Intel recruited, promoted, and trusted with consequential decisions.

• His diagnosis of the derailment: business leaders replaced technologists, promoted more business leaders, and gradually hollowed out technical authority. Satya Nadella and Sundar Pichai need not be founders to fit his preferred model; what matters is being deeply technical enough to judge investments whose economics look poor before the underlying technology trend becomes obvious.

• Capital allocation made the cultural failure concrete. In the five or six years before Gelsinger returned as CEO in 2001, Intel distributed $100 billion through dividends and buybacks, had not built a new factory in a decade, and had not bought EUV machines. “What I wouldn’t have done for another hundred billion dollars.”

• Jason’s pushback broadened the criticism to Apple’s buybacks and small acquisitions, but Gelsinger kept Intel’s lesson narrower: every leader makes good and bad calls, yet “this is a technology business.” Technologists must run it, hire technologists onto staff, and keep funding capabilities before a spreadsheet can prove their value.

2. Apple, Nvidia, and TSMC compounded small advantages into platforms

• Steve Jobs initially made extraordinary size-and-power demands of Intel’s Centrino chips. Once unconvinced Intel could remain far enough ahead, Apple bought PA Semi and expanded small internal chip efforts gradually; the strategic shift was not an angry supplier rejection but the conclusion, “I can supply myself better,” while optimizing silicon and operating system together.

• Gelsinger’s defining Jobs story came from Apple’s earlier move from PowerPC to Intel. When Intel offered help porting the operating system, Jobs replied that Apple had already ported its previous four releases to x86. Gelsinger was stunned: Jobs had quietly maintained an option for years before the external switch became necessary.

• Intel similarly dismissed Nvidia’s GPUs as niche graphics machines while Jensen Huang steadily improved CUDA, SIMT, and the surrounding software stack. Japanese high-performance-computing researchers then recognized the cards as computationally dense general-purpose devices. Gelsinger’s competing x86 project, Larrabee, was killed one week after his first Intel departure: “The world would have been so much different.”

• TSMC’s equivalent insight was organizational: factories costing $20 billion–$30 billion could serve the whole industry through standardized PDKs, EDA tools, and manufacturing access. Intel’s proprietary IDM system saw foundry work as trivial; by Gelsinger’s 2001 return, TSMC produced 5× Intel’s wafers, and he put the current gap near 7×.

3. Taiwan turns manufacturing scale into a macroeconomic risk

• Gelsinger credited the CHIPS Act with tangible, incomplete progress: U.S. leading-edge production moved from about 12% when the act began to roughly 18% today. Intel is becoming a real foundry, TSMC’s factories are operating at scale, and Samsung is also present—but 18% remains nowhere near resilience.

• The chilling constraint is Taiwan’s under-three-week energy reserve. A blockade that stops oil and LNG could brown out the island without combat; once a fab shuts, Gelsinger said it needs 90 days to return. His estimate for the resulting global economic damage was categorical: “greater than the Great Depression.”

• Jason asked whether the flashpoint comes in 2027, 2030, or 2035; Gelsinger declined false precision because he lacks situation-room intelligence. He said he thought China had blockaded the Taiwan Strait seven times over the last four years, making this more than a theory: supply-chain diversification must become faster and more meaningful.

4. AI’s physical ceiling is energy, while quantum extends the runway

• Asked whether AI infrastructure is a bubble, Gelsinger found reassurance in electricity: companies will not buy GPUs or build data centers without power. He put global energy-capacity expansion around 5%, after a U.S. decade near 1%; that physical upper bound constrains how far spending can outrun deployable capacity.

• Demand could remain vast because, if a token is a measure of intelligence, its potential value is “somewhat infinite” across supply chains, finance, logistics, and labor-constrained economies. Gelsinger therefore expects “not a couple of years, a couple of decades” of buildout, with the objective of making AI 10,000× better and reducing token cost and energy consumption by five orders of magnitude.

• Jason questioned extraordinary valuations, but Gelsinger distinguished today’s companies from dot-com speculation through “real revenues” and “real margins.” He still expects repeated corrections and industry disruptions—including more episodes like the “SaaS apocalypse”—and welcomes them: “Every time we have one of those corrections, say thank you.”

• Jason’s sharpest challenge was that quantum has been five years away for 25 years. Gelsinger answered “this decade”: useful chemistry, biology, and logistics results before 2030, with solving things like encryption probably around 2032–33. He disclosed his PsiQuantum portfolio bias but noted four to six improving modalities, proven error correction, and engineering scale as the remaining race.

5. Lovable has moved from generating apps to operating businesses

• Osika defined two gaps: enabling anyone to build a product, then helping that product become a business. After 20 months, Lovable was producing one million new projects weekly across more than 50 million applications, with 700 million monthly visits to those apps; enterprise was its fastest-growing segment.

• About 20% of users are technical and four out of five are non-technical. Engineers value Lovable’s opinionated architecture, payment setup, continuous security scans, and monitoring; non-engineers use the same structure to discover what should be built. Some customers now run businesses generating more than $1 million on the platform.

• Jason’s Founder University team built an intranet independently in four to eight hours, then added an economic-impact model covering employment, taxes, housing, and salaries. He compared the result with a $500,000 build and estimated total cost below $2,000 in a year; Lovable starts at $25, with the business plan discussed at $50.

• Security was Jason’s initial concern, and Lovable’s team reviewed the deployment; Osika now wants penetration testers comparing competing tools. He emphasized that even free users receive background security scanning. Meanwhile, Lovable’s hosting product is growing faster than app creation, and the company is working with AWS and Red Hat.

6. The software moat moves from code generation to context and judgment

• Osika’s next product is an “AI co-founder” with access to a company’s applications and operating data when customers run those apps and tools on the platform. It could work overnight, then propose strategic directions, growth optimizations, or better customer service. Jason compressed the proposition neatly: customers “come and build the software, but you stay to build the business.”

• Bespoke software will sometimes replace SaaS, but Osika expects coexistence. At Nursa, an employee built a nurse-education product plus scheduling, licensing, certification, and administrative tools, then replaced more than 10 internal tools, saving over $1 million annually. Elsewhere, Lovable can retain Salesforce, HubSpot, Google, Microsoft, or Slack underneath a custom interface.

• People have repeatedly declared Lovable dead with each new frontier-model release; Osika said, in that revenue context, “We reached 500 in May.” Lovable routes tasks across multiple commercial and open-weight models, while a Stockholm research team focuses on post-training and uses reinforcement learning on failures with the greatest customer impact.

• Jason inferred an all-in open-source strategy; Osika corrected him toward a portfolio approach based on speed, cost, and measured customer outcomes. Lovable keeps usage caps and top-ups, but Osika would not substitute a cheaper model when measurably worse. Its million weekly projects supply signals for improving both the agent harness and internal software-building skills.

• Cheap engineering also changes organizational design. Osika endorsed teams independently attacking the same problem, recalling CERN groups that withheld results until publication to avoid a shared local minimum. Lovable can later import the best features and run split tests; with “engineering less of the bottleneck,” the scarce capability becomes choosing the right product and experiment.

• On Anthropic’s “Fable,” Osika saw sophisticated, attractive first-attempt outputs and even 3D games. Yet iteration still requires humans to plan with the agent, provide the right data, and choose strategic direction. Visual and technical generation is accelerating faster than judgment about what will improve a business.