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

Pat Gelsinger argued that Intel lost its way over roughly 15 years of being led by business rather than technical executives, during which the company returned approximately 100 billion dollars to shareholders through dividends and buybacks, failed to build a new factory for a decade, and stopped purchasing EUV machines. He contrasted this with the deeply technical leadership of Andy Grove, Gordon Moore, and Bob Noyce, and argued that founder-led technical companies make better hardware decisions than those driven by spreadsheet analysis.

Intel's decision to pass on making chips for the iPhone was described as a consequential missed opportunity. Steve Jobs had already been quietly porting Apple's operating system to x86 architecture across four releases before formally announcing the switch to Intel chips. Jobs eventually moved Apple toward designing its own silicon not because Intel failed as a supplier but because he wanted to optimize system and silicon design together rather than rely on chips optimized for a Windows environment, building that capability gradually by acquiring small companies including P.A. Semi. Intel also internally dismissed Nvidia's GPUs as gaming hardware at the height of its CPU dominance, and had a competing project called Larabee that attempted to use x86 architecture to replicate what Nvidia was doing with CUDA. That project was killed one week after Gelsinger's first departure from Intel. Nvidia's advantage compounded once it built a real software stack on top of its hardware.

When Gelsinger returned to Intel in 2021, TSMC was producing roughly seven times the wafers of Intel, up from five times the ratio when he first became CEO in 2001. TSMC was founded on a pure foundry vision of serving the entire semiconductor industry, while Intel operated as an integrated design and manufacturing company with highly proprietary processes and no standardized process available to third parties. Apple's role as a customer was identified as a key driver in making TSMC a truly significant foundry operation.

On geopolitical risk, Gelsinger cited a Wall Street Journal article reporting that Taiwan holds less than three weeks of energy reserves. A semiconductor fab that goes offline takes 90 days to restart, and a brownout of Taiwan's fabs would produce an economic impact exceeding the Great Depression without a single military engagement. China has blockaded the Taiwan Straits approximately seven times over the last four years. He argued that progress through the CHIPS Act, which has grown US leading-edge chip production from roughly 12 percent to 18 percent, needs to accelerate. He described the current AI infrastructure buildout as the largest technology buildout he has witnessed, surpassing the PC, server, and internet eras, and identified energy capacity as a natural ceiling on any AI bubble, noting global energy capacity is growing four to five percent annually compared to roughly one percent annual growth in the US grid over the prior decade and a half. He said current AI companies differ from dot-com era speculation because they have real revenues and real margins, though he acknowledged valuations have become extraordinary and periodic corrections are inevitable. He predicted quantum computing will deliver meaningful results before 2030 and that Q-day implications around encryption will emerge around 2032 to 2033.

Anton Osika, CEO of Lovable, reported the company reached 500 million dollars in annualized revenue as of May, with more than 50 million apps built on the platform, one million new projects created every week, and more than 700 million monthly visits to applications built on the platform. Approximately 80 percent of users have no technical or engineering background, some businesses on the platform are generating more than one million dollars in annual revenue, and enterprise is currently the fastest growing segment. Around 60 percent of customers hit their usage caps and top up, which Osika described as a signal of strong demand rather than a pricing problem.

Lovable's strategic direction is evolving from a tool that builds software into one that operates businesses, with a new hosting product line growing faster than the core building product. Hosting gives Lovable access to all company data, enabling a planned co-founder feature that would work overnight and deliver strategic recommendations each morning. A large US nursing company replaced more than ten internal tools with bespoke applications built on Lovable and is saving more than one million dollars per year. Jason Calacanis estimated that a piece of bespoke software that previously cost 500 thousand dollars to build can now be built for between 600 and 6,000 dollars per year on platforms like Lovable.

Lovable uses a routing system that directs queries to whichever frontier or open-weight model is best suited for a given task and has never chosen a cheaper model if it is measurably worse for customers. The company has a research team in Stockholm applying reinforcement learning specifically to problems where frontier models make mistakes on Lovable's platform, feeding improvements back into a system shared with engineers. A pattern emerging among organizations using Lovable is that two people independently build separate projects solving the same internal problem rather than collaborating on one codebase, which participants framed as a form of parallel competition that prevents teams from settling into a local minimum. The CEO and other speakers agreed that engineering is less of the bottleneck now and that the harder question is determining what is the right thing to build.