Intel's collapse, Taiwan risks and the quantum future: A look back from former CEO Pat Gelsinger
Recently, former Intel CEO Pat Gelsinger had an extremely frank sharing session on the All-In Podcast, revealing the hidden corners behind the decline of a semiconductor monument that once dominated the globe. From a unique position with the legendary "Intel Inside" era, this empire has lost its competitive advantage by giving leadership to financial people instead of technical experts. The consequence was a series of fatal strategic mistakes: from being ousted by Apple, conservatively ceding the outsourcing throne to TSMC, to leaving NVIDIA completely far behind in the race to shape the AI era.
Below is a summary of the brightest points in Pat's sharing, organized for you to easily follow. Not only stopping at dissecting Intel's stumbles in the past, Pat also provides a perspective on the risks that directly threaten the chip supply chain in Taiwan, the true limits of the current AI fever and the turning point where quantum computing will officially redefine the rules of the technology world.
Fatal mistake: Letting "finance people" lead technology companies
Intel used to be a monument led by great technical minds like Andy Grove or Gordon Moore. In the boardroom of the board of directors at that time, 15 out of 20 people were technical doctorates. However, things began to go wrong when the company fell into the hands of "bean counters" (finance and accounting people).
Apple Silicon and Steve Jobs's secret project
The relationship between Intel and Apple during the Centrino architecture chip era was very good, but Steve Jobs was an extremely strict leader. Apple continuously requires Intel to make chips that are smaller in size and more energy efficient. When realizing that Intel could not meet the demand, Apple quietly prepared the ability to design its own silicon. Pat recalls a memory: when Intel offered to help Apple convert the operating system code from PowerPC architecture to x86, Steve Jobs calmly replied that Apple had internally assigned someone to silently compile macOS to the x86 platform during the previous 4 versions. Apple is always prepared to be technologically autonomous instead of being completely dependent on one supplier.
Missed the AI train and the rise of NVIDIA
At its peak, Intel internally used to laugh at Jensen Huang's GPUs, considering them just graphics processing machines for gamers. However, NVIDIA has persistently built the CUDA software platform and optimized the multi-threaded architecture across thousands of graphics processing cores. Step by step, their GPU became perfect until high-performance computing (HPC) researchers realized its huge computing power and applied it to AI. Pat revealed that Intel once had a Larrabee project to use the x86 instruction set to counterbalance GPUs, but this project was killed just a week after he left Intel for the first time, closing the company's opportunity to compete early.
Be conservative with IDM and give the throne to TSMC
TSMC from the beginning had a clear vision: to become the processing powerhouse of the entire industry. Meanwhile, Intel still sticks to the IDM (self-designed, self-manufactured) model. Intel's design tool system (EDA) is closed, does not share standards at all, and they are not interested in opening up processing workshops to third parties. TSMC goes the opposite way, they standardize the process and are open to all designs. As a result, when Pat returned to the CEO position in 2021, TSMC's wafer output was five times larger than Intel's. Currently, the entire semiconductor industry operates under the Foundry model, forcing Intel to pivot strategically in this direction to survive.
Taiwan and the global supply chain bottleneck
The CHIPS Act is having an effect as advanced chip manufacturing capacity in the US has increased from 12% to 18%. However, the world's biggest risk today lies in Taiwan. Citing a Wall Street Journal report, Pat said the island's energy reserves are only enough to last less than 3 weeks. If a blockade scenario occurs, semiconductor factories will lose power. A shutdown fab will take at least 90 days to restart, and the global economic fallout from this disruption will be worse than the Great Recession.
The AI Bubble and the Age of Quantum Computing
Faced with concerns about an AI bubble as companies invest frantically in infrastructure, Pat believes that the natural barrier keeping this development from going too far is the capacity of the national power grid. Data center construction depends entirely on available energy sources. We are at the beginning of an AI development cycle that will last several decades, with the technical goal of making the cost of processing each token 10,000 times cheaper than it is today. Regarding quantum computing, this technology will create practical breakthroughs within this decade (before 2030). In the first phase, it will solve extremely large-scale optimization problems such as logistics, then unlock frontiers in chemistry and biology. Around 2032-2033, quantum computing power will be strong enough to crack current encryption systems.