66-Year-Old Lip-Bu Tan Has Run Intel for 14 Months — and the Stock Is Up 6x

Source: No Priors | Published: 2026-06-18T10:00:05Z

Tan stabilized Intel's balance sheet with a 'crawl-walk-run' playbook. Jensen Huang's $5B bet has ballooned to $25B, and agentic AI is shifting the CPU-to-GPU ratio from 1:8 toward 1:4.

Born in Malaysia, raised in Singapore, and MIT-educated, this semiconductor veteran's career spans EDA, chip design, and venture capital. He spent 14 years at Cadence (12 as CEO, 2 as executive chairman), delivering over 30x returns to shareholders, while simultaneously founding Walden International and backing more than 140 IPOs. Now he's bringing startup speed to a chip giant with over 50,000 employees.

The Morning Trump Asked Him to Resign

The biggest surprise after Tan took office wasn't a technical problem or a market challenge — it was a phone call from the President. Trump cited conflicts of interest and demanded his resignation as CEO. No exceptions.

He talked himself through it first: I don't need this job. I'm here purely to save Intel. Once he set aside personal stakes, he secured a chance to explain in person. He got the call Thursday morning and was sitting in front of the President by Monday, telling his story — born in Malaysia, raised in Singapore, studied at MIT, never left America since. Trump listened and gave him the green light to continue.

"Crawl, Walk, Then Run"

Tan's go-to management philosophy is "crawl, walk, run" — stabilize, build momentum, then sprint. His first move at Intel was shoring up the balance sheet, because "the balance sheet was really in bad shape."

He lined up three key backers: the U.S. government became a major shareholder — he explained to Trump that TSMC had the Taiwanese government as a shareholder at its founding, and Japan and Singapore's semiconductor industries worked the same way. This is infrastructure; government support is natural. His old friend Jensen Huang put in $5 billion, now worth $25 billion. And SoftBank's Masayoshi Son — Tan had served on SoftBank's board — stepped in as well.

With the balance sheet stabilized, he did two things: slashed product lines and had all engineering teams report directly to him. As an engineer-turned-CEO, he wanted to personally understand where things had gone wrong.

CPUs Staging an Unexpected Comeback: Agentic AI Is Shifting the Compute Mix

A trend that has Tan excited is unfolding. During training, the CPU-to-GPU ratio used to be roughly 1:8. He's now seeing it shift toward 1:4 or even lower.

The reason is agentic AI. After talking to multiple AI model developers, the feedback is clear: in reinforcement learning scenarios and when orchestrating large numbers of agents at speed, CPUs actually perform better. Servers and PCs used to serve humans. Now there's an entirely new dimension — millions of agents need access to compute and software stacks. This opens a market for Intel's CPU business that simply didn't exist before.

Co-Building Terafab with Musk

Musk decided to build his own wafer fab — that's Terafab. Intel's role is providing technology and process support to help him reach volume production faster. Tan meets with Musk's team weekly.

"He questions every single step and asks why things are done the traditional way. It's incredibly refreshing."

As for the rumor that Musk wants to smoke in the cleanroom, Tan laughed and said "I wouldn't go that far," but added he keeps an open mind and seriously evaluates which conventions can actually be broken. The logic is straightforward: Musk's robots and vehicles need massive quantities of chips, and both he and Tan share the conviction that semiconductor capacity hasn't kept pace with AI's growth.

Foundry: A Trust Business

There's plenty of skepticism about whether Intel should keep investing in its foundry business — too expensive, can't be done. Tan's view: it's critical for America and the entire industry.

He defines foundry as a "trust business": customers hand you their wafers, and if yields are poor, their revenue suffers. So the core metrics are yield, defect density, and cycle time. Intel's 18A process is progressing, 14A (equivalent to 1.4nm) is ramping up, and 1nm and 0.7nm are already on the roadmap.

But process technology alone isn't enough. A complete foundry service needs a robust IP library — mobile chips, for instance, require low-power IP. Ultimately, it has to become full-stack: not just silicon, but software too, with some customers outright demanding "just give me the whole rack."

Moore's Law, Next Chapter: New Materials and Advanced Packaging

As process nodes shrink, physical limits loom closer. Tan's strategy runs on two parallel tracks.

The first is new materials. He's betting on three simultaneously: gallium nitride, silicon carbide, and indium phosphide — he invested in Inphi (later acquired) and a power management company (acquired by ADI). Power conversion is a real bottleneck: stepping down from 48 volts to 1 volt wastes enormous amounts of energy.

The second is advanced packaging. Beyond TSMC's CoWoS, Intel has its own next-gen packaging technology, EMIB. More frontier explorations include glass substrates — glass is an excellent thermal insulator, and he invested in a company called 3DGS — as well as synthetic diamond, through his investment in Diamond Foundry. Intel holds over 1,000 patents in substrates and modules and just announced a partnership with the Indian government to manufacture advanced packaging in India and New Mexico.

"The beauty of being an engineer is that you always hit a wall, and then you figure out how to climb over it or go around it."

From Nobody Wanting to Invest to Everyone Piling In: Semiconductor Investment Comes Full Circle

Eighteen years ago, when Tan pitched semiconductor deals to top VCs, half the partners would leave mid-meeting. The ones who stayed asked, "Got any software SaaS deals?" The stragglers stuck around out of pity.

Now Huang's NVIDIA is worth $5.3 trillion. Broadcom and TSMC are at $2 trillion each. Lisa Su's AMD is approaching $800 billion. Intel itself is nearing $600 billion. VCs are flooding into semiconductors.

But Tan's investment methodology hasn't changed just because the sector got hot. He always focuses on three questions: Where's the bottleneck? Are customers screaming about this problem? Who's the first target customer? He favors hyperscaler customers — if they love your product and are willing to pay millions, it's worth giving them warrants, because one major customer lets you scale.

He specifically mentioned Israeli entrepreneurs. Even during wartime, they show up to calls on schedule. Sometimes they say, "Sirens going off, heading to the shelter, connectivity might be rough — let's switch to audio" — and then keep going. That resilience left a deep impression.

Nine Out of Ten Companies Changed Their Business Plan Midstream

Looking back on his investing career, Tan shared a statistic: nine out of ten companies he backed changed their business plan along the way, because the market shifted.

So he values several things: the founding team, not a solo founder; open-mindedness and willingness to take feedback; the ideal scenario is you give enough input that the founders reach a conclusion on their own — and that conclusion happens to be one you agree with, or at least a different direction you can get behind. He also values co-investors who stick around through adversity — plenty of people are happy in good times but vanish when things go south. He looks for partners who've been through multiple "nearly bankrupt but eventually took off" cycles.

Applications Decide Everything — Building Infrastructure Doesn't Mean Winning

The current massive buildout of AI infrastructure is the right move, and Tan believes it won't slow down in the near term. The only brakes are supply-side constraints: power, helium (most people don't realize helium's impact on semiconductors), memory shortages, and CPU/GPU production capacity.

But he has a clear-eyed view on who ultimately wins: not everyone building infrastructure will win. It depends on how large the applications you serve are, whether they're sustainable, and whether the space is too crowded. It's the same as the internet era — Amazon and Netflix were real applications, and they won. Many others either disappeared or got acquired. AI will go through the same explosion, then consolidation, with one or two emerging as the real winners.

His bet for Intel: edge computing and on-device processing will make a comeback. Robots, autonomous vehicles, defense — compute on these devices is non-negotiable. You can't rely entirely on the cloud. This was forgotten for a while during the SaaS era, but physical AI is pulling it back.