Intel CEO Lip-Bu Tan used his first long-form appearance on the No Priors podcast to describe a sweeping rebuilding plan for the company. His stated return target is “10x in 5 to 10 years,” and he said Intel has already generated roughly a 6x return for shareholders over the past 14 months. In his words, however, “this is just the beginning.” The plan he laid out extends beyond Intel’s traditional PC client base and into data centers, edge computing, physical AI and agentic AI.
The interview, hosted by No Priors with Allad, introduced Tan as the legendary investor from Walden, the former CEO of Cadence and the current CEO of Intel. Asked why he accepted what the host called a difficult job at a critically important American semiconductor company, Tan said he is 66 and that many people told him he should retire. He gave two reasons for taking the role: Intel is an iconic company that is highly important to the semiconductor ecosystem and to the United States, and after Cadence he decided to take on one more big mission.
From “crawl-walk-run” to balance-sheet repair
Tan framed Intel’s turnaround through his familiar “crawl-walk-run” model: stay humble, listen to customers, then move faster step by step. He also recalled the most unexpected episode of the past year. One morning, President Trump asked him to resign, saying there was a conflict of interest and “no exception.” Tan said he first put personal emotion aside and reminded himself that he did not need the job and was doing it to save Intel. He then secured a meeting on Thursday morning and another on Monday, during which he explained that he was born in Malaysia, grew up in Singapore, graduated from MIT and had lived in the United States ever since. Trump listened and gave him the opportunity to continue, Tan said.
Inside Intel, Tan said the early work focused on changing the culture, clarifying accountability and speeding up decision-making. He described Intel as having layers of bureaucratic meetings, while he is used to the pace of a startup. From day one, he decided that all engineering teams would report directly to him. As an engineer, he said, he wanted to know personally where problems existed and what needed to be corrected. The other priorities were listening to customers, satisfying them, making sure Intel had the right products, simplifying the product line and setting a clear roadmap and vision for the next five to ten years.
He also described the balance sheet as being in poor condition when he arrived. Tan said he was pleased that the U.S. government became a major shareholder, and that he explained to Trump that countries such as Japan and Singapore treat semiconductor manufacturing capacity as infrastructure that deserves government support. He thanked Jensen Huang, who invested $5 billion in Intel; Tan said that investment has grown to $25 billion or more. SoftBank’s Masayoshi Son, with whom Tan had worked while serving on SoftBank’s board, also helped. These moves, he said, stabilized Intel’s balance sheet.
CPU demand, foundry discipline and Terafab with Elon Musk
Tan said agentic AI and inference are driving a strong rebound in CPU demand. In earlier training workloads, he said the CPU-to-GPU ratio was roughly one to eight; now he is seeing it move toward one to four or even lower. Some AI model developers told him that CPUs perform better in reinforcement learning and in coordinating the speed of all agents. After strengthening the data-center server product line, Tan said the other major business is Intel’s foundry operation.
On foundry, Tan acknowledged that it is capital intensive and difficult. He nevertheless chose to continue because advanced manufacturing in the United States has strategic value for supply-chain security. No large semiconductor company, he argued, should concentrate its supply chain in one or two geographic regions. At the execution level, he is focused on yield, defect density and cycle time. Foundry, he said, is a service business and a trust business: “Before customers hand you their wafers, they must trust you.” If yield is not good enough, customers lose revenue and leave, and winning them back becomes very hard.
Tan also said Intel and TSMC are partners, not merely competitors, and that the industry needs more capacity to meet growing customer demand. He expects the real potential of Intel’s foundry business to become visible in the 2030 to 2032 timeframe. The gap with TSMC remains large, he said, so Intel has to stay humble and build foundations such as IP, yield, defect density and cycle time.
Another major initiative discussed in the interview was Terafab, Intel’s collaboration with Elon Musk. Tan said both he and Musk share the view that semiconductor infrastructure has not kept pace with AI growth in capacity, production efficiency and power efficiency. Under the collaboration, Musk decided to build his own fab, while Intel will contribute technology and process support to help accelerate production. Tan said he meets Musk’s team every week and described the cooperation as exciting. Musk, he added, challenges traditional practices and once discussed allowing smoking in certain areas of a clean room. Tan said he would not go that far, though some areas might be considered; the key is to keep an open mind.
Advanced packaging, glass substrates and synthetic diamond
As traditional process scaling approaches physical boundaries, Tan is turning to materials science and advanced packaging. He said Intel now has 18A and is pushing 14A into production, while he can see a path to 10 nanometers and 7 nanometers. That path, however, will become more expensive and more difficult. This is why, in his view, Intel must work closely with substrate suppliers and equipment makers to improve yield and performance.
Advanced packaging is another bottleneck. TSMC has CoWoS, and Intel is pushing its next-generation packaging technology, EMIB. Tan said he must make sure EMIB reaches the yield that customers require in mass production. In packaging materials, he invested in 3DGS, a glass-substrate company, because glass has useful properties as a heat-dissipating insulating material. Intel has roughly 1,000 patents in modules, and integrating substrates with modules is one of the core engineering questions he emphasized. Intel has also announced advanced-packaging manufacturing collaboration projects in India and New Mexico in the United States.
Tan also discussed new semiconductor materials. He said he has invested in gallium nitride, silicon carbide and indium phosphide, and that some of those portfolio companies have been acquired by large semiconductor companies such as ADI. He has also invested in a synthetic diamond wafer company, seeing diamond as another strong insulating material for chip packaging. “That is the spirit of engineers,” he said. “You keep encountering bottlenecks, and then you find a way to cross them or go around them.”
When asked whether process-node convergence could flatten the performance differences among foundries, Tan returned to the essence of Moore’s Law: transistor density doubles, but power and cost do not fall at the same rate. Performance can double, he said, but area and cost do not necessarily decline equally unless new materials and new design methods are found. This is why he is hiring more materials-science talent, which he called central to innovation in the field.
AI changes semiconductor investing, design and teams
Tan said AI’s impact on the global semiconductor supply chain will be greater than the internet’s. AI can help people complete work more efficiently and can improve timing optimization, speed to market and costs in semiconductor design. At the same time, AI demand faces bottlenecks: power constraints in some countries, the impact of helium on the semiconductor industry, and memory shortages. Even if companies expand capacity now, new capacity takes years to come online, and CPUs and GPUs are also in short supply, pushing prices and costs up toward the client.
As an investor and operator, Tan described his track record: 159 IPOs, 126 merger-and-acquisition exits, more than 200 semiconductor investments and 38% of those investments in the United States. His framework begins with a question: where is the bottleneck, and what problem is being solved? He invested in Cradle Semiconductor because interconnect became a bottleneck, and in Celestial AI because optical interconnect is becoming increasingly important inside clusters. He said Jensen Huang has invested in nearly every photonics-related company, and he does not view that as a coincidence.
Tan also pointed to EDA as a large opportunity, asking whether AI and machine learning can reduce complexity and improve design quality. In materials, gallium nitride, silicon carbide and indium phosphide remain investment directions for him, with some companies already acquired by ADI and other large players. Power management is another bottleneck he highlighted, especially the large loss involved in converting from 40V to 1V. For early customers, he prefers hyperscalers that have the ability and willingness to pay millions over several years, and once one large customer is secured, scale becomes easier.
He also named Silicon Valley, Austin and Israel as areas of focus for talent. He praised Israeli entrepreneurs as disruptive, hardworking and resilient, even during wartime, when they would pause a meeting for an alarm, move to a basement and continue by voice if the network was weak. Beyond agentic AI, he described physical AI as the next major frontier and said he remains deeply involved in frontier-model-related investments and open frontier technologies for physical AI.
Asked about Cadence and AI in chip design and testing, Tan said he spent nearly 15 years at Cadence and that one of the achievements he is most proud of was identifying and training his successor. That successor, now CEO, is actively embracing AI and bringing agentic AI into tools to raise efficiency. Synopsys’s Sassine is doing similar work, supported by Nvidia’s $2 billion investment, and Synopsys has acquired Ansys to expand toward full-system design. Tan said large companies are moving, but startups still have room to do more disruptive work and may ultimately go public or be acquired by the two major companies.
On organizational change, Tan said Intel is still in the “crawl” phase. He has recruited top semiconductor talent and is now thinking about the software talent required to build full-stack capabilities. He also noted that the team’s average age is in the 40s to 50s, so he wants to bring in younger people who understand workloads and frontier open-source models. He said his son has become his teacher on AI and machine learning, and that he tries to turn what he learns into investment judgment and hiring decisions. Intel, he said, has long been an old-school, spreadsheet-dependent company, and he is working to turn it into an AI-enabled enterprise across the organization, including design, not only sales and marketing.
Capital, investor misconceptions and where compute will live
For capital-intensive businesses and infrastructure projects, Tan said access to capital is critical. Some venture capital firms are now willing to put $1 billion into a single company, which he said was unimaginable before. Early-stage strategy therefore requires either entering very early, when valuations are still reasonable, or investing at Series A, although Series A valuations are now above $1 billion and therefore difficult. For AI factories and foundries, he said government funding, sovereign wealth funds and large infrastructure funds are necessary, and sovereign and government capital will become increasingly important.
As a public-company CEO, Tan said he is intentionally focusing on investors with a longer-term growth orientation, rather than short-term capital that asks every quarter when the company will repurchase shares. Shareholder return is a legitimate concern, he said, but he also has to build the business, and balancing the two matters.
Asked what investors misunderstand most about Intel, Tan returned to the “crawl-walk-run” framework. Over the past few months, he said, Intel has still been crawling, although people have started to see the potential. In products, the PC client business still has share, but performance must improve sharply. He is quietly building CPU architecture, GPU architecture and software architecture teams to prepare for leapfrog leadership, moving with the speed of a large startup.
Tan said Intel’s foundry business still trails TSMC by a large margin and must stay humble, with attention on IP, yield, defect density and cycle time. These efforts take longer, but he believes people will begin to see Intel’s true potential around 2030 to 2032. The PC client business is the base, but Intel is extending into edge, physical AI and agentic AI. In the past, servers and PCs served humans; now millions of agents need access to compute and software stacks.
He compared the Intel target with his experience at Cadence. From interim CEO through retirement, Cadence’s share price rose from $2.4 and generated about a 76x shareholder return; by the end of his time as executive chairman, the return was about 85x. Intel is larger and harder to replicate, he said, but his goal remains 10x in 5 to 10 years. The direction he described is to combine Intel’s XPU, advanced packaging and foundry capabilities to build customized chips for different workloads.
On the future distribution of compute, Tan said today’s large-scale AI infrastructure buildout is correct and that he sees no reason for it to slow because workloads continue to grow. Current constraints, in his view, are mainly on the supply side rather than the demand side. After infrastructure is built, however, the important question becomes which applications will run on it at scale. He compared this with the internet era, when Amazon and Netflix emerged while other applications disappeared or were acquired. AI, he said, will also see growth followed by consolidation, with one or two real winners emerging.
Some applications are better suited to edge or client computing, Tan added, including robotics and defense. In those settings, the choice of on-device compute, assumptions about connectivity and assumptions about built-in device capability determine what can be done. His investment method remains to identify a real problem, find the right partners and evaluate whether the application’s market size is sustainable. If he truly believes in it, he said, he doubles or triples down, including on applications that have not yet reached large-scale deployment.