Intel CEO Pat Gelsinger's Latest Interview: After Missing Mobile, Cloud, and AI, We Can't Miss the Next Wave

Video Title: Intel CEO Lip-Bu Tan on the Comeback of American Chip Industry
Video Author: TechSurge: Deep Tech VC Podcast
Translation: Peggy, BlockBeats

Editor's Note: Against the backdrop of Generative AI driving a new round of computational power investment, industry discussions are shifting from "who has the most powerful chip" to "who can organize a more complete computing system." With GPU demand, advanced processes, and data center capital expenditure now a consensus, a more fundamental question is emerging: what will determine the efficiency of the next stage of AI infrastructure, individual chip performance or the synergy between computation, storage, interconnect, packaging, and manufacturing?

In the Tech Surge podcast under Celesta Capital, host Michael Marks had a conversation with Intel CEO Pat Gelsinger. The discussion ranged from his early semiconductor investment experience to Cadence's transformation, and Intel's product, manufacturing, and platform strategy in the AI era.

In this conversation, Pat Gelsinger not only laid out a product roadmap for Intel's resurgence but also broke down semiconductor competition into a set of more foundational structural issues: why hardware has become a capital focus again, how the AI bottleneck has spilled from chips to infrastructure, whether vertical integration can recreate system value, and how a large company that missed multiple technology waves can regain its perception of cutting-edge changes.

First, the value of semiconductors is shifting from a single product back to a technological foundation. Over the past 20 years, software has received higher valuations and more venture capital, while semiconductors, due to their long R&D cycles, heavy capital investment, and limited exit pathways, were once seen as unsuitable for venture investment. Today, AI turns computing power, power consumption, and bandwidth into direct constraints on application scaling, where chips are no longer just carriers for running software but the infrastructure that determines model costs and commercialization boundaries. This means that the return of hardware is not simply a rotation of valuation but a capital repricing driven by technological bottleneck changes.

Second, AI competition is expanding from a single accelerator to system engineering. The previous AI investment cycle mainly revolved around GPUs and model training. However, as workloads shift to inference, intelligent agents, and edge AI, bottlenecks are starting to spill over to CPUs, storage, high-speed interconnects, advanced packaging, and cooling. The shift from air cooling to liquid cooling and even microfluidic cooling, electrical interconnects to photonics, traditional packaging to glass substrates and new materials all reflect the same change: cluster efficiency can no longer be improved solely by enhancing the performance of individual chips. This also means that the next wave of hardware value may not be concentrated only in top GPU companies but could be distributed across the weak links of the entire computing system.

Third, the key for Intel is to maintain vertical integration, not in terms of how many businesses to retain, but in terms of whether the scattered capabilities can be reorganized into a platform. Previously, the division of design and manufacturing drove the fabless model and the rise of foundries, while Intel simultaneously operated its own products and offered foundry services, increasing the complexity of organization and capital allocation. Chen Liwu still emphasizes the integration of CPU, GPU, software, advanced packaging, and foundry services because product optimization in the AI era increasingly relies on cross-layer collaboration. The potential value lies not only in in-house manufacturing but in co-optimizing architecture, packaging, and process according to customer workloads; the corresponding risk is that if product competitiveness and manufacturing execution cannot be improved simultaneously, vertical integration may continue to escalate costs.

Fourth, the relationship between CPU and storage is being redefined by AI. In the past, CPU has been Intel's most stable core business, while storage has been seen more as a cyclically volatile commodity. As AI shifts from training to inference, general-purpose computing requirements have not disappeared; the CPU still needs to handle tasks such as data processing, job scheduling, and agent operation. Meanwhile, memory bandwidth, capacity, and power consumption have become important constraints on system performance. Chen Liwu mentioned CPU and memory stacking and new storage architectures, indicating that this does not necessarily point to Intel re-entering the traditional storage market but to the need for computing and memory to re-coordinate at the architectural and packaging levels.

Fifth, what Intel truly needs to fix may not be a particular product generation but the organization's ability to acquire external information and respond quickly. Chen Liwu's management style established at Cadence involves directly listening to employees and customers, transforming the company's relationship with customers from a "supplier" to a "partner" who can share roadmaps. For Intel, reconnecting with customers, universities, AI labs, venture capital firms, and startups is also part of restoring this perceptual capability. A large tech company usually misses a technological trend not because it cannot see the new direction at all but because external changes cannot be promptly translated into internal resource allocation and product decisions.

If we were to condense this conversation into one assessment, it would be this: AI hardware competition has shifted from a single-point performance race to a system competition determined by computing, storage, interconnection, packaging, manufacturing, and organizational capabilities. In this sense, the subject of this article is no longer just about whether Intel can complete a corporate transformation but about whether a traditional chip giant can rebuild the ability to participate in the next generation of computing platforms.

The following is the original content (reorganized for ease of reading comprehension):

TL;DR

Chen Liwu believes that semiconductors have once again become the core of the tech industry, and AI competition has expanded from a single chip to encompass packaging, storage, interconnectivity, cooling, and the full software stack.

He brings Cadence's transformation experience to Intel: staying humble, listening to customers, responding quickly, and upgrading customer relationships from 'vendor' to 'partner'.

Intel will continue to maintain its vertical integration model for product design, advanced packaging, and wafer fabrication, creating greater value for customers through platform capabilities.

CPU remains at the core of Intel's competitiveness rebuilding. AI for intelligent edge, inference, edge computing, and physical AI may bring about a new wave of demand.

Patrick Gelsinger revealed that Intel is researching CPU and memory stacking and a new storage architecture, but is not yet ready to announce specific plans.

Having missed the mobile internet, cloud computing, and AI waves, Intel will reconnect with universities, venture capital firms, and startups to avoid falling behind in cutting-edge innovation again.

Key Points of the Interview

Global semiconductor sales are approaching $1 trillion.

In a conversation on the Tech Surge podcast under Celesta Capital, Intel CEO Patrick Gelsinger stated that AI is making hardware the core of the tech industry again. However, this wave of opportunity is no longer limited to GPUs but extends to CPUs, memory, advanced packaging, high-speed interconnects, photonics, and cooling systems.

For Intel, this is not just a product cycle but a rebuilding of platform capabilities.

Gelsinger admitted that Intel missed out on significant waves such as the mobile internet, cloud computing, and AI in the past. Therefore, he has now set a goal for himself: "From now on, I will not miss any more waves of innovation."

Semiconductors Transition from 'Sunset Industry' to AI Innovation Hub

Gelsinger started investing in chips in 1987, with investments in nearly 550 companies. However, for a long time, semiconductors were not favored by venture capital firms.

He recalled that 20 years ago, when visiting a top-tier venture capital firm, the entire partner team would usually be present at the beginning of the meeting. But when he started talking about semiconductors, half of them would politely excuse themselves, leaving only a few people to continue listening with 'sympathy.'

At the end of the meeting, they would often ask him, "Do you have any software or service-oriented startups?"

At that time, mainstream venture capital firms considered the semiconductor industry a sunset industry, with capital continuously shifting towards software and internet services. Even some of Chen Liwu's investors believed that his decision to double down on chips while other institutions were exiting was a nearly crazy choice.

However, Chen Liwu believed that chips were always the underlying foundation of the tech industry. Without chips and platforms with the right performance, power consumption, and cost, many upper-layer applications simply could not be viable.

This judgement also led him to persist as a contrarian investor.

He mentioned that in the past, co-investors asked him, "Can you name a semiconductor company with a market cap exceeding $1 trillion?" Today, this question is no longer valid because semiconductor companies have emerged among the world's most valuable tech firms.

But Chen Liwu's focus is not only on large companies like NVIDIA. In his view, the semiconductor industry is a vast technological ecosystem, and many key innovations come from unnoticed small companies: some are focused on reducing Chiplet power consumption, some are addressing high-speed interconnect issues, and some are betting on photonics, advanced packaging, and new cooling materials.

The real investment opportunities often lie within these underappreciated bottlenecks.

Investing in SambaNova: AI Inference Cannot Solely Depend on High-Power GPUs

The assessment of AI hardware is a microcosm of Chen Liwu's investment approach.

Having previously invested in companies like S3, a graphics chip company, he early on recognized the high power consumption issue of GPUs. He also predicted that as AI transitioned from model training to actual deployment, the market size for inference and edge AI could be much larger than the training market.

Approximately nine to ten years ago, he supported two different computing architecture paths for this.

The first was Cerebras' wafer-scale chip solution. Chen Liwu believed that this technology was very challenging to implement, but the problem Andrew Feldman, the founder, was trying to solve was worth supporting, so he began investing from the Series A round.

The second path was SambaNova's RDU, the reconfigurable dataflow unit. Its dataflow architecture aimed to maintain computing performance while reducing power consumption, providing another path for AI computing beyond GPUs.

In 2017, driven by Chen Liwu, Celesta made its first $2 million investment in SambaNova, valuing the company at around $12 million. Subsequently, he participated in multiple rounds of SambaNova's funding and helped the company bring in new investors.

Chen Liwu stated that SambaNova is currently advancing in its Series F funding round, with the funding amount expected to be between $8 billion and $10 billion. This refers to the funding size, not the company's valuation.

He emphasized that investing in startups should not solely rely on one founder, but instead look for a complete team that can adapt to changing circumstances. As the market evolves, out of the ten companies he has invested in, approximately nine have changed their initial business plans during development.

What truly deserves long-term support are teams that can adapt to changes, establish the right culture, and ultimately build world-class companies.

Transforming Cadence: Turning Suppliers into Customer Partners

When Chen Liwu took over Cadence, the company's stock price had plummeted to around $2.42.

Initially agreeing to serve as an interim CEO for only three months, while the company was concurrently searching for a permanent CEO, those three months turned into 15 years. During this time, Cadence underwent a transformation in corporate culture and product strategy, leading to a significant increase in the stock price from its low point.

Chen Liwu summarized the core of this experience in three words: humility, listening, and responsiveness.

Upon becoming CEO, he told employees at an all-hands meeting, "This is my first time as a CEO, so if you have any good ideas, feel free to email me." Subsequently, he started receiving around 300 emails per day, replying to each one individually. For suggestions that warranted deeper exploration, he would directly approach employees at their desks to discuss.

This approach helped him uncover internal information gaps within the company and allowed the management team to hear genuine feedback from the product frontlines.

The customer relationships also needed to change. Chen Liwu recalled that some of Cadence's customers at the time were extremely angry, not only demanding refunds but also explicitly stating their unwillingness to continue using the company's products. Some customers complained that there was never a response when they raised product issues in the past, and the company team only appeared near contract renewal time.

As a result, Chen Liwu drove Cadence to establish a rapid response mechanism. Later, a customer informed him that after lodging a complaint, someone was already in their office resolving the issue in less than 24 hours.

One of Chen Liwu's major competitors at Cadence once told him, "The same customer views me as a supplier, but views you as a partner."

In Chen Liwu's view, this is the key distinction between the two relationships. Once customers perceive a company as a partner, they are more likely to share product roadmaps and genuine needs. The company can also leverage feedback from other customers to provide more valuable recommendations.

He is bringing this approach to Intel, but Intel's business is more complex: the company needs to both rebuild product competitiveness and drive the success of its foundry business.

Intel Needs to Rebuild Compute Platform, Not Just Make Chips

Regarding why Intel still needs to undertake chip design, manufacturing, and sales simultaneously, Chen Liwu's answer is: the combination of product, advanced packaging, and foundry services can create greater value for customers.

Intel could theoretically further transition to an outsourcing model and no longer manufacture chips in-house. However, Chen Liwu still believes in vertical integration because the future of computing competition is not just a comparison between individual chips but the synergy of an entire system.

However, the premise for this strategy to work is for Intel to first have competitive enough products.

Chen Liwu acknowledges that Intel used to have a very strong market position in the CPU and computing fields, but has made many mistakes over the past years, gradually losing some advantages. To rebuild Intel, it is necessary to re-attract excellent CPU architects, GPU architects, system architects, and software talents to establish full-stack capabilities from chips to systems to software.

CPU remains the core of this strategy.

As AI shifts from training to inference and further into embodied AI, the demand for general-purpose CPUs may rise again. Chen Liwu stated that he now frequently receives calls from other company CEOs who hope Intel can provide more CPUs. Intel needs to increase the supply on one hand and develop new CPU architectures on the other to meet the requirements of future workloads.

This wave of demand is not only coming from traditional servers and data centers but will also extend to PCs, edge computing, and physical AI. Intel also needs to more closely connect cutting-edge research institutions, AI labs, and software development ecosystems to drive chip and system design based on actual application needs.

AI Bottleneck Overflow: Interconnects, Packaging, and Cooling Need Redoing

Chen Liwu believes that as AI compute scales up, bottlenecks are spreading from the chip itself to the surrounding infrastructure.

First is high-speed interconnects. As AI cluster scale increases, the performance of a single chip is no longer able to determine overall efficiency, making data transmission between chips, servers, and racks more critical.

Based on this assessment, he has invested in companies like Credo Semiconductor and Astera Labs and has also entered the photonics interconnect field. Some related companies were later acquired by Marvell, Credo, and other firms.

Next is heat dissipation. With the continuous increase in power consumption of CPUs and other AI chips, the heat dissipation solution is transitioning from air cooling to liquid cooling, and further advancing to microfluidic cooling.

Advanced packaging follows the same logic. Intel already has packaging technologies like EMIB-T, and Liwu Chen is also focusing on new materials such as glass substrates and synthetic diamond, aiming to enhance the packaging, insulation, and heat dissipation capabilities of high-performance chips.

Not all of these technologies necessarily need to become standalone businesses within Intel.

Li Wu's approach is that technologies that can be developed internally will be driven by Intel itself; for those that are not suitable for internal development, they can first support the growth of external startups and then potentially integrate them into Intel's platform through cooperation, integration, or acquisition in the future.

What Intel needs to establish is not a set of disparate products, but a larger platform that covers computation, interconnectivity, packaging, and manufacturing.

A Return to Storage? Intel is Exploring CPU and Memory Coordination

Intel was founded in 1968 and initially focused on memory chip business rather than microprocessors.

When asked if Intel might re-enter the storage market, Li Wu did not disclose specific plans but hinted at something worthy of attention: Intel is researching CPU and memory stacking, as well as new storage architectures.

In the past, he did not favor investing in storage chips because traditional storage products have strong cyclical characteristics and are considered commodities. However, as AI computing sets new requirements for bandwidth, capacity, power consumption, and packaging, storage is shifting from standardized components to a critical part of system performance.

Li Wu stated that new technologies are reshaping the storage industry. Exploring new storage architectures has become one of his focus areas.

He also mentioned that he has hired the former CEO of SK Hynix, Lee Seok-hee. As for whether this personnel arrangement indicates that Intel will expand its storage business again, he mentioned that the company is not yet ready to announce specific plans.

However, from an overall strategic perspective, what Intel is contemplating is not simply reentering the traditional storage market but whether CPU, storage, packaging, and manufacturing can form a new system architecture.

After Missing Out on Mobile, Cloud, and AI, Intel Bets on the Next 15 Years

Li Wu taking over Intel was not just a career choice.

He mentioned that at his age, he could have easily retired. However, as Intel is both an iconic company and holds significant importance for the semiconductor industry and the United States, he wanted to be personally involved and truly make an impact.

This also determined his time scale.

When he joined Intel, he told the board that he was not someone who only focused on the short term. He was thinking about what Intel would look like 10 or 15 years later, how to build a bigger platform, and how this platform could truly benefit the entire industry.

This long-term thinking also influences his understanding of competitive relationships.

Pan Xuebao has known industry figures such as Micron CEO Sanjay Mehrotra and NVIDIA CEO Jensen Huang for many years. Today, these companies compete with each other, but also have investment and partnership relationships. NVIDIA has become an investor in Intel, and the U.S. government and SoftBank are also on Intel's shareholder list.

But Pan Xuebao said he wouldn't simply see these people as competitors. The market is large enough, and the more important issue is how to jointly create a bigger market.

For Intel, the real challenge is not to briefly lead in a round of product competition, but to re-enter the constantly emerging frontier of technology.

This means connecting university professors, startups, venture capital firms, and AI labs to find next-generation opportunities in new materials, computing architecture, and system bottlenecks.

As Pan Xuebao put it, Intel has missed out on major waves such as mobile internet, cloud computing, and AI. Now, he wants to ensure one thing: when the next wave comes, Intel must be there.