Qualcomm Business Development Director Andy Li: AI is not "winner-takes-all," and Web3 is not a "monster"
TechFlow Selected TechFlow Selected
Qualcomm Business Development Director Andy Li: AI is not "winner-takes-all," and Web3 is not a "monster"
Qualcomm's role in Web3 is that of an enabler. We hope our customers can use Qualcomm's technology and products to create products that benefit societal progress and development within legal and ethical frameworks.
Navigating Web3 tides with focused insightsInterview: Sima Linwei
Article: Chen Xiaorui
On April 18, during Token 2049 in Dubai, Aethir launched the Aethir Edge product powered by Qualcomm technology, featuring the Qualcomm Snapdragon 865 chip.
Qualcomm is a semiconductor and communications technology company founded in 1985 and headquartered in San Diego, California, USA. Its main businesses include research, development, and sales of mobile communication chips, as well as licensing of related patented technologies. Its products and technologies are widely used in smartphones, tablets, automobiles, IoT devices, and various other wireless communication devices. Aethir is a decentralized GPU cloud infrastructure provider established in 2021 with headquarters in Singapore. Public information shows that in July 2023, Aethir completed a $9 million Pre-A round financing, achieving a valuation of $150 million at the time.
Recently, DeThings conducted an exclusive interview with Andy Li, Director of Business Development at Qualcomm. Below is the full transcript, edited for clarity.
DeThings: Could you briefly introduce and explain Qualcomm's technology and business?
Andy Li: Over its more than 30 years of development, Qualcomm has accumulated extensive technological expertise. Most people know us best for our mobile communications technologies—from early CDMA to 3G UMTS and CDMA2000, then to 4G LTE, and now 5G. Recently, we've announced our 5.5G evolution plan—what we call 5.5G, which is essentially an advancement of 5G. Our labs have already achieved breakthroughs in 6G technology, though commercial deployment will depend on market readiness.
Beyond mobile communications, we also possess significant computing technologies such as CPUs and GPUs, including the Snapdragon CPU and GPU. Our GPUs achieve a strong balance between high performance and low power consumption. Qualcomm developed its own CDSP architecture long ago—previously known as CDSP (with "C" standing for computing)—a digital signal processor designed for computation tasks. As AI algorithms like computer vision, CNN, TensorFlow, and others have evolved rapidly in recent years, Qualcomm has continuously refined the CDSP architecture to keep pace. More recently, we’ve introduced Qualcomm NPUs, including matrix accelerators, tensor accelerators, and vector accelerators, along with numerous optimizations in memory sharing bandwidth and access speed.
Therefore, Qualcomm’s expertise extends beyond mobile communications into computing. When applying these technologies to the Internet of Things (IoT), the first step is connecting physical objects—using Qualcomm’s connectivity solutions, not only cellular networks but also short-range communication technologies like Bluetooth and Wi-Fi. Once thousands of devices are interconnected, data silos are eliminated. While each object may be just a small sensor, it constantly generates data. In the past, when unconnected, this data existed in isolation, requiring manual extraction and analysis. With connectivity, this data forms a network where large AI models can automatically learn from and analyze the data, extracting valuable insights.
DeThings: From the PC era to the mobile internet era, the term “Internet of Things” (IoT) emerged. What are your thoughts on the concept of IoT?
Andy Li: For Qualcomm, IoT is not a new term—it has existed for quite some time. By definition, IoT means connecting physical objects into a network, which is the foundational step. We have a long history in this field. Many associate Qualcomm primarily with mobile communications, especially the Snapdragon platform in smartphones. But in reality, Qualcomm’s business is far broader. We see ourselves as an enabler for countless industries, leveraging our portfolio of products and technologies. While the Snapdragon brand is most familiar to consumers, we’ve also recently launched branded products for PCs and XR (VR/AR).
Whether it’s Depins or so-called Web3, fundamentally they share the same principle as traditional IoT—connecting all devices. The difference lies in the past, when individual devices had limited computing power, necessitating a powerful centralized computing platform for control. However, as each node becomes more capable computationally, decentralization—or what we call Depins—becomes feasible, because true decentralization requires every node to be sufficiently powerful; otherwise, a strong central brain would still be needed. Once even the smallest endpoints possess adequate computing power, a flat, distributed network becomes possible. This is where Qualcomm sees opportunity: through our technological capabilities at the edge and endpoint devices, empowering each node with robust processing power, combined with connectivity to link them together—this architecture enables genuine decentralized networks.
DeThings: After ChatGPT burst onto the scene, everyone’s attention shifted toward AI. How does Qualcomm view computing in the AI era?
Andy Li: I believe AI could genuinely be the force—since the Second Industrial Revolution—that reshapes human society’s production structure and, consequently, our entire social organization—an intelligent productivity. However, the general public might hold a broad understanding of AI as simply “artificial intelligence.” So how do we concretize AI into tangible applications in daily life?
First, we’ve seen general-purpose generative AI like ChatGPT. But ChatGPT represents general AI—an omniscient model. To become truly general-purpose means knowing everything. Hence, the parameter scale of such general AI models keeps growing, with some already exceeding 200 billion parameters.
But returning to IoT or industrial terminal domains, I question whether we actually need a universal generative AI like ChatGPT—this deserves further discussion. In IoT or industry-specific devices, these systems aren’t expected to handle every task. When designing a device, we already define a narrow set of functions it should perform within its domain. Just as we wouldn’t ask a radio, “What gift should I give my wife for her birthday?” we only pose operationally relevant questions to the device. In such cases, do we really need a general AI? Probably not. Instead, what’s needed is a customized, vertical AI model tailored specifically to its application domain.
This allows us to narrow the scope of AI. From a hardware perspective, requirements are correspondingly reduced. Running models with tens of billions of parameters currently demands data center-level infrastructure. On smartphones, you might run models with a few billion parameters. For IoT devices, even a 7-billion or 4-billion-parameter model may suffice. Our focus shouldn't be stacking hardware to run massive models, but rather optimizing efficiency in running AI models.
Qualcomm has strategic initiatives in this area. Beyond continuously enhancing chip capabilities—enabling devices that previously couldn’t run large models to do so—we’ve developed the Qualcomm AI Stack. Importantly, Qualcomm doesn’t build large models ourselves. Instead, we provide a complete suite of tools. Last week at the embedded world exhibition in Nuremberg, Germany, we launched the AI Hub—a collection of hundreds of open-source AI models ported and optimized for the Qualcomm Snapdragon platform, made freely available to developers. We aim for partners and developers to use our platform to create truly personalized, industry-specific AI solutions applicable across diverse sectors.
DeThings: In other words, do you or Qualcomm believe AI is a “winner-takes-all” domain? For instance, OpenAI is leading efforts to develop trillion-parameter-scale models. Given such powerful AGI (Artificial General Intelligence), why would other AIs still be necessary?
Andy Li: That’s an excellent question—one I actually address in my presentations. As you said, if AI were truly “winner-takes-all,” it would imply one monolithic AI existing solely in the cloud, since neither edge nor endpoint devices could support it.
First, specialized AI models are typically derived from a general large model, customized and fine-tuned according to specific industry needs. Thus, while we need general AI as a foundation, we must then specialize it.
Returning to your question: Why not connect all terminal and industrial devices directly to the cloud? This comes down to economic efficiency. You know that operating a massive data center in the cloud incurs enormous costs. Even a simple inference requires full activation, consuming vast amounts of energy—not just electricity, but also cooling systems like air conditioning. That’s why data centers are increasingly built in colder regions or near power sources—to minimize these expenses. That’s point one.
Second, consider that AI use cases and users will grow exponentially. Suppose one billion people worldwide use AI in the future—multiply the energy cost per inference by such massive scale, and the total economic burden becomes staggering. Under those conditions, is putting all computation in the cloud truly optimal? Or should some processing move to the edge and endpoints? I believe the latter is necessary.
More importantly, many industrial applications prioritize latency and latency reliability. If relying on the cloud, even if average latency is under 20 milliseconds, actual delays might vary from 10ms to 100ms. Such unpredictability is unacceptable for many industrial uses. In contrast, edge and endpoint devices offer consistent, ultra-low latency—ensuring reliable connectivity.
Third is data security—the very reason we advocate decentralization—to ensure data safety and integrity. Of course, the primary concern is security. For reliability, we prefer keeping data locally or within a controlled, accessible perimeter, acting as a node in a larger decentralized network.
Therefore, from an AI-era perspective, general large models are economically inefficient for meeting all specific needs. Moreover, they fall short in reliability and security, underscoring the continued necessity of industry-specific, customized solutions. General large models aren’t a universal answer.
DeThings: We know global smartphone shipments are declining. There’s a stereotype about Qualcomm: that its Snapdragon platform excels at mobile phone chips. So where does Qualcomm see its next growth engine?
Andy Li: First, although global smartphone shipments have declined compared to pre-pandemic levels, they are now slowly recovering.
Second, we continue to regard smartphones as our core business—no change there. Much of our R&D originates from smartphones, which represent historically the largest single-demand market. Though only a few major brands exist, their needs remain broadly similar. Thus, the smartphone market remains fertile ground for cultivating new technologies.
Beyond that, Qualcomm is actively pursuing business diversification. You can see our expansion into IoT, XR (extended reality), PCs, and automotive sectors. In automotive, for example, Qualcomm leads in both advanced driver assistance systems (ADAS) and infotainment systems. We’ll continue advancing this diversified strategy.
These emerging markets are also substantial in scale. Take IoT, for instance—we estimate its total market size at $720 billion, surpassing even the smartphone market. While IoT is highly fragmented, with modest volumes per industry, each sector has unique demands. Yet, beneath these varied applications, underlying technical needs are often aligned. Our flexible portfolio of technologies and patents enables us to tailor combinations of products and solutions across industries. I believe this adaptability offers tremendous value to customers, delivering cutting-edge technology in the most efficient way possible.
DeThings: Aside from smartphones as the core business, what do you see as Qualcomm’s greatest potential growth area in the future?
Andy Li: Beyond smartphones, we see our biggest growth opportunities in IoT and the automotive sector.
IoT is indeed one of the largest growth vectors. Additionally, both new energy vehicles and internal combustion engine vehicles present huge potential. Even conventional vehicles demand increasing levels of intelligence. While new energy vehicles don’t face battery constraints and allow for cleaner design integration, legacy vehicles also require smart upgrades. For us, the propulsion method is secondary—we aim to make every vehicle smarter, enabling seamless connectivity between vehicles, infrastructure, and, most importantly, drivers. This is precisely where Qualcomm excels.
In fact, we view cars as part of the IoT ecosystem—as intelligent connected devices. Therefore, automotive fits naturally within Qualcomm’s broader IoT strategy.
We believe Qualcomm holds unique advantages in these areas due to our extensive partnerships, vast customer base, and collaborative relationships. Qualcomm has always been sincere in sharing our latest technologies openly and generously. We believe that widespread customer benefit ultimately benefits society as a whole.
Dethings: What’s your take on the concept of Web3? Besides AI, Web3 has been a hot topic over the past few years.
Andy Li: Regarding Web3, we see it as a promising and emerging form of economic activity. For Qualcomm, we play an enabling role—because Web3 is decentralized, with no single dominant entity, and every participant contributes equally.
As I mentioned earlier, Qualcomm has always been—and will become even more—open. We’re opening up our hardware platforms, even open-sourcing software components, embracing open-source software more deeply, and contributing our own open-source code. We aim to empower our partners and customers to thrive in Web3 and potentially even future iterations like Web4. We provide the tools so they can innovate in this expansive landscape. Because if AI represents a transformation in productivity, then Web3 signifies a shift in production relationships. Qualcomm embraces this emerging industry as an open enabler.
Dethings: Why did you choose to collaborate with a Depin platform like Aethir? For a company like Qualcomm, involvement in Web3 seems unusual. Web3 carries strong financial attributes, involving token economics and related aspects. Does Qualcomm have any concerns in this regard?
Andy Li: Aethir is actually a partner we highly value. They have deep experience in both cloud computing and edge computing. If you visit Aethir’s website, you’ll see real-time data displays of their resources—for example, GPUs currently in operation. This transparency is impressive. They’re not just a conceptual project; they operate a real, functioning business grounded in commercial practice. As a pragmatic company, Qualcomm wants to work with equally practical commercial partners—whether building character models or real production tools.
Additionally, we don’t get involved in the economic layer you mentioned—that’s beyond Qualcomm’s scope. We focus primarily on the technology layer. We believe technology is borderless and inherently neutral—neither good nor bad. We enable technology for others to use. Personally, I don’t see Web3 as a “monster” to fear. We need to understand and embrace new developments—only through understanding can we apply them correctly.
Qualcomm’s role in Web3 is that of an enabler—we want our customers to leverage Qualcomm’s technologies and products to build socially beneficial innovations within legal and ethical boundaries.
Dethings: We’ve noticed companies like Google, Microsoft, and NVIDIA already have some Web3 clients in infrastructure. Is Qualcomm observing or exploring further opportunities in the Web3 space?
Andy Li: We maintain an open stance and welcome all customers to reach out. We believe Web3 represents a new form of production relationship. As a technology enabler, Qualcomm is willing to share our capabilities and technologies.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
@DethingsNews
