This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies. Read our privacy policy

Huawei's Eric Xu: Innovating Nonstop for Faster Digitalization

Sep 23, 2021

[Shenzhen, China, September 23, 2021] Today, Huawei kicks off its annual flagship event, HUAWEI CONNECT, which is being held entirely online for the first time. Themed "Dive into Digital", this year's event brings together industry visionaries, business leaders, top tech experts, and ecosystem partners to explore how to better integrate digital technology with business scenarios and industry know-how, and how stakeholders can work together more effectively to foster an open industry ecosystem and drive shared success. Eric Xu, Huawei's Rotating Chairman, opened the event with a keynote speech. The full text is as follows:

Eric Xu delivers a keynote speech

Huawei Rotating Chairman Eric Xu delivers a keynote speech at HUAWEI CONNECT 2021

Thank you for tuning in to HUAWEI CONNECT online. Having the opportunity to connect with you online is a testament to the progress the world has made in digital transformation. Today, I'd like to talk about how Huawei is innovating nonstop to speed up that progress further.

The world is changing fast, and so is digital technology. There's a global consensus on the importance of going digital these days. It might be the only thing everyone can agree on. More than 170 countries have released a national digital strategy. Recent developments have reminded us that digital transformation is now more real and urgent than ever.

First of all, the lingering pandemic has accelerated the digitalization of products and services over the last two years. According to McKinsey, COVID-19 has caused digital transformation to jump ahead by seven years globally and by a full 10 years in Asia Pacific. Their research also found that going digital is no longer viewed as some tedious and impractical endeavour. On the contrary, companies have proceeded 20 to 25 times faster than expected. And now it's pretty much accepted that hybrid onsite and remote work will be the future of the workplace.

The second major catalyst is the proactive global response to carbon emissions and global warming. The EU has announced its plan to achieve carbon neutrality by 2050, and China has pledged to hit peak emissions by 2030 and become carbon neutral by 2060. Digital technology holds the key to reducing emissions across all industries. According to a survey by the World Economic Forum, by 2030, ICT technology will help reduce industrial emissions by 12.1 billion tons, roughly 10 times the amount emitted by the ICT industry itself.

Third, the increasingly complex global environment has caused companies to place greater strategic priority on business resilience, for which digital technology is a key enabler. Economic recovery from the pandemic and low-carbon development are both pushing organizations worldwide to pick up the pace of their digital transformation.

The good news is that the underlying technology has never been more ready. Many countries came to realize this in their efforts to grapple with the pandemic. The development of the underlying digital technology and infrastructure has laid a solid foundation for digital transformation.

Globally, there are already 176 5G commercial networks and more than 10,000 projects exploring how 5G can drive industrial digitalization. On the consumer end, there are already more than 490 million 5G users worldwide. According to IDG, 81% of organizations worldwide are already using cloud or have applications in the cloud. AI is developing faster than ever. A study by Roland Berger shows that AI has already penetrated every sector. In certain sectors, such as high-tech, telecoms, finance, automotive, and assembly, AI adoption exceeds 60%, and in business services, healthcare, and retail, adoption rates are roughly 50%, 40%, and 38% respectively.

Just as digital transformation is a shared vision, the technology itself can be shared too. So what's the best way forward? All countries, businesses, and industries are different, and they have unique challenges. So of course their understandings, strategies, pace of development, and approaches to digital transformation tend to differ as well. So true digital transformation is still a long way ahead.

Huawei's mission and vision is to bring digital to every person, home and organization for a fully connected, intelligent world. I believe helping industries go digital is a critical aspect to our mission. You could even say we wouldn't be able to accomplish our mission without the successful digital transformation of industries.

There are four pillars to our value proposition:

  • Ubiquitous connectivity: We believe that every person has the right to be connected, and Huawei's role is to bring connectivity to all people and things, and to keep raising the bar for user experience.
  • Pervasive intelligence: We see AI as a general-purpose technology that can help all industries bring every step of their end-to-end value creation process to the next level.
  • Personalized experience: Every person is unique. We are committed to providing personalized products and services so that the individuality of every person is fully respected and their potential is fully unleashed.
  • Digital platform: We firmly believe that digitalization will bring civilization to entirely new heights. We are committed to providing open, secure, flexible, and easy-to-use digital platforms to spark innovation across industries, drive industrial upgrade, and advance social progress.

Digital development relies on digital technology. For digital technology to stay relevant, we must continue to innovate and create value. Cloud, AI, and networks are three critical digital technologies. I'd like to talk about these three today and then touch on what we're doing to enable low-carbon development. I'll share some of the progress we've made and our thoughts on where the industry is heading.

First, let's take a look at cloud services.

On September 1, 2016, at the very first HUAWEI CONNECT, I delivered a keynote called "Embrace and Integrate with the Cloud to Become a Digital Enterprise".

On March 19, 2017, we announced the establishment of our Cloud BU at the Huawei Eco-Partner Conference, held in Changsha, China. At the time we also said that, from 2017 onward, Huawei would ramp up efforts to build open public cloud platforms with public cloud services. We were to focus on select industries and work with partners to build a cloud ecosystem and grow the pie together.

Now, four years later, HUAWEI CLOUD has brought together more than 2.3 million developers, 14,000 consulting partners, and 6,000 technology partners, and also made more than 4,500 services available in the HUAWEI CLOUD Marketplace. It has become an important platform for Internet companies, traditional enterprises, and governments alike to take their organizations digital. HUAWEI CLOUD, together with partner public clouds, now serves 27 Regions in more than 170 countries around the globe. According to a study by Gartner, HUAWEI CLOUD was the fastest growing cloud in the IaaS market in 2020, and made it into the top 2 for cloud service providers in China, and top 5 globally. We've come a long way, and that's just the beginning.

With HUAWEI CLOUD, our mission is to build the cloud foundation for an intelligent world through ubiquitous cloud and intelligence. As industries speed up their digital transformation, HUAWEI CLOUD is primed to develop even further.

As digital transformation begins to take root, and with the growing diversity and sophistication of digital applications, traditional cloud services that simply provide the basics like elastic resources and simplified O&M are no longer going to cut it. Super-elastic resources, alongside agile application development and iteration, are the way forward for cloud services. This is why tech companies, traditional enterprises, and governments have all started embracing the idea of Cloud Native. The shift towards cloud-native applications will allow traditional enterprises and governments to benefit from greater resource elasticity and agility. Beyond that, they will be able to create greater value by harnessing the power of big data and AI that come with new, cloud-native services.

As an advocate and early adopter of Cloud Native, HUAWEI CLOUD has released a great number of cloud-native services since 2016, helping Internet companies, traditional enterprises, and governments become cloud natives themselves. Building on this experience, we brought up the concept of Cloud Native 2.0 in 2020, the next phase of development where we hope to enable all organizations to become new cloud natives.

As cloud-native applications become more widespread across all sorts of different scenarios, the need for distributed deployment, unified management across clouds and regions, and ensuring a consistent experience will become increasingly important.

To address this need, after years of hard work, today we're launching the industry's first distributed, cloud-native service called HUAWEI CLOUD UCS. UCS stands for "ubiquitous cloud native service". With HUAWEI CLOUD UCS, we want to provide organizations with a consistent experience while using cloud-native applications that are not constrained by geographical, cross-cloud, or traffic limitations. UCS aims to bring cloud-native capabilities to every business scenario and accelerate the adoption of cloud-native applications in all industries.

Now, let's move on to artificial intelligence.

In October 2018, Huawei launched its full-stack, all-scenario AI portfolio at HUAWEI CONNECT in Shanghai. On August 23, 2019, we announced the open source plan for our AI computing framework, MindSpore, in Shenzhen. Since these announcements, we've stuck to our plans and have managed to meet our targets.

First, in terms of hardware, more than 10 of our partners have launched AI hardware products that use our Ascend modules and cards.

Second, MindSpore went open source in March 2020, as scheduled. From that time to the end of August 2021, it has been downloaded more than 600,000 times, making it the most vibrant AI community in China. There are also more than 100 universities that include MindSpore in their curriculum. It's fair to say that MindSpore has become the mainstream AI computing framework in China.

In addition, more than 500 of our partners have developed over 600 AI solutions based on Ascend. These solutions are used across a wide variety of industries. Generally speaking, our full-stack, all-scenario AI portfolio is moving forward as expected.

At HUAWEI CONNECT 2019, we released the Atlas 900 cluster. At the time, a single cluster used 1,024 Ascend 910 AI processors, delivering 256 PFLOPS of computing power. Now, a single Atlas 900 cluster can use up to 4,096 Ascend 910 AI processors, delivering 1 EFLOPS on non-blocking networks.

On top of these clusters, HUAWEI CLOUD ModelArts can use inter-cluster dynamic adaptive routing technology to expand the computing power of a cluster by a factor of between 4 and 32, depending on the power constraints. That adds up to computing power of up to 32 EFLOPS and an increase in linear speedup ratio to more than 85%. The Atlas 900 cluster, as well as the cloud services based on it, currently serve more than 300 enterprises across all sorts of industries, including transportation, finance, energy, manufacturing, and healthcare. They are used by many enterprises and research institutes for large model training.

These are some of the more noteworthy pre-trained large models supported by the Atlas 900 AI cluster:

  • HUAWEI CLOUD Pangu Chinese NLP large model
  • HUAWEI CLOUD Pangu CV large model
  • HUAWEI CLOUD Pangu drug molecule large model
  • HUAWEI CLOUD Pangu scientific computing large model
  • LuojiaNet, a dedicated remote sensing framework
  • Pangu Chinese NLP large model for Peng Cheng Laboratory
  • Biopharmaceutical large model for Peng Cheng Laboratory

In Huawei's full-stack, all-scenario AI strategy, ModelArts is positioned as an enabler of AI applications. Its goal is to enable incredibly simple AI application development, helping address the growing shortage of AI professionals and experts. Our hope is that ModelArts will equip each and every engineer who has a basic grasp on AI to independently develop their own AI models and applications. Over the past three-plus years, ModelArts has been used in thousands of AI application projects across different industries. Throughout this time, we have continued innovating and accumulating industry know-how to better adapt the framework to meet the needs of organizations at different stages of digital transformation and AI adoption. This has resulted in a series of full-pipeline, scenario-based services based on ModelArts. The creation of these services marks the first step in realizing our goals for ModelArts.

For most enterprises, there are three different stages of AI application development, and ModelArts provides targeted services for each.

In the early stage, most enterprises take a more experimental approach to AI for either generalized or more specific tasks. In this stage, they're mostly focused on model development and feasibility. The AI capabilities of enterprises in this stage are often quite limited. To address this, ModelArts provides services and development tools like domain suites, example scenarios, Pangu large models, and pre-trained models, allowing engineers to quickly pick up, train, and verify AI models without having to mess around with too much code.

The second stage is the "quick win" phase where, building on successful experimentation, enterprises tend to focus on how they can use AI to create immediate value. During this phase, AI development is no longer about developing models, but about facilitating one or more real-world tasks, catering to specific deployment environments and specific industry requirements, and adopting trustworthy designs. So for organizations in this stage, ModelArts provides things like trusted components and security algorithms, ModelBox, AutoSearch, and Pangu large models, allowing AI engineers to adapt solutions to diverse deployment environments and rapidly develop AI applications that can be used in real-world scenarios.

The third stage is characterized by the development of systemic AI applications or AI sub-systems. This stage often requires collaboration between multiple applications, tools, and systems. ModelArts supports simplified and efficient AI system development by offering MLOps, OptVerse AI Solver, scientific computing, Pangu large models, heterogeneous distributed system schedulers, and a wide range of industry-specific components and tools from ecosystem partners. With ModelArts, we are committed to equipping every engineer with the tools and support they need to develop AI applications, and we look forward to achieving this goal as soon as possible.

Even the most seasoned AI experts find it challenging to integrate AI into real-world scenarios across different industries. This is because industrial scenarios are varied and don't often have much in common. Even with highly automated tools, AI model development has to be done on a case-by-case basis, which is extremely labor intensive and time consuming. A general lack of good data to train models presents an even bigger challenge. For AI models to be accurate, we have to feed them huge amounts of data. A lack of scenario-based data means that models often fail to meet real-world requirements, rendering AI useless in certain scenarios. In situations like this, large models are a good solution.

With pre-trained large models, developers don't have to start from scratch when developing AI for any scenario. Instead, smaller, scenario-specific models can be automatically extracted through targeted training based on a large model. This shortens the development cycle from months to days, marking a shift from manual AI model development to industrial-grade development.

More importantly, targeted training based on large models greatly improves model performance and AI usability. These are actual results from our Southern Factory. With only 40 data samples, the accuracy of AI models trained with conventional methods can only reach about 80%, which definitely doesn't meet our requirements. But if we train them based on large models, we're able to boost the accuracy to as high as 99.5%. So now using AI for detecting defects is a practical reality.

Next, I'd like to talk about enterprise networks.

As organizations go digital, they tend to see exponential growth in network complexity, because they have to deal with:

  • More connected branches and access locations in a hybrid work environment
  • More dynamic changes to experience due to greater employee mobility
  • More connections as office networks converge with IoT
  • More performance requirements and frequent network changes due to cloud and new apps
  • More types of equipment from more vendors, leading to greater management scope
  • More demand on network assurance as the focus shifts from connection-centric to experience-centric

At the same time, these organizations won't see linear growth in their number of O&M engineers – or they won't see any growth at all. The gap between the complexity of network O&M and the availability of O&M engineers will only continue to expand. This challenge to network O&M can be addressed by applying digital technology. Innovating, not expanding the O&M team, is the right way to cope with the growing complexity of network O&M.

Based on this belief, we came up with the concept of autonomous driving networks, or ADN. We believe that networks of the future will be like autonomous vehicles that are able to automatically operate and maintain themselves without manual interference. ADNs will have four characteristics:

  • First, they will be self-fulfilling. That means the network can automatically deploy services based on user intent. The ultimate goal is to fully automate service deployment.
  • Second, they will be self-healing, meaning that the network can predict and prevent faults, and automatically recover from incidents. Fully automated O&M is the ultimate goal in this respect.
  • Third, self-optimizing, with networks that can automatically adjust and optimize themselves to provide a superior experience. Ultimately, we want to aim for fully automated network optimization.
  • Fourth, autonomy. This means that network functions will be able to autonomously adapt, learn, and evolve.

This is our vision for ADNs and what we ultimately hope to achieve.

Over the past two years, we've been innovating solutions for global networks based on the ADN concept. We've also been working together with customers from the finance, education, and healthcare sectors to innovate and deploy new applications.

In the finance sector, Huawei and China CITIC Bank have applied ADNs to upgrade the bank's data center network. In 2020, our ADNs enabled end-to-end automation of services across over 40 scenarios on one data center network with equipment supplied by a single vendor. This year, we will keep pushing the envelope to support heterogeneous scenarios that involve multiple clouds and equipment from multiple vendors. Take the bank's services for sending money transfers to international students, for example. In the past, launching a new service like this took an average of more than 30 days to do collaborative design and evaluation across domains and manage any changes. Now the entire process can be completed within just 30 minutes.

Being able to rapidly locate faults is the biggest headache when managing any data center network, but it's no longer the case with ADNs. ADNs offer end-to-end visibility into network quality. They also make it possible to locate 75 types of typical faults within three minutes, and provide advice for fixing the fault within five minutes. This year, our ADNs began using knowledge graphs for self-learning. They are able to analyze live network data to continuously learn about new faults, boosting overall coverage to 97%.

In the education sector, Huawei and Xi'an Jiaotong University are using ADNs to redefine campus networks. As the university deploys more and more smart teaching and campus services, it has to manage a growing number of IoT devices like cameras, barrier gates, smart doors, and lecture recording devices. Across their four campuses, they have more than 500,000 devices across 50 categories. They're used in all sorts of different places and can access the campus network, which could potentially pose a security risk.

Our ADN technology helps the university's networks automatically identify devices and grant access within seconds. Powered by AI, the ADNs can also learn about and label unknown devices online. Now it can identify 98% of devices on its own. Wireless access over campus networks has already become mainstream. In the past, Wi-Fi inference, roaming, and application support were big challenges to the university. They had to manually optimize the Wi-Fi network, which was very inefficient. Now with Huawei's AI-based optimization functionality, they have reduced the need for manual interference and increased their Wi-Fi RSSI fulfillment rate from 64% to 90%.

Finally, I'd like to talk about how Huawei is supporting low-carbon development with digital technology.

Like I said earlier, digital technology is key to going low-carbon. Huawei will continue to innovate in digital technology to support low-carbon development. We have three priorities in this area:

  • First, we invest and innovate in energy-saving technologies to deliver more energy-efficient ICT products for a low-carbon ICT industry.
  • Second, we invest in innovations where power electronics and digital technologies converge to promote clean energy and the digitalization of traditional energy.
  • Third, we provide digital technology to help all sectors go digital and low-carbon.

Helping the ICT industry go low-carbon is our first priority. For decades, Huawei has been developing equipment and solutions around the goal of reducing energy consumption and emissions. New requirements are emerging on combating climate change and achieving low-carbon development in all sectors. These requirements raise the bar for ICT equipment. In response, we will set higher goals of realizing energy conservation through innovation.

Our second priority is to promote clean energy and the digitalization of traditional energy. As part of global efforts to achieve peak CO2 emissions and carbon neutrality, we aim to use our digital technology to promote clean energy and the digitalization of traditional energy. For this purpose, we recently set up a subsidiary called Huawei Digital Power. Our vision has several layers. We want to promote clean energy and the digitalization of traditional energy. We're also working to more effectively integrate digital and power electronics technologies, and converge information and energy flows to drive an energy revolution for a better, greener future.

Specifically, Huawei Digital Power converges power electronics and digital technologies, in order to use bits to manage watts. That means we use digital technology to control power electronics equipment. Our main focal areas span from clean power generation, energy digitalization, and transportation electrification, to green ICT infrastructure, and integrated smart energy. We provide secure, efficient, green, and intelligent solutions for these areas. We also provide the energy sector with widely-used enablement platforms, including those for embedded power, intelligent power distribution, and energy storage. In addition, we will develop a unified Energy Management Cloud Service platform for all digital power scenarios. This will be an open application platform for all of our customers and partners.

By providing these products and solutions, Huawei Digital Power aims to support low-carbon development in all areas, including households, buildings, factories, campuses, villages, and cities. We will work to drive the shift towards low carbon, and ultimately zero carbon.

Our third priority is to reduce emissions in traditional sectors, especially those with typically heavier emissions. This is central to the transition towards a low-carbon world. It's also a key part of our innovation strategy. We remain committed to providing digital technology to help all sectors go digital and low-carbon. It's fair to say that cutting emissions is a shared goal between Huawei and almost all the sectors we work with. We've made some exciting progress so far.

In smart transportation, our traffic light management solution reduces traffic jams and emissions in cities. Our smart highway solution supports free flow tolling and has helped reduce fuel use by 321,700 tons thus far.

Our smart heating solution is already reaping benefits for Harbin, a city in northeast China. In the city's Daowai District, on-demand heating reduces energy use by an average of 12.1%. If smart heating is used across 13 billion square meters of building space in China, then CO2 emissions will decrease by 16.199 million tons per year.

Our smart agriculture solution is also creating real value for farmers in Switzerland. Drones that use big data and 5G are able to inspect farmland 20 times faster than before. Through precision weeding, farms can reduce weedicide use by 90%.

There's no doubt about it: Digital transformation is a long-term process that won't happen overnight. Fortunately, the tech sector is more dynamic and vibrant than ever. Nonstop innovation has been the driving force behind digitalization thus far. Moving forward, if we hope to reach more ambitious goals for digitalization, nonstop innovation will continue to be key. So let's innovate nonstop for a better future.

Thank you!