Where have all the research and development funds of Chinese internet giants been invested, with nearly 200 billion yuan invested in a year?

With the boom of large model alchemy, Nvidia recently became the seventh American company with a market value of more than $1 trillion. At the previous Taipei International Computer Show, CEO Jensen Huang boasted about China's progress in cloud computing, digital payment, electric vehicles and other fields at the round table forum. When talking about the technical blockade of the United States, his statement implied some concerns:

No matter what regulations there are, of course we will definitely abide by them. But I think China will also use this opportunity to cultivate local enterprises. That's why there are so many GPU startups in China

This may seem like a warning to Chinese companies about the pace of progress, but it actually reflects the gap in cutting-edge technology between China and the United States. While keeping pace with international first-class standards in some fields, domestic companies still have significant shortcomings in areas such as general-purpose chips, chip manufacturing, and computer science.

During the same period, from Microsoft's Copilot to Google's I/O conference, American stock market peers seemed to be able to come up with some new ideas every year, combined with the technological weaknesses exposed in trade frictions, causing the struggle between model innovation and technological innovation, which quickly became rampant. When a group of overseas companies stand in the field of technological innovation, domestic internet companies have once again stood at the forefront of countless magnifying glasses.

In fact, from the financial reports of various companies, whether it is the absolute value of research and development expenses or the proportion of revenue, the investment of mainstream internet companies is not small and is increasing year by year. Roughly speaking, mainstream internet companies have invested as much as 200 billion yuan in research and development in the past year, but from the actual results, it seems that they are indeed lacking in quality.

On the one hand, when many high-tech fields in China started, overseas peers had already become industry oligarchs that could rival the rich, which was also the source of many technological gaps. On the other hand, in the era of rapid growth of mobile internet, the focus of domestic companies is indeed on the engineering and application levels. With the gradual exposure of technological weaknesses in trade frictions, both public opinion and industry have begun to systematically examine their own arrears.

Facing this systematic gap, while facing the gap, it is more important to explore sustainable paths for technological innovation.

Market is a keyword

In the book 'All in One Catch', it records an important opportunity for the birth of Amazon's cloud business:

In 2002, a publisher approached Amazon founder Bezos and hoped that the latter could open up their sales data to the public, allowing publishers to arrange book distribution plans accordingly. Meanwhile, after Amazon opens its own interface, other websites can also embed Amazon's products, which in turn promotes Amazon's sales.

At that time, there was no precedent for this type of interface openness in technical engineering for other companies. During this process, Bezos gradually realized that Amazon's IT infrastructure, computing resources, etc. could be open to third parties. Ultimately, this became the prototype of Amazon Web Services (AWS), the Amazon cloud business.

Neither Bezos nor the book publisher named Tim O'Reilly may have foreseen the future development of cloud computing. Bezos' original intention was to expand Amazon's revenue capabilities, as stated in the book: "At first, he replied that he did not see what benefits Amazon would receive

Market is a keyword

In 2016, Google's AlphaGo defeated Li Shishi in a Go game, and Google also released its self-developed AI chip TPU. Compared to the GPU commonly used in AI training, TPU significantly reduces storage and connection requirements, but more importantly, it is low-cost.

In a similar context, the difficulty of many domestic technological breakthroughs lies in the natural lack of competitiveness during product launch. Due to the lack of market buyers, the product loses the funds for subsequent optimization iterations, and ultimately ends up in a messy manner. If a technology that consumes a huge cost cannot be pushed to the market and is shelved in the laboratory, it will undoubtedly cause the greatest harm to the industry.

In the context of the low starting point of the industry itself, the core of some breakthroughs in domestic underlying technologies also lies in the market's feedback. The domestic cloud computing industry, as well as its parallel databases, is gradually stepping into world-class development in this context.

The exploration of databases in China began around 2008. At that time, Ant and Alibaba were both major customers of Oracle. However, as the business scale grew, Oracle's products were unable to meet some customized needs in technology, which also brought a huge cost burden. In this context, Ant began to develop its own database products.

The so-called relational database is a flagship product of Oracle, which can be simply understood as a bookshelf storing millions of books. Each book has many different attributes, such as book title, author, publisher, ISBN number, etc. These books may also have some "relationship", such as the same author writing several books, many of which come from the same publisher. The storage and retrieval of this database can be very complex.

The breakthrough of such products, on the one hand, is technically difficult and requires a very high amount of investment. On the other hand, the data based business is running all the time, so it is impossible for Alipay system to maintain for a month and wait for migration. In consideration of security and stability, migration costs are very high, and it is easy to rely on suppliers.

For independent teams, due to the fact that the large customers of such products are often concentrated in areas such as finance and communication that require high security and stability, they are very afraid of potential risks. In this situation, even if there are formed products, it is difficult to obtain profits in the market to support subsequent research and development. Therefore, many domestic products have passed the research and development stage and are still on the eve of industrialization, unable to take the next step.

At that time, Oracle's research and development center in Beijing had nearly 2000 employees, while most of its domestic peers had only a few dozen or hundreds of employees. The products were not competitive and they were unable to invest in subsequent research and development due to a lack of orders, leading to a vicious cycle.

Ant's self-developed database product has a natural advantage: due to the rich business of the entire Alibaba system, it can test its edge business and participate in the product validation process. In other words, it means being the first party to oneself, allowing self-developed products to cross the stage of difficulty in participating in market validation and self generating.

In 2010, Ants began building the OceanBase database, replacing MySQL and Oracle bit by bit from Taobao's favorites business. At the same time, Ali began to go to IOE aggressively. At the same time, due to the popularity of cloud computing, the one-time investment in enterprise IT has been significantly reduced, driving the market space for underlying software such as databases and proprietary clouds, creating a path for Chinese enterprises to overtake on a curve.

By May 17, 2013, Alipay had completely offline IBM minicomputers and EMC storage devices. As Taobao's advertising business stopped using Oracle's database in July of that year, Alibaba had achieved the autonomy and control of the database, and Oracle had lost its largest database customer in Asia. In August 2019, Ant Group's distributed relational database OceanBase successfully passed the TPC-C test of TPC, an authority in the database field, surpassing Oracle, an old database giant. A year later, OceanBase once again reached the top, increasing the record by 11 times.

The growth of OceanBase within the Ant and Alibaba systems is actually a microcosm of the technological innovation path of domestic internet companies:By leveraging its diversified business advantages, we aim to continuously validate our self-developed products in diverse business scenarios and ultimately achieve substitution.

This is also an innovative path for domestic internet companies in many technological fields.

Starting from the Edge

Last April, Google released its self-developed video processing chip ArgosVCU, replacing tens of millions of Intel CPUs, saving Google 20 billion yuan in capital expenses in one fell swoop. Coincidentally, Amazon has also successfully replaced Nvidia's chip by processing the operation of Alexa's voice assistant through its self-developed chip, reducing the cost by 30%.

The biggest characteristic of this type of chip is that they are specialized chips with low universality.

For example, the so-called general-purpose chips, such as Nvidia's GPU, can be competent for multiple scenes such as games, rendering, AI, mining, etc; And most specialized chips are only used for a very specific scenario, such as AI training. Simply put, specialized chips sacrifice universality and enhance a specific function. When making chips, Nvidia should not only serve Google and Amazon, but also Tesla and Weilai. It should also consider the vast number of game players and mine owners. This is "universality".

In the context of core making, most of the public are distressed by general-purpose chips such as CPU and mobile phone SoC, which are also the advantageous projects of Nvidia, Intel and other large companies. In some specialized chips for segmented scenarios, domestic internet companies have actually made significant achievements.

Alibaba's Pingtou Ge once developed a chip called "Hanguang 800", which is specifically designed to serve Taobao's feature of identifying products based on images, "Pailitao". It can complete the recognition of 1 billion "Pailitao" product images in 5 minutes, while traditional GPU processing takes 1 hour.

Afterwards, Pingtouge successively developed four product series, namely Xuantie (processor IP), Feather Array (RFID chip), Yitian (ARM server chip), and Hanguang (high-performance AI inference chip). It is said that the shipment volume of the Xuantie series processors has exceeded 2 billion, and more than 200 enterprises are using them.

It is not difficult to develop a chip, but rather to develop a high profit chip for the consumer market. This not only requires technological breakthroughs, but also the cooperation of terminal products, corresponding software ecology, and long-term capital investment.

Even for companies like Apple, their self-developed M1/M2 chips actually start with some segmented scenarios, such as highlighting stronger performance in specific scenarios such as editing and video processing, and adapting to macOS, rather than competing head-on with x86 architecture in all aspects.

In terms of Apple's size, it is still necessary to avoid sharp differentiation competition. For domestic companies with lower starting points in industrial chain integration capabilities and ecological construction, if they hastily enter GPU and other chips with high barriers, they are likely to repeat this cycle - the product is not competitive, lacks order feedback and R&D iteration, and ultimately falls behind step by step.

Therefore, the approach of domestic large factories is actually to start with segmented and edge based scenarios, and make targeted investments, while relying on their own business volume to continuously support product iteration. In other words, the internal business of internet companies is a large market that can support R&D teams to gradually explore from the edge to the center.

For example, Tencent, Byte and Kwai are all investing in special chips for audio and video cloud processing to transmit content at a higher compression rate and smaller bandwidth. It should be known that the bandwidth expenditure of major manufacturers accounts for nearly 10% all the year round, and the bandwidth magnitude of Kwai is close to hundreds of TB. With the help of special chips, it can save a lot of money.

Whether it is hardware represented by chips or software represented by databases, the vast business system of internet companies can enable some products to pass the difficult stage of market validation and be pushed to the market after being polished and matured.

At the Zhongguancun Science and Technology Expo in May this year, Ant announced for the first time a complete open-source layout centered on key basic software. In the database and middleware fields, which are two of the three major software components, Ant has accumulated nearly 100 community leading open-source projects.

These technological accumulations often have a common characteristic: relying on ants' own business volume to gradually grow, and participating in competition in the process of opening up to the outside world.

In summary, in the context of technology research and development, the most important role played by internet companies is actually to create a market that can validate products for technological breakthroughs through their own business system. Expanding to the entire technological breakthrough in China, a large-scale market is also the strongest backing for Chinese companies.

Moving towards a thousand lines and various industries

In 2002, Fortune magazine conducted a survey and the results showed that the total profits of 10 pharmaceutical companies among the top 500 companies in the United States exceeded the sum of the other 490 companies.

Before the popularization of mobile Internet, the most profitable industry in the world was pharmaceutical industry, and almost all the top pharmaceutical companies were in the United States. In 2022, five of the top ten global pharmaceutical companies with the highest revenue scale belong to the United States.

The strength of technological research and development is naturally an important reason, but the most core factor is that the United States has a huge pharmaceutical consumer market. In 2019, the per capita health expenditure in the United States reached as high as $10000, far exceeding not only China, but also many developed economies such as the UK, Germany, and France.

Technological breakthroughs often require a huge market as support. TSMC is invincible in the advanced manufacturing process, which is not only the investment of countless engineers, but also because Apple, Nvidia and AMD are lining up to buy TSMC's production lines and reimburse TSMC for its R&D expenses.

Similarly, in the ascent of high-end technology, China's biggest confidence still lies in market size.

In 2020, OceanBase, as an independent company, began commercial operations and served industries that were extremely sensitive to data security, such as finance and communication. It achieved significant cost reductions and upgrades to overseas products to a large extent.

Shi Wenhui, the head of database products of Ant Group, OceanBase solutions and general manager of products, once shared a case - Shandong, a populous province with OceanBase ranking second in telecom operator business, realized the replacement of Oracle:

"The operator's business is very complex, with many complex queries and many high-level features of Oracle, which poses a great challenge to any database. After full communication with the operator, both parties want to try to smoothly migrate from Oracle Database to OceanBase. After several months of efforts, it took only one hour to complete the switch of the entire business. Finally, the query efficiency of BOS detailed list was improved by 30%, and the storage cost was reduced by 90% Only 1/10 of the original

After 2018, the trade frictions between China and the United States exposed the difficulties caused by technological shortcomings. At the same time, the trend of anti globalization and supply chain autonomy gradually emerged. This continuous external pressure ultimately led to the domestic substitution of the core areas of underlying software from a concept to a market direction, and its penetration rate has shown explosive growth in the past few years.

From a commercial perspective, the trend of domestic substitution has essentially created a huge market increment for domestic products, and its growth space has also multiplied.

At the same time, all kinds of underlying software products also began to go out of the Comfort zone of large factories and began to participate in market-oriented competition. At present, OceanBase has started serving multiple core areas related to data security, such as finance, communication, and transportation.

With the continuous infiltration of digitalization into the physical industry, the accumulation of technology by many internet companies has also begun to show value. Whether it is databases, cloud computing, or various video conferencing and office software, they are gradually breaking away from the business system of large factories and playing the role of the chassis of the digital economy and an autonomous and controllable infrastructure.

On June 1, Ant Group, which had not made a public appearance for a long time, released a "Sustainable Development Report of Ant Group in 2022". In addition to detailing the company's many actions in the ESG field, Ant also disclosed some key financial indicators. Throughout 2022, ants invested 20.46 billion yuan in technological research and development,

Based on this calculation, Ants' research and development expenses have already ranked among the top among major domestic internet companies.

The advantages of internet companies include not only their own profitability, but also stronger talent attraction and capital operation capabilities. Whether it's internal incubation or external investment, the exploration of high-tech fields is going from point to surface.

In 2022, Tencent announced that it will invest 10 billion yuan in 10 years to support China's basic science innovation research and launch the New Cornerstone Science Foundation to support scientific research innovation; Baidu has a full stack layout in the field of artificial intelligence, with an investment of up to 100 billion yuan over the past 10 years; Meituan Longzhu, which previously focused on investment in the consumer sector, has also shifted its focus to hard technology fields such as robotics, semiconductors, and autonomous driving since 2021.

The more core technology it is, the more it requires continuous exploration, failure, and reflection. Exploring the pinnacle of technology has never been a smooth sailing adventure, but once you start, everything is not too late.

Author: Yuanchuan Research Institute

Visual design: Shurui

Editor in charge: Li Motian