Microsoft, Google, and Amazon are starting the cloud war in the era of large models

Author: Wan Chen Editor: Zheng Xuan

Image source: Generated by Unbounded AI

With the tightening of cloud spending on Internet enterprise software, slowing growth has gradually become a dark cloud over the heads of cloud vendors.

The birth of ChatGPT broke this bottleneck, and AI will reshape software. Customers of cloud vendors—software companies are actively embedding the AI capabilities brought by large models into existing workflows to complete higher-level automation.

With the gradual drying up of new cloud customers, software companies no longer go to the cloud for the sake of going to the cloud, but strive to use AI to improve productivity. ** "This is the biggest increase in the cloud computing market in the next ten years. Computing power infrastructure is the absolute beneficiary of the big model." **A person in the cloud computing industry who has been in the industry for more than ten years explained to Geek Park.

Under such a prospect, several major overseas cloud service giants—Microsoft, Amazon, Google, and Oracle quickly made changes. **In the past few months, cloud giants have spent real money to develop large-scale models, strategic investments, and self-developed AI chips... The era of large-scale models is in the ascendant, and they have already targeted a new generation of AI software customers. **

The past is far from unbreakable, the cloud market is rapidly reshuffled, and the giants have opened up a new curtain of competition.

After all, the decline of the big brother in the mobile Internet era is imminent. In a few years, Nokia has gone from 70% of the mobile phone market share in its heyday to no one cares about it. It is only between the thought of making a wrong decision. As for the large model, the cloud industry quickly reached a consensus: this time AI is by no means a small variable. Judging from the rapid development speed of the industry, the current leading players may also be left behind.

**Half of 2023 has passed, this article will sort out several major overseas cloud giants, what is the key to the competition among cloud vendors today? **

01 Research and development of AI-specific chips, you can't give all your "life" to Nvidia

After the advent of the era of large models, for cloud service providers, the most scarce resource today is computing power, or AI chips. **Investing in the underlying infrastructure - AI acceleration chips, has also become the first focus of competition among cloud vendors today. **

Scarcity and high cost are considered to be the primary reasons for cloud vendors to speed up self-developed chips. Even powerful figures in the technology circle like Musk commented that "this thing (Nvidia GPU) is more difficult to deal with than medicine", and secretly bought 10,000 cards from Nvidia for his AI company X.AI, and also received Lot of idle equity in Oracle.

This degree of scarcity is reflected in the business of cloud giants, which directly corresponds to the business loss caused by "stuck neck". Even Microsoft, which is the first to act first, has been exposed to rumors that due to the shortage of GPUs, the internal AI R&D team implements a GPU rationing system, various new plans are delayed, and new customers have to queue for months to go to Azure.

Even venture capital institutions have to rely on Nvidia chip inventory to grab projects. For the sake of N cards, the forces of all parties have reached the point where "everything is used".

**Another name for scarcity is expensive. **Considering that the large model requires more than ten times the computing power, the card will only be more expensive. Recently, an investor told Geek Park, “At the beginning of the year, the A100 single card was 80,000, but now it has been sold to 160,000, which is still out of reach.” Correspondingly, the tens of thousands of cards of the cloud giants have to pay The "Nvidia tax" will only be an astronomical figure.

It's hard to feel that "fate" hangs in the hands of others, and Microsoft, which is the most popular, has the most say. A month ago, The information exclusively reported that Microsoft established a 300-person "Tiantuan" to accelerate the pace of self-developed AI chips. The server chip code-named Cascade may be launched as early as next year.

Not only because of the "stuck neck", cloud manufacturers' self-developed chips, but also another layer of meaning-GPU is not necessarily the most suitable chip for running AI, and the self-developed version may optimize specific AI tasks.

Admittedly, most of the current advanced AI models are powered by GPUs, because GPUs are better at running machine learning workloads than general-purpose processors. **However, GPUs are still considered general-purpose chips, not truly native processing platforms for AI computing. **As Yuanchuan Research Institute pointed out in "A Crack in the Nvidia Empire", GPUs are not born for training neural networks. The faster artificial intelligence develops, the more these problems will be exposed. Relying on CUDA and various technologies to "magic change" scene by scene is an option, but it is not the optimal solution.

Amazon, Google and Microsoft have been developing chips known as ASICs — application-specific integrated circuits — that are better suited for artificial intelligence. The Information interviewed multiple chip industry practitioners and analysts and concluded that Nvidia GPUs helped train the model behind ChatGPT, but ASICs generally perform these tasks faster and consume less power.

As shown in the figure above: Amazon, Microsoft, and Google have all raised the importance of in-house self-developed chips, and developed two types of chips for the data center sector: standard computing chips and chips dedicated to training and running machine learning models. These models can power chatbots such as ChatGPT.

At present, Amazon and Google have developed customized ASICs for key internal products, and have provided these chips to customers through the cloud. Since 2019, Microsoft has also been working on developing custom ASIC chips to power large language models.

Some chips developed by these cloud providers, such as Amazon's Graviton server chips and AI-specific chips released by Amazon and Google, are already comparable in performance to chips from traditional chipmakers, according to performance data released by cloud customers and Microsoft. Google TPU v4 is 1.2-1.7 times faster than Nvidia A100, while reducing power consumption by 1.3-1.9 times.

02 Strategic Investment Competition: Giants spend money to "buy customers"

In addition to research and development of chips, the second key point in the competition of several major overseas cloud giants is to invest in foreign strategic investments to grab AI customers and AI projects. **

Compared with venture capital, the strategic investment of giants has an absolute advantage. The combination of OpenAI and Microsoft serves as an excellent example, opening a precedent for large-scale models and strategic investment. This is because the resource barriers required for large models and related applications are extremely high. Only money, limited money, is not enough to grab AI projects. After all, Google, Microsoft, AWS, Oracle or Nvidia can not only write huge checks, but also provide scarce resources such as cloud credits and GPUs.

From this perspective, grabbing projects and grabbing customers all happen among cloud giants, and there are no other rivals. They are playing a new game — seeking promises from AI companies that they will use their cloud services rather than those of competitors.

Microsoft is sitting on the position of OpenAI's exclusive cloud service provider. While paying a huge cloud bill for OpenAI, Microsoft has exchanged a series of enviable rights such as OpenAI's equity and product priority.

**Microsoft's rivals are also scrambling to win over other AI customers. **These cloud providers offer deep discounts and credits to AI companies to win their business. Critics have pointed out that this is akin to buying customers, although the practice of taking equity in future or current customers is not uncommon in the enterprise software world.

Oracle has also offered computing credits worth hundreds of thousands of dollars as an incentive for AI startups to rent Oracle cloud servers, The Information previously reported.

Google may be the most active of these major cloud vendors, offering AI startups a combination of cash and Google Cloud credits in exchange for equity. Earlier this year, Google invested $400 million in Anthropic, one of OpenAI's main entrepreneurial challengers. Google Cloud said in February that it had become Anthropic's "preferred" cloud provider.

Recently, Google invested US$100 million in Runway, an AI company in the field of "Vensheng Video". But before that, Amazon AWS touted Runway as a key AI startup customer. In March of this year, AWS and Runway announced the establishment of a long-term strategic partnership, becoming its "preferred cloud provider." Now, Runway seems to be one of the "pawns" in Google's duel with Amazon, because Runway is also expected to rent cloud servers from Google.

Earlier, Google Cloud also announced the establishment of partnerships with two other popular AI companies, namely: Midjourney in the Vincent graph field and the chat robot App Character.ai, which was previously a key cloud customer of Oracle.

It's too early to tell whether these deals will help Google catch up with its larger cloud computing rivals AWS and Microsoft, but Google Cloud is on the mend.

Of the 75 (AI) software companies in The Information database, Google provides some cloud services to at least 17 companies, more than any other cloud provider. Amazon is not far behind, with at least 15 companies using AWS for cloud computing. Microsoft and Oracle provide cloud services to six companies and four companies, respectively. Of course, using multiple clouds is also a habit in the industry, and at least 12 of these 75 companies use a mixture of multiple cloud providers.

03 large model is the real key to winning or losing

Computing power and combat investment are the high ground that must be contested in the early stage of this cloud war. But in the long run, the big model is the real key to the success of market competition.

Microsoft's ability to become a leader is due to the cooperation with OpenAI. Coupled with the excellent engineering capabilities of the Microsoft team, GPT-4 was embedded in Microsoft's "family bucket" within a few months. In the past six months, Microsoft first took advantage of the priority use of OpenAI products and the price reduction of enterprise software products to seize more cloud markets. Then rely on the product line upgrade to Microsoft 365 Copilot to increase the price to obtain greater revenue.

According to Yunqi Capital's research, Microsoft's underlying model basically relies on OpenAI, and after accessing the large model, Microsoft began to package and sell Teams, Power BI, Azure and other application layer products at a lower price.

Microsoft Chief Financial Officer Amy Hood told investors in April that OpenAI will generate revenue for Azure as more people start using its services.

New reports indicate that Microsoft is charging some Office 365 customers a 40% premium to test AI capabilities that automate tasks like writing text in Word documents and creating PowerPoint slides, and at least 100 customers have paid up to $100,000 fixed fee. Data shows that within a month of its launch, Microsoft earned more than $60 million in revenue from Microsoft 365 Copilot's AI capabilities.

**In stark contrast to Microsoft, the once leader Amazon Cloud, one step behind in the big model, is facing more severe challenges today. **

AWS was an early developer of AI cloud services, and it has been in place since around 2016. But customers don't find the services, which include facial recognition, converting text to lifelike speech, and primitive forms of chatbots for tasks like customer service, to be very useful. AWS also launched SagaMaker, an AI digital tool for engineers in 2017, to help them develop and use machine learning models, which once became AWS's most important AI product.

However, in the following years, AWS's AI products failed to keep up with the wave of large language models. Since November 2021, Microsoft has started selling AI products developed based on the GPT series of models for use by enterprise customers. At the same time, Google has also snatched up major AI startups as cloud customers and sold proprietary AI software to its cloud customers. Even cloud computing laggard Oracle has its own advantages in providing computing resources to AI startups.

AWS, belatedly, is trying to catch up. In April, it announced a cloud service that allows customers to use large models from Stability, Anthropic and AI 21 Labs as bases for their own products. In return, AWS will share a portion of the revenue with these partners.

At the 2023 Google I/O conference, CEO Sundar Pichai introduced Google's latest AI progress | Image source: Google's official website

**Google got up early, but caught up late. **As a major manufacturer with the deepest accumulation in the field of large-scale models, Google’s reaction after the release of ChatGPT is not unpleasant. It quickly released the conversational intelligent robot Bard and a new generation of large-scale language model PaLM 2 as a response. As a result, the press conference directly overturned , The subsequent product release speed is not ideal, which is in stark contrast to Microsoft's strong engineering capabilities.

**The last thing worth mentioning is that Oracle, which fell out of the forefront of the cloud market very early, unexpectedly has a tendency to counterattack in this wave of upsurge. **

Oracle, long a laggard in the cloud space, has been surprisingly successful in leasing cloud servers to high-profile AI startups that compete with OpenAI. That's partly because Oracle Cloud can run complex machine learning models more economically than Amazon Web Services or Google Cloud, The Information reported.

Oracle Cloud's approach to the AI race appears to be similar to that of AWS, which develops its own AI software to sell to customers, but will also sell access to open-source AI software and products from other AI developers.

In addition, some people familiar with the matter revealed that Oracle has begun testing OpenAI's products to enrich its product line for B-end customers, including human resources and supply chain management software, but Oracle is more likely to develop its own software for this purpose. features help Oracle customers quickly generate job descriptions and schedule meetings between recruiters and candidates, though the company is still deciding which products to improve first.

Reference materials:

"A Crack in the Nvidia Empire" | Yuanchuan Research Institute

Big factory and big model: real business is the last word｜"Yunqi FutureScope"

Google and Microsoft’s Other AI Race: Server Chips｜The information

Skepticism Rises Over AWS’s AI Strategy｜The information

Google, Nvidia and Microsoft Offer What VCs Can』t｜The information

Pro Weekly: Generative AI Spurs Cloud Demand—and Competition｜The information

Google Invests in AI Startup Runway to Wrest Cloud Business From AWS｜The information

Microsoft Is Charging Some Office 365 Customers 40% Extra to Test AI Features｜The information