OpenAI, which was “raised” by Microsoft, decided to bite back Microsoft

Original source: Future Science and Technology Power

Author:ViniWang

Image source: Generated by Unbounded AI‌

The competition is close to a showdown.

On August 29, OpenAI released the enterprise version of ChatGPT. Following the competition for individual users, OpenAI competed head-on with Microsoft for enterprise users.

As a result, the "dissension" between Microsoft and OpenAI is no longer hidden by both parties, and the support story that has been shaped into a good story by both parties quickly enters the stage of direct competition.

The rift between OpenAI and Microsoft dates back to November last year.

Originally, Microsoft was busy integrating ChatGPT into its own series of software such as Microsoft Bing. Unexpectedly, Open AI released the public beta version of ChatGPT in advance and gained 100 million users within two months. According to media reports, this caused dissatisfaction among Microsoft executives because its own software was overshadowed by ChatGPT.

On August 29, OpenAI "repeated its old tricks." OpenAI launched the enterprise version of ChatGPT, which coincided with the time when Microsoft began to support the Bing Chat enterprise version in the Windows Copilot preview version, and the functions are also close. OpenAI’s strategy seems to be precisely aligned with Microsoft at every step.

The reason is simple. At the strategic priority level, if your opponent provides more valuable functions earlier than you do, you will lose your first-mover advantage in this field. **The release of the enterprise version of ChatGPT is a sign that OpenAI is reaching into the hinterland of Microsoft. **

With the help of Microsoft's business structure, it is not difficult to understand why OpenAI focuses on enterprise services.

Windows Copilot is part of Office365 and belongs to Productivity and Business Processes. Among them, Office 365 is the product of a new business model adopted by traditional Office. By learning from Salesforce paid subscription, it has become Microsoft's fastest growing business. In the 2018 Q4 financial report, enterprise-level Office 365 service revenue increased by 38%, and consumer-level Office service revenue increased by 38%. %, contributing $17 billion in revenue to Microsoft.

It is also Microsoft's most profitable business unit. Microsoft's financial report for the fourth quarter of fiscal year 2023 shows that this department generated revenue of US$18.3 billion for Microsoft, and 1/3 of Microsoft's total revenue comes from this business. Despite the slowdown in Microsoft's personal computing business, the business grew by 10% year-on-year, with Office business product revenue alone growing by 15%.

This logic exactly explains why OpenAI wants to follow Microsoft's strategy - because it wants to compete with Microsoft for users with the most commercial value and the easiest way to grow in the most profitable and stable market.

There are three categories of market users: enterprise users, individual users, and small and medium-sized enterprise start-up company developer ecosystem users. The most price-sensitive among them are the small and medium-sized enterprise developer ecosystem users and individual users. Their base is relatively large, but their payment is unstable because they sometimes turn to competitors out of the temptation of low prices. In contrast, enterprise users are the least sensitive to price, have strong demand and willingness to pay, have the most stable payment ability, and have the greatest commercial value.

The stock price changes before and after Microsoft announced the pricing of Microsoft 365 Copilot can also reflect the premium space for "enterprise users". On July 18, after Microsoft announced pricing of $30 per user per month, the stock price rose 6% to as high as $366.78.

In fact, Microsoft seems to have already anticipated that competition will occur here. After all, this is the company it supports and the area it is best at. Previously, it had begun to make various defensive actions, and the methods were equally precise, targeting OpenAI's initial "seed customers", the small and medium-sized enterprise developer user group.

You may be curious, for developers, Microsoft and Open AI, one is an API supplier and the other provides cloud servers, what conflict of interest can they have?

Readers who understand the product development process know that if small and medium-sized enterprise developers develop a product and need to write code, they need to purchase a cloud server that hosts the product code.

If the product developed involves a specific function, you need to pay the API provider to purchase the API interface required to call the function. Sometimes cloud server providers will also sell APIs together, but usually the two companies make money together through some kind of agreement.

Now imagine this situation: If the cloud server provider happens to bundle a certain popular API interface, without giving the API interface supplier a profit share, it takes advantage of the demand of small and medium-sized enterprise developers for this popular API interface and in turn promotes it. Your own cloud server increases your own cloud server sales, but in turn affects the API provider's profits. What should you do?

You may say that it is very simple. API providers no longer authorize their own popular API interfaces to cloud servers, and no longer let cloud server providers suck their blood. They are the only ones who are beautiful. They sell their own APIs and let cloud server providers walk independently. Isn’t that enough?

But the problem is not that simple, because of these two technology suppliers, the API supplier is OpenAI, and the cloud server supplier is Microsoft. They cannot unbundle because Microsoft is a 49% shareholder of OpenAI. Since the birth of OpenAI, they have There is a deeply binding strategic alliance.

Before OpenAI became famous with ChatGPT in December last year, Microsoft spent huge sums of money to help Open AI get through the long night of no revenue but urgent need to continue to burn money. In July 2019, Microsoft made a strategic investment of 1 billion in OpenAI and provided OpenAI with servers with powerful computing power needed to support large model development. All OpenAI technologies run on Azure cloud servers.

**There is no free lunch in the world. The price OpenAI needs to pay for this $1 billion is to allow Microsoft to sell its API through its Azure cloud business. **

According to "The Information", after cooperating with OpenAI, Microsoft's cloud service Azure OpenAI can directly call OpenAI models, including ChatGPT, Codex and DALL.E. Customers of Microsoft cloud services can use chat robots and search engines without going through OpenAI. Use these large AI models in other products.

This is a foreshadowing that the two companies will compete sooner or later. Originally, developers from small and medium-sized enterprises purchased APIs through OpenAI and accessed OpenAI's large models. But now, Microsoft not only provides the same API that can access the Open AI large model, but also bundles its own Azure cloud server. When two companies sell the same thing, competition begins and discord arises.

According to The Information, in March this year, an internal Microsoft document showed that Microsoft required its sales staff to "take advantage of the market's enthusiasm for the OpenAI model" to promote Azure OpenAI and count Azure OpenAI service transactions into the assessment requirements of Azure sales staff.

Microsoft instructs sales staff to intervene in potential customer selection, telling potential customers that OpenAI's standalone access rights are "ideal for experimentation" but have "limited enterprise-grade capabilities" and fewer "security/privacy features." Claiming that "Azure OpenAI can provide the same model as OpenAI".

This move affected the choice of developers of small and medium-sized enterprises and caused the loss of OpenAI developer users. Then OpenAI began to fight back, first restricting Microsoft's use of some models. On March 1 this year, OpenAI began selling the right to use its speech recognition model Whisper, but the model is not provided by the Azure OpenAI service. On March 14, OpenAI released the latest GPT-4 language model. Microsoft has not yet announced plans to provide GPT-4 on Azure.

**In line with the principle of "the enemy of my enemy is my friend", OpenAI and Microsoft have also cooperated with their respective competing products. Following in March this year, OpenAI established a cooperation with Microsoft competitor Salesforce to provide chatbots and generated artificial intelligence services for its business software. Not to be outdone, Microsoft directly defended OpenAI’s competitors. In July this year, Microsoft, as the first partner, appeared at the launch of Meta's next-generation open source large language model Llama 2. **

OpenAI's response killed two birds with one stone: On August 23, OpenAI launched a fine-tuning function based on GPT-3.5 Turbo and updated the open API, using more convenient functions to compete with the open source Llama2, while also continuing to compete with Microsoft for small and medium-sized enterprise developers.

Behind these increasingly explicit competitions is the revenue pressure brought about by the huge costs of large models.

Oren Etzioni, a board member and former CEO of the Allen Institute for AI, a nonprofit research organization, said, “Both sides need to make money, which makes them more prone to conflict, and the point of conflict is that they are trying to use something like products to make money.”

For OpenAI, the operating cost of large models is high, which is an industry consensus.

According to a recent research report by an Indian news media platform in July this year, OpenAI costs about US$700,000 a day just to run ChatGPT. In fiscal year 2022, OpenAI’s total revenue will be only US$30 million. The meager revenue is a drop in the bucket compared to the high operating costs - this year's revenue alone is only enough to maintain OpenAI operations for 42.85 days.

Meager revenue is not enough to cover development costs. According to "The Information", including the capital costs of poaching talents from Google, OpenAI's total development costs last year were as high as US$540 million. Although it received an additional investment of US$10 billion from Microsoft last year, the money seems to be worthless in the face of high operating costs and development costs.

Not only that, don’t forget that according to the investment agreement signed between OpenAI and Microsoft, before Microsoft recovers its investment costs, it must obtain 75% of the profits from OpenAI’s sales.

OpenAI seems to be in a hurry. It has decided to counterattack Microsoft, and it also wants the outside world to know that it has this ability. The recent series of actions are very thought-provoking. The company, which once kept everything strictly confidential, is beginning to have a lot of information revealed. For example, according to The Information, OpenAI is currently expected to earn more than $1 billion in the next 12 months from selling artificial intelligence software and the computing power to power it. income.

Sam Altman wants people to know that he is not being held hostage by Microsoft and that he can still make money in Hezonglianheng.

**And another news released by Sam Altman not long ago is also very interesting. He announced on his social account that on November 6, OpenAI will hold its first developer conference in San Francisco - this was once a patent for large platforms such as Microsoft and Google to showcase their ecological capabilities, but now Altman wants everyone to know , your own company can be like them. **