Source: Deep Web · Tencent News
Author: Hu Shixin
On May 6th, Beijing time, OpenAI announced that the company will continue to be controlled by a non-profit organization, the existing for-profit entity will be transformed into a public benefit corporation, and the non-profit organization will control this public benefit corporation and become an important shareholder.
Just a few days ago, OpenAI CEO Sam Altman posted a photo with CEO Satya Nadella on social platform X and captioned that they were discussing some recent developments. $Microsoft (MSFT.US)$ In the comments section, Nadella responded: "It was great to see you today, I really like the new office."
Looking back over the past six years, Microsoft has invested over 13 billion dollars in the company OpenAI; due to its deep ties with OpenAI, Microsoft is perceived as an imaginative veteran Technology company and even briefly reached the top of the Global Market Cap.
Ultraman once stated that OpenAI and Microsoft have the "strongest partnership in the Technology industry." During their closest period, Nadella would send Ultraman five to six text messages every day, and Ultraman would reply to each one.
However, according to foreign media reports on April 29, this "Silicon Valley marriage" is now in trouble. People familiar with the relationship revealed that the two CEOs have increasingly escalating conflicts over issues such as the computing power provided by Microsoft to OpenAI, OpenAI's permission for Microsoft to use its models, and whether OpenAI can quickly develop General AI.
At the same time, Nadella has prioritized increasing the sales and usage of Microsoft's self-developed AI model Copilot, which is a competitor to ChatGPT, and last year hired an AI expert to secretly initiate the development of a new model.
According to foreign media reports, nowadays, the text message exchanges between Ultraman and Nadella have also decreased, as they now mainly communicate through their scheduled weekly phone meetings.
Microsoft is reducing the risk of dependence on OpenAI.
The collaboration between Ultraman and Nadella began in the summer of 2018, when they met in a stairwell during an annual meeting. According to Ultraman, he took only five minutes to introduce Nadella to OpenAI's fundraising plan.
A year later, Microsoft invested 1 billion USD in OpenAI. This investment also granted Microsoft exclusive rights to use OpenAI technology and made it its only Cloud Computing Service provider.
OpenAI launched its groundbreaking product—ChatGPT—in November 2022. In early 2023, Microsoft invested another 10 billion USD, enabling OpenAI to rent Datacenter from Microsoft to train new models. In November of the same year, Ultraman was briefly removed by the Board of Directors, and Nadella proactively offered to hire him.
According to foreign media reports, this incident is referred to internally at OpenAI as an 'unexpected episode.' However, for Nadella, this episode serves as a reminder that he must develop a 'contingency plan' for Microsoft's dependence on Ultraman and OpenAI.
Later, Nadella recruited Mustafa Suleyman, one of the three co-founders of Google DeepMind.
Suleyman dropped out of Oxford University at the age of 19 and co-founded DeepMind with two friends in 2010, which was later acquired by Google in 2014. In 2019, Suleyman left DeepMind entirely and began focusing on Google’s AI-related work. In 2022, he left Google again to start the AI Venture Inflection AI.
Ultimately, Microsoft spent $0.65 billion to successfully bring Suleyman and his colleagues from Inflection into the company. According to informed sources, Suleyman is working on developing a large language model aimed at competing with the advanced model GPT-4 that OpenAI had released at the time.
However, the project got off to a rocky start. An early training test showed that constructing a model comparable to OpenAI's was more challenging than anticipated.
But according to the latest news, Suleyman's team has developed an AI model family codenamed 'MAI', which has demonstrated performance on par with products from industry leaders such as OpenAI and Anthropic in internal benchmark tests. Additionally, they are also developing reasoning models similar to ChatGPT -o1.
Microsoft has already begun to replace the OpenAI models in Copilot with these MAI models and is considering releasing the MAI models in API form for external developers later this year.
'Microsoft cannot rely solely on OpenAI technology, or it may be left behind,' said D.A. Davidson Analyst Gil Luria. 'This is a real competition, and OpenAI may not win.'
另外，微软已开始多元化其AI供应链。在苏莱曼的指导下，微软正在测试来自Anthropic、xAI、DeepSeek和 $Meta Platforms (META.US)$ 等公司的模型，作为Copilot的潜在替代品。而这种“多厂商”战略旨在降低对OpenAI的依赖风险。
OpenAI也开始“去微软化”
而据外媒报道，OpenAI与微软之间的重要矛盾之一还集中在双方云计算合作协议上。
作为OpenAI的最大投资方，微软掌握着OpenAI模型训练的算力“命门”——OpenAI的大部分电力和数据中心算力均由微软提供。根据合同，微软不仅是OpenAI的独家云服务器供应商，还是唯一获准向云计算客户转售OpenAI模型的公司。
目前，微软已向OpenAI投入超过13 billion美金，获得其20%的股权，并有望获得未来最高达92 billion美金的利润。旗下Azure是OpenAI的独家云服务提供商，年支出费用超过1 billion美金。
此外，微软可以使用OpenAI开发的所有技术，并将其集成到产品Copilot中；Azure则垄断了OpenAI模型向云计算客户的转售权，并向OpenAI支付20%的收入。OpenAI则负责训练新模型（包括GPT-4、Sora等），并通过ChatGPT订阅和销售API获利。
As the demand for computing power from OpenAI continues to rise, Ultraman keeps asking Microsoft for more funding and computing resources. However, since the second half of 2023, Microsoft's responses to OpenAI's requests have become perfunctory. They stated that they have provided everything they can and have relaxed the exclusivity agreement.
OpenAI is dissatisfied with this, believing that Microsoft is unable to meet its server requirements. Subsequently, Ultraman began seeking new investors, frequently visiting Middle Eastern countries and negotiating partnerships with sovereign funds from nations such as Saudi Arabia and the UAE.
In 2024, Ultraman collaborated with Masayoshi Son to develop AI Chips. In October of the same year, OpenAI completed a fundraising of 6.6 billion USD, with investors including Thrive Capital,$NVIDIA (NVDA.US)$the UAE Sovereign Wealth Fund and SoftBank, among others. Although Microsoft also invested 0.5 billion USD, it no longer holds a dominant position.
Additionally, Ultraman bypassed Microsoft to directly promote OpenAI's service products. Insiders revealed that during promotional events, he emphasized that companies working directly with the OpenAI team could gain access to the latest models and more customized AI products.
In January of this year, Ultraman, along with Trump, Masayoshi Son, and Larry Ellison, announced the establishment of the 'Interstellar Gateway' joint venture at the White House, with total investments expected to reach 500 billion USD. This project is similar to Ultraman's original plan to collaborate with Nadella, who was attending the World Economic Forum in Davos at that time.
According to insiders, Microsoft and OpenAI had discussed a Datacenter project, but after Ultraman's brief dismissal, Microsoft shelved the project.
OpenAI announced that by 2030, the "Stargate" Datacenter, led by SoftBank, will support three-quarters of its computational power needs. In the upcoming $40 billion financing, SoftBank will invest at least $30 billion, nearly half of which will be allocated to the "Stargate" project.
Additionally, OpenAI expects this year's company revenue to more than double from last year's, increasing from $3.7 billion to over $12.5 billion, with one-third coming from SoftBank's spending on AI agent tools.
When exactly will AGI be achieved?
The estrangement between Microsoft and OpenAI is inevitable. Another major contradiction between the two companies is: when exactly can OpenAI achieve AGI?
According to the agreement between OpenAI and Microsoft, once OpenAI successfully develops AGI, its Board of Directors will have the right to readjust the cooperative relationship with Microsoft.
Altman has repeatedly claimed that OpenAI is still some distance from achieving AGI and cannot do so within one or two years, but he also believes that "the arrival of AGI will happen sooner than most people expect." In Altman's view, the industry has touched upon the "known unknowns," and the overall direction is clear; what is needed now is rapid advancement.
According to informed sources, during private closed-door negotiations, Microsoft's negotiators told OpenAI that the current technology is still far from achieving AGI. Previously, Nadella dismissed the hype around AGI milestones in a popular podcast, arguing that such claims are unfounded and merely toy with benchmark testing.
Nadella believes that the development of AI should not overly pursue AGI milestones.
In December last year, foreign media exposed a "secret agreement" stating that when OpenAI develops an AI system capable of generating at least 100 billion US dollars in profit, it would mark the arrival of the AGI era. However, the document also stipulates that the realization of AGI still requires the OpenAI Board of Directors to determine it based on "reasonable discretion."
In fact, OpenAI is still in a loss-making state.
On one hand, since the departure of OpenAI's co-founder and chief scientist Ilya Sutskever, there have been concerns about the lack of a technical soul figure at OpenAI. Moreover, almost every month, technical talents announce their voluntary resignations, some joining other companies, some choosing to start their own ventures, and others returning to academia.
On the other hand, from the continuous twelve-day live streaming event at the end of last year, OpenAI did not release any stunning model works like before (the only released o3 model was revealed to have cheating behavior), leading people to begin questioning whether OpenAI has run out of talent.
Foreign media analysis pointed out: the sudden rise of DeepSeek at the beginning of the year has left OpenAI somewhat flustered.
Although the ChatGPT 4o model led a wave of "Ghibli-style" craze in image generation, it cannot fulfill people's expectations for AGI. AGI has still not been realized, but Microsoft and OpenAI are growing increasingly distant.
Editor/rice
