Microsoft AI boss confirms development of “off-frontier” AI models, but they’ll be 3 or 6 months behind OpenAI: “Our strategy is to really play a very tight second”

by

Towards the start of this year, Salesforce CEO Marc Benioff forecasted that Microsoft could break free from a heavy dependence on OpenAI for its artificial intelligence capabilities. He suggested that Microsoft might be currently working on creating cutting-edge AI models independently.

Not long after OpenAI introduced the Stargate project, a $500 billion initiative aimed at building data centers nationwide for its AI endeavors, it was reported that Microsoft had reservations. Besides aiming for greater autonomy from OpenAI, Microsoft allegedly expressed concerns about the costliness and slow pace of OpenAI’s GPT-4 model in fulfilling consumer demands.

An independent source confirmed that Microsoft is developing its unique AI systems. After the unveiling of OpenAI’s massive $500 billion Stargate initiative, Microsoft no longer held the position of exclusive cloud provider and largest shareholder, as

In the recent financing round, SoftBank took the lead, contributing a massive $40 billion. This investment significantly increased the market value of the company behind ChatGPT, setting it at an impressive $300 billion.

Despite limited information on Microsoft creating their own AI models, Microsoft AI CEO Mustafa Suleyman shared additional insights about their progress in the field during an interview with CNBC’s Steve Kovach. He suggested that developing AI models a few months late offers certain benefits.

To clarify, the executive emphasized that by creating models in this manner, it lowers the overall cost of development, thereby enabling the company to concentrate on particular application scenarios.

According to Microsoft’s AI CEO:

It’s more cost-effective to provide a solution after waiting for the initial three or six months, once the frontier has moved first. We refer to this as ‘off-frontier.’ Essentially, our approach is to take a patient stance and strategically position ourselves, considering the high capital requirements of these models.

At Microsoft’s 50th anniversary and Copilot event, they unveiled numerous new features set to roll out for Copilot, such as Copilot Vision, Deep Research, Pages, Copilot Avatar, and Memory. This suggests that Microsoft intends to exert greater control over the AI technology they’ve incorporated into their tech infrastructure. The most effective method to achieve this seems to be by creating their own AI models.

Suleyman continues:

It’s essential for us at Microsoft that we become self-reliant in AI technology over the long term. I usually consider these matters on a timescale of five to ten years, with 2030 as a minimum goal. For now, we are closely collaborating with OpenAI, who have had a very successful partnership with us. While Microsoft is working on developing its own AI internally, we are not in a rush to create the most advanced models.

Our AI team is exceptionally skilled, we have abundant computational resources at our disposal, and it’s not essential for us to create the cutting-edge model or the best one globally from the get-go. That would be excessively costly and unnecessary as it could lead to unnecessary duplication in the field.

As an analyst, I find myself eagerly anticipating the strategies Microsoft will employ to address the escalating challenges in artificial intelligence (AI) as it continues its ascent towards higher altitudes, requiring increasingly substantial computing resources, graphics processing units (GPUs), and cooling systems.

Read More

2025-04-07 13:39