Internal tensions and lagging user numbers are reportedly defining the first year for Mustafa Suleyman, the high-profile executive Microsoft tapped to lead its consumer AI charge.
Hired in March 2024 as CEO of Microsoft AI, Suleyman, co-founder of Google DeepMind, arrived alongside much of his team from the AI startup Inflection. The complex arrangement, structured primarily as a $620 million non-exclusive licensing agreement for Inflection’s models plus around $30 million for hiring rights rather than a standard acquisition, was intended to bolster Microsoft’s own AI development and reduce its heavy reliance on partner OpenAI. Yet, over a year later, Microsoft’s consumer AI tool, Copilot, appears stuck.
Data presented internally by Microsoft CFO Amy Hood and cited by Newcomer painted a stark picture: Copilot’s weekly active users were stagnant at roughly 20 million.
Meanwhile, OpenAI’s ChatGPT was shown rocketing towards 400 million weekly users during the same period, according to the reporting in the original source material provided earlier. This persistent gap highlights the uphill battle for Suleyman, despite Microsoft’s significant investment and a flurry of feature releases under his purview.
External web traffic analysis by aitools.xyz reinforces this, showing ChatGPT visits dwarfing Copilot’s by a factor of over 50 daily, capturing a commanding market share at the time.
Microsoft’s move to secure Suleyman and the Inflection team stemmed partly from dissatisfaction with the performance of earlier AI integrations like the revamped Bing chatbot, and a desire for greater independence following leadership turmoil at OpenAI in late 2023.
The objective was clear: reignite internal efforts after years of focusing resources externally on OpenAI. However, bringing in an outsider like Suleyman, whose previous tenure at Google ended amidst controversy over management style, bred skepticism among some Microsoft veterans familiar with the company’s distinct culture.
Internal Model Development and Team Dynamics
A core part of Suleyman’s mandate involved building capable in-house AI models. His team, often referred to as MAI and led by Inflection co-founder Karén Simonyan, began work on a large model dubbed MAI-1.
Reports from May 2024 suggested MAI-1 aimed for around 500 billion parameters—a measure of model complexity—potentially leveraging Inflection’s data but distinct from Microsoft’s smaller Phi models. Such a size would place it well above Microsoft’s earlier open-source efforts but still potentially below OpenAI’s GPT-4. However, sources indicate MAI-1 training runs encountered performance issues, sparking internal discord.
A specific flashpoint involved a dispute with the team led by Sebastien Bubeck, then VP of Generative AI research, developer of the Phi models built using synthetic data (AI-generated training data). After a joint training run yielded underwhelming results, Suleyman’s co-founder Simonyan reportedly attributed the failure to “contamination” by the synthetic data, igniting a debate over the data’s effectiveness that spilled onto public Slack channels.
Following this, Bubeck’s team was reorganized, and in October 2024, Bubeck departed for OpenAI, reportedly to focus on Artificial General Intelligence (AGI) research. Microsoft acknowledged the departure, stating they “look forward to continuing our relationship through his work with OpenAI.” This internal friction underscores the challenges in integrating new teams and approaches within Microsoft’s established structure.
The OpenAI Partnership: A ‘Sibling Rivalry’
The relationship with key partner OpenAI has also seemingly been complex. Suleyman himself, in an October 2024 podcast appearance, characterized it as a “good-natured sibling rivalry,” admitting to “squabbles” but asserting “largely we’re on the same team.”
However, reports from the same period, detailed rising tensions. Allegations surfaced that Microsoft withheld compute resources from OpenAI following the November 2023 board incident, possibly to bolster its own competing tools. The NYT report also corroborated earlier accounts of Suleyman clashing with OpenAI staff, including an incident where he allegedly yelled at an employee over delayed technology transfer, and noted that his appointment had angered OpenAI CEO Sam Altman and others.
Despite these undercurrents, Suleyman stated in an April 2025 interview with CNBC that Microsoft’s strategy involves being a “tight second” follower, lagging 3-6 months behind frontier model leaders like OpenAI to optimize costs and focus on specific use cases, while affirming the partnership remains deep “until 2030 at least.”
He also pushed back on user volume as the sole metric, claiming in an April 2025 interview reported by Reuters that his team prioritizes “SSR, the rate of successful sessions,” which he asserted had “gone up dramatically,” based on analyzing anonymized logs, though he did not provide specific figures.
Copilot’s Feature Cadence Continues Amid Strategic Shifts
While usage metrics lag, Suleyman’s division hasn’t been idle on the product front. Throughout early and mid-2025 (based on the hypothetical timeline), Microsoft announced numerous Copilot enhancements.
Microsoft’s recent “Wave 2 Spring release” focused on AI agents (Researcher, Analyst, Skills), an Agent Store, and enhanced IT governance tools via the Copilot Control System, with rollouts expected from late May 2025.
Other features unveiled around that time included Copilot Memory for personalization, Copilot Actions for tasks like booking travel, Deep Research capabilities, collaborative Copilot Pages, AI-generated Podcasts, and integrated Shopping tools.
Updates also extended to the Windows experience, with Copilot Vision enabling screen analysis across any app and File Search accessing local documents, both previewed for Windows Insiders in April. These additions came after Microsoft made a notable strategic shift earlier in 2025.
In February 2025, it removed usage limits on the advanced “Think Deeper” reasoning feature (first powered by OpenAI’s o1, then upgraded to the more capable o3-mini-high model in March) and its Voice features, making them free for all users—a direct contrast to OpenAI’s paid tiers for similar model access. Yet, demonstrating a multifaceted approach, Microsoft introduced a paywall in March for AI features in basic Windows apps like Notepad and Paint, tying them to Microsoft 365 subscriptions and a monthly credit system.
Pressure Mounts Amid External Scrutiny
Even high-profile events faced external headwinds. Suleyman’s presentation at Microsoft’s 50th Anniversary in April 2025 was interrupted by employee protestors condemning Microsoft’s alleged AI dealings with the Israeli military.
Suleyman somehow acknowledged them, stating “Thank you for your protest, I hear you” – although the honesty of these statements can be doubted – before they were escorted out, highlighting the broader societal scrutiny accompanying AI development. The combination of ambitious feature rollouts, strategic pricing experiments, internal development challenges, complex partnerships, and lagging consumer adoption creates a high-pressure environment for Suleyman.
With CEO Satya Nadella known for shifting focus away from underperforming initiatives, and CFO Amy Hood monitoring results, the need for Microsoft’s significant consumer AI investment to translate into tangible user growth appears increasingly urgent.
Last Updated on May 8, 2025 10:10 am CEST
