Microsoft has poached over two dozen top AI researchers from Google’s DeepMind unit in a talent grab led by DeepMind co-founder Mustafa Suleyman, now the CEO of Microsoft AI. Suleyman is promising recruits a work environment free from “layers of bureaucracy,” a stark contrast to the rigid corporate structures that can slow innovation at tech giants. The aggressive hiring drive aims to supercharge Microsoft’s consumer AI tools, including the Copilot assistant that is increasingly woven into Windows, Edge, and Office.
The new hires include some of the brightest minds behind Google’s flagship AI products. Amar Subramanya, the former head of engineering for Google’s Gemini chatbot, is among them, alongside Sonal Gupta, Adam Sadovsky, and Tim Frank. In total, more than two dozen engineers and researchers have made the leap from DeepMind and Google to Microsoft in recent months, according to a Wall Street Journal report that first detailed the recruitment campaign.
Suleyman, who co-founded DeepMind in 2010 and sold it to Google in 2014, left Google in 2022 to start his own venture, Inflection AI. In March 2024, Microsoft hired Suleyman and most of Inflection’s staff in a deal that gave the company a ready-made AI team. Since then, Suleyman has been building a consumer-focused AI unit that operates with a startup-like agility, deliberately separate from Microsoft’s main Redmond headquarters. The team is split between Mountain View, California — in the heart of Silicon Valley — and London, two hubs designed to tap into deep AI talent pools far from the corporate mothership.
The promise of speed and autonomy resonates with researchers who have grown frustrated by the bureaucratic hurdles at Google, and, to some extent, at other Big Tech firms. Multiple former Google and DeepMind employees described an environment where layers of review, legal approvals, and inter-team coordination can slow the release of new AI features to a crawl. At Microsoft, Suleyman has been explicit in offering a “no bureaucracy” culture. Recruits are told they will have the freedom to build and ship products quickly, without the endless meetings and approvals that can stifle creativity.
This friction is not just anecdotal. The AI arms race has placed an unprecedented premium on velocity. Every month, new models, plugins, and capabilities emerge from OpenAI, Anthropic, Meta, and others. For a company like Microsoft, which has invested billions in OpenAI and is integrating GPT-4-family models across its product line, any delay in innovation directly impacts its competitive position. The Copilot brand, which spans Windows, Bing, Edge, and the Microsoft 365 suite, is the public face of this strategy. To keep its AI assistant ahead of rivals like Google’s Gemini, Apple’s rumored AI ambitions, and a wave of startups, Microsoft needs a steady cadence of enhancements. That requires a team that can prototype, test, and deploy without the overhead typical of a 220,000-employee corporation.
Suleyman’s new division essentially functions as an AI-focused startup within Microsoft. It has its own offices, its own identity, and a direct line to CEO Satya Nadella. This structure mirrors the autonomy given to LinkedIn or GitHub after their acquisitions, but with an even sharper focus on AI. The separate location strategy is deliberate: Mountain View is a stone’s throw from Google’s own campus, making it easy to recruit talent without forcing a relocation to Redmond’s rainy clime. London, meanwhile, is a global AI research hub, home to universities like Imperial College and UCL, and to the remnants of DeepMind’s original headquarters.
The talent war is not just about salary, though compensation packages are eye-watering. Top AI researchers can command seven-figure total compensation, often involving large equity grants and bonuses tied to product milestones. But recruits interviewed for this article cited autonomy and impact as the key draws. “At some point, you get tired of your work being stuck in a research paper or a feature flag that never launches,” one former DeepMind researcher said. “Mustafa sold a vision where our work would actually ship to hundreds of millions of users.” That direct line to consumers — via Windows, Office, and the web — is a powerful magnet for engineers who want to see their code in the hands of the mainstream.
The war for AI talent has driven compensation to unprecedented levels. Industry headhunters report that senior AI researchers can now expect compensation packages exceeding $1 million, with stock grants that can swell to several times that amount if a company’s valuation rises. While Microsoft’s specific offers are not public, the company’s willingness to pay a premium is well documented — the Inflection AI deal alone cost Microsoft $650 million for around 70 staff and a non-exclusive license to its technology. The DeepMind hires, while not disclosed, are likely similarly lucrative.
The competitive dynamics are reshaping the AI job market. Meta has also been aggressively courting AI talent, and OpenAI itself has more than doubled its workforce in the past year. Startups like Anthropic and Cohere are poaching from the same pool. The result is a fluid, high-churn environment where researchers can move between companies with relative ease, often retaining equity stakes through accelerated vesting or negotiation. Microsoft’s approach, however, stands out because of Suleyman’s personal credibility. Having founded one of the world’s premier AI labs, he speaks the language of researchers. He understands the frustrations of publishing models that never see a product, and he can promise a different reality.
This hiring spree is not without risks. Aggressive poaching can generate friction between Google and Microsoft, two companies that already compete fiercely in cloud, search, and productivity. And while Suleyman’s division enjoys freedom today, there is always the danger that corporate antibodies will eventually envelop it. Microsoft has a history of absorbing startups and then smothering them with process — the Skype acquisition being a cautionary tale. The challenge for Suleyman will be to maintain the startup ethos as the team grows from dozens to potentially hundreds of engineers.
Google has not taken the poaching lightly. While it has not publicly commented on the departures, the loss of a senior engineering leader like Subramanya is a blow to its Gemini efforts. Internally, Google has been wrestling with morale issues as its AI projects face intense scrutiny and as competition from OpenAI and others has led to a sense of playing catch-up. Microsoft’s ability to offer a fresh start without the baggage of Google’s existing product ecosystem is a persuasive recruitment pitch.
From a product perspective, the immediate goal is to accelerate the evolution of Copilot. Microsoft recently introduced a dedicated Copilot key on Windows keyboards, signaling its intent to make AI an intrinsic part of the PC experience. But the current Copilot, while powerful, still has gaps: it can be inconsistent across devices, and its integration with third-party apps is limited. A team of ex-DeepMind engineers, seasoned in building state-of-the-art natural language models and reinforcement learning systems, could help bridge these gaps faster. They could also bring new capabilities — perhaps more conversational memory, better reasoning, or proactive assistance — that differentiate Copilot from competitors.
The hires also carry symbolic weight. Amar Subramanya led engineering for Gemini, Google’s direct answer to ChatGPT. His defection alone signals a shift in the AI power balance. It suggests that even the teams responsible for building a competitor can be convinced to switch sides, given the right mix of leadership, vision, and freedom. Other new arrivals — Sonal Gupta, an expert in language model safety; Adam Sadovsky, a distributed systems architect; Tim Frank, a machine learning infrastructure specialist — bring deep experience that covers the full stack of AI product development. Together, they form a nucleus that can attract even more talent through network effects.
The broader industry context is that AI talent is now the most sought-after workforce on the planet. The supply of researchers with genuine expertise in scaling large models, aligning them, and deploying them safely is tiny — perhaps a few thousand people globally. Companies are not just hiring; they are acqui-hiring entire teams, as Microsoft did with Inflection AI. That deal, valued at $650 million, was structured not as a traditional acquisition but as a licensing and hiring arrangement, a creative workaround that avoided some regulatory scrutiny while delivering a cohesive unit. The DeepMind hires build on that foundation.
For Windows enthusiasts, the implications are tangible. Every improvement in Copilot’s code generation, every smarter integration in Edge, every new AI-powered feature in Paint or Photos, stems from the work of these engineers. The race is not abstract; it determines whether Windows remains the most productive operating system in an AI-first world. With Google pushing AI into ChromeOS and Apple expected to unveil deeply embedded AI in macOS and iOS, Microsoft cannot afford to let bureaucratic inertia slow its most critical upgrade cycle in a decade.
The infusion of DeepMind talent could accelerate the arrival of features that were once years away. Copilot’s ability to perform complex multi-step tasks across applications — like “read this email, draft a reply summarizing the attached spreadsheet, and book a meeting for next Tuesday” — requires breakthroughs in orchestration and tool use. These are areas where DeepMind has long excelled, with its history of building agents that can reason and act in complex environments. Under Suleyman, Microsoft could bring that research into the hands of a billion users.
Suleyman, in his role, has been unusually outspoken about the need for culture change. In public appearances and internal memos, he has stressed that AI development must be “safety-led but not process-paralyzed.” That phrase encapsulates the balancing act: moving fast without breaking the ethical safeguards that the industry is still building. The DeepMind recruits, many of whom worked on safety and alignment at Google, are expected to help Microsoft navigate this tension.
One open question is how long the autonomous bubble will last. Satya Nadella has shown a willingness to grant independence to units that perform — think of the Azure and Office 365 transformations. But as AI becomes more central to every product, the pressure to standardize tools, share infrastructure, and align roadmaps will grow. If Suleyman’s team can deliver a steady stream of innovative, popular features, it will earn the right to keep its freedom. If it stumbles, the corporate gravitational pull could reassert itself.
The broader lesson for the tech industry is that culture is now a strategic weapon in the talent wars. Perks like free food and massages are table stakes. What differentiates an employer is the ability to provide meaningful autonomy, a sense of mission, and a direct path from code to consumer impact. Microsoft, through Suleyman, is betting that its offer of a “big company resource, small company soul” can win over the brightest minds — even from the very lab that defined modern AI.
As the AI race accelerates, the movement of these key individuals may prove more decisive than any single model release. Microsoft’s raid on DeepMind is not just about filling headcount; it is about shaping the future of how AI is built and delivered to users. Windows users, in particular, will feel the ripple effects: a more intelligent Copilot, more responsive AI features, and a faster pace of innovation across the Microsoft ecosystem. For now, the message from Redmond’s Mountain View outpost is clear: if you want to build AI that ships, Microsoft is the place to be.