Koray Kavukcuoglu

Koray Kavukcuoglu is a Turkish computer scientist and machine-learning researcher, born in Turkey in 1980. He is the Chief Technology Officer of Google DeepMind and, since June 11, 2025, the first Chief AI Architect at Google, a senior vice president role reporting directly to Sundar Pichai. As of May 2026, he holds both titles in parallel from Mountain View, California, after relocating from London in 2025, and is a Fellow of the Royal Academy of Engineering with co-authorship credit on the DQN, AlphaGo, and WaveNet papers from his early DeepMind period.

At a glance

• Education: BS in Aerospace Engineering, Middle East Technical University (METU), Ankara, 1999; MS in Aerospace Engineering, METU, 2003; MS in Computer Science, New York University, 2005; PhD in Computer Science, NYU Department of Computer Science, January 2011, supervised by Yann LeCun, thesis Learning Feature Hierarchies for Object Recognition.
• Current roles: CTO, Google DeepMind, since 2018; Chief AI Architect, Google, since June 11, 2025.
• Key contributions: co-authored Human-level control through deep reinforcement learning (Nature, February 2015), the DQN paper that established deep reinforcement learning; senior co-author on Mastering the game of Go with deep neural networks and tree search (Nature, 2016); senior co-author on WaveNet: A Generative Model for Raw Audio (2016); founded DeepMind's deep-learning research team.
• Awards: Fellow of the Royal Academy of Engineering (FREng), elected September 20, 2022.
• X / Twitter: @koraykv
• LinkedIn: Koray Kavukcuoglu
• Google Scholar: Koray Kavukcuoglu
• Google blog page: blog.google/authors/koray-kavukcuoglu

Origins

Kavukcuoglu was born in Turkey in 1980. He studied aerospace engineering at the Middle East Technical University (METU) in Ankara, taking a BS in 1999 and an MS in 2003. During the four-year METU graduate-study period he held a research-and-development engineering role at Roketsan, the Turkish guided-missile and rocket-systems manufacturer.

In 2003 he moved to New York University, taking an MS in computer science in 2005 and continuing toward the PhD under Yann LeCun at the Courant Institute. The dissertation, Learning Feature Hierarchies for Object Recognition, was filed in January 2011 and worked through unsupervised feature learning, predictive sparse decomposition, and convolutional sparse coding, with applications to object recognition, handwritten-digit classification, and pedestrian detection. During the doctoral period Kavukcuoglu interned at Siemens (2007), NEC Laboratories America (2008 and 2009), and Google (summer 2010), and was a co-developer with Ronan Collobert and Clément Farabet of the Torch5 machine-learning library.

Career

After the PhD, Kavukcuoglu spent a short period at NEC Laboratories America in Princeton, New Jersey, then joined DeepMind Technologies in London in November 2012, two years after the lab's September 2010 founding by Demis Hassabis, Shane Legg, and Mustafa Suleyman, and fourteen months before Google's January 2014 acquisition. He founded the lab's deep-learning research team during this period.

The early DeepMind period produced the deep Q-network research that became the lab's first widely-cited public artifact. Kavukcuoglu was a co-author on the December 2013 Playing Atari with Deep Reinforcement Learning preprint, lead-authored by Volodymyr Mnih, and on the February 2015 Nature paper Human-level control through deep reinforcement learning, which extended the system to fifty-seven Atari 2600 games. Through the mid-2010s, he held a senior author position on the January 2016 Nature paper Mastering the game of Go with deep neural networks and tree search, lead-authored by David Silver and Aja Huang; and the September 2016 WaveNet: A Generative Model for Raw Audio paper, lead-authored by Aäron van den Oord, which shipped in production as the Google Assistant voice.

Kavukcuoglu held the title of Vice President of Research and Technology at Google DeepMind through the late 2010s, was elevated to Chief Technology Officer in 2018, and held the CTO title through the April 2023 merger of DeepMind with Google Brain, the launch of the Gemini family from December 2023 onward, and the 2024 Nobel Prize awarded to Hassabis and John Jumper for AlphaFold.

On June 11, 2025, Sundar Pichai announced in an internal memo that Kavukcuoglu would become the company's first Chief AI Architect, a newly created senior vice president role reporting directly to Pichai, with a remit to accelerate the integration of Google DeepMind's models into Google's product surface. He retained the DeepMind CTO title in parallel and relocated from London to Mountain View, California, and in November 2025 co-signed the Gemini 3 launch announcement alongside Pichai and Hassabis.

Affiliations

• Roketsan: R&D Engineer, 1999 to 2003.
• NEC Laboratories America: Research staff member (machine-learning department), 2010 to 2012.
• Google DeepMind: Research scientist (founded the deep-learning team) from November 2012; Vice President of Research and Technology through approximately 2018; Chief Technology Officer, 2018 to present.
• Google: Chief AI Architect (first), June 11, 2025 to present.

Notable contributions

Kavukcuoglu's published record concentrates on deep learning, deep reinforcement learning, and generative models. His Google Scholar profile lists more than 270,000 citations and an h-index above 85.

• Human-level control through deep reinforcement learning (Nature, February 26, 2015). Co-author on the deep Q-network paper that established deep reinforcement learning as a viable research program. The system reached human-level performance on forty-nine Atari 2600 games from raw-pixel input and laid the groundwork for the AlphaGo and AlphaZero programs.
• Mastering the game of Go with deep neural networks and tree search (Nature, January 28, 2016). Senior author position on the AlphaGo paper, which produced the first program to defeat a professional Go player on a full nineteen-by-nineteen board. The system's March 2016 four-to-one win over Lee Sedol in Seoul is a recurring reference point in deep-learning histories.
• WaveNet: A Generative Model for Raw Audio (September 2016). Senior author on the autoregressive raw-audio generative model; the architecture shipped in production as the Google Assistant voice.
• DeepMind deep-learning research team. Founded in the early 2010s. The subsequent VP-of-Research and CTO roles cover the broader portfolio that produced the AlphaGo, AlphaZero, AlphaFold, and Gemini lines.
• Torch5 (late 2000s and early 2010s). Co-developer with Ronan Collobert and Clément Farabet of the Lua-based machine-learning framework that supported large parts of deep-learning research before the field consolidated around Python.
• Gemini family (December 2023 onward). Senior research and technology leadership across the Gemini line through Gemini 1, 1.5, 2, 2.5, 3, and 3.1 Pro. He co-signed the Gemini 3 launch announcement in November 2025 alongside Pichai and Hassabis.
• Public-speaking record. Recurring appearances at Google I/O 2024 and Google I/O 2025; the November 2025 long-form interview This Is How We Are Going to Build AGI with Logan Kilpatrick on the Google AI: Release Notes podcast. Podcast and conference appearances are limited relative to peer figures at frontier laboratories.
• Royal Academy of Engineering Fellowship (September 20, 2022). Elected as a Fellow (FREng) for leadership in deep learning, unsupervised learning, generative models, and deep reinforcement learning.

Investments and boards

• Google DeepMind (AI): Chief Technology Officer, 2018 to present. Subsidiary of Alphabet; merged with Google Brain in April 2023.
• Google (AI / Software): Chief AI Architect (first), June 11, 2025 to present. Senior Vice President reporting directly to Sundar Pichai.

No public personal angel-investor activity on record in AI, semiconductors, datacenters, software, or energy as of May 2026.

Network

Kavukcuoglu's longest-running professional relationship is with Yann LeCun, his NYU PhD advisor across 2005 to 2011, the 2018 Turing Award co-recipient now running AMI in Paris. Within DeepMind, his closest collaborator is Demis Hassabis, the chief executive he has worked alongside since 2012, with cofounders Shane Legg (continuously at the lab) and Mustafa Suleyman (departed in 2019).

His DeepMind colleagues across the early DQN, AlphaGo, and WaveNet research arc include Volodymyr Mnih, the lead author on the deep Q-network papers; David Silver, the AlphaGo and AlphaZero principal investigator who departed in late 2025 to found Ineffable Intelligence; Aäron van den Oord, the lead author on WaveNet; and John Jumper, the AlphaFold research lead and 2024 Nobel laureate. Among DeepMind senior leadership his peers as of May 2026 include Jumper, Legg, and Lila Ibrahim. At Google, his principal counterparty since June 2025 is Sundar Pichai, alongside Hassabis on the DeepMind side and Jeff Dean, Google's chief scientist.

Position in the field

As of May 2026, Kavukcuoglu is one of a small group of senior AI research leaders whose career has run continuously inside a single frontier laboratory since the early 2010s, alongside Legg and Hassabis at DeepMind and Dean at the broader Alphabet research organization. The fourteen-year DeepMind tenure, from November 2012 through the Google acquisition (2014), the AlphaGo Lee Sedol match (2016), the Brain merger (2023), and the Gemini cycles, is a defining structural fact of his record.

The June 2025 elevation to Chief AI Architect is the principal recent event. The senior vice president role formalizes a research-to-product authority structure that had not previously existed at the parent-company level. Industry coverage characterized the dual-title structure as a deliberate organizational design intended to accelerate the rollout of DeepMind's models across Google Search, Workspace, Pixel, Android, and the Gemini app, while preserving research-organization continuity at DeepMind.

His scientific footprint sits in the production-research line. The DQN, AlphaGo, and WaveNet authorship credits across 2013 to 2016 are his most-cited published works; the subsequent decade has been concentrated in research-leadership and organizational roles. The 2022 Royal Academy of Engineering Fellowship is his principal external scientific recognition, and the LeCun-advised PhD places him in the same research-genealogy line as Yoshua Bengio and other senior figures in the modern deep-learning lineage.

Outlook

Open questions over the next 6 to 18 months:

• Chief AI Architect remit. Whether the role translates into observable acceleration of Gemini integration into Google Search, Workspace, Android, and the Gemini app.
• Apple-Siri rollout. The Gemini-into-Siri partnership announced in 2026 is among DeepMind's largest distribution events and the kind of cross-organizational integration the Chief AI Architect mandate was created to coordinate.
• Gemini 4 cycle. Whether the next-generation line narrows the gap to OpenAI and Anthropic on coding benchmarks.
• Public-speaking cadence. Whether the elevated visibility of the Chief AI Architect role produces more frequent long-form interviews, podcast appearances, and conference keynotes.
• Senior-research departures. Whether the late-2025 to early-2026 departures of David Silver to Ineffable Intelligence and Tim Rocktäschel to Recursive Superintelligence are followed by further senior-bench moves.
• Continuity at Alphabet. Whether Kavukcuoglu remains in the dual role through the next Gemini cycle, given the broader 2025 to 2026 pattern of senior researchers leaving frontier labs to found independent ventures.

Sources

• Koray Kavukcuoglu LinkedIn profile. Public profile listing the CTO and Chief AI Architect titles, education, and employment history.
• Koray Kavukcuoglu, Google blog author page. Google's official byline page identifying him as CTO of Google DeepMind and Chief AI Architect at Google.
• Koray Kavukcuoglu, Google Scholar profile. Citation profile listing publications, citation count, and h-index.
• Learning Feature Hierarchies for Object Recognition. The January 2011 NYU PhD dissertation, supervised by Yann LeCun.
• Human-level control through deep reinforcement learning. The February 2015 Nature deep Q-network paper.
• Mastering the game of Go with deep neural networks and tree search. The January 2016 Nature AlphaGo paper, with Kavukcuoglu in a senior author position.
• WaveNet: A Generative Model for Raw Audio. The September 2016 raw-audio generative-model paper, with Kavukcuoglu as senior author.
• Koray Kavukcuoglu FREng, Royal Academy of Engineering. The 2022 Fellowship citation page.
• Exclusive: Google names new chief AI architect to advance developments. Semafor's June 11, 2025 report on the Chief AI Architect appointment.
• Google taps DeepMind's Kavukcuoglu for new chief AI architect role. CNBC's parallel reporting on the appointment, with biographical detail.
• Introducing Gemini 3. The November 2025 Gemini 3 launch announcement, co-signed by Pichai, Hassabis, and Kavukcuoglu.
• Koray Kavukcuoglu: This Is How We Are Going to Build AGI. The November 24, 2025 long-form interview on the Google AI: Release Notes podcast with Logan Kilpatrick on Gemini 3 and the path to AGI.
• Google I/O 2025: Koray Kavukcuoglu. Google I/O 2025 speakers page listing the dual title.
• Photo: Koray Kavukcuoglu headshot from the Google I/O 2025 speakers page, Google press use.