Alphabet's AI empire: Gemini models, DeepMind research, custom TPUs, and a hyperscale cloud.
Google, under parent Alphabet, is the only company that is simultaneously a frontier lab (Google DeepMind), a hyperscaler (Google Cloud), and a chip designer (TPUs). Its Gemini models power consumer products reaching billions of users, while Vertex AI distributes them to enterprises. Google invented the transformer architecture that underpins modern AI, and its vertically integrated TPU stack makes it the strongest structural rival to the NVIDIA-centric ecosystem.
Founded in 1998 by Larry Page and Sergey Brin, Google built the modern internet's dominant search, advertising, video, and mobile platforms before becoming an AI-first company. Its researchers published the 2017 "Attention Is All You Need" paper that introduced the transformer, and its 2014 acquisition of DeepMind — merged with Google Brain in 2023 to form Google DeepMind under Demis Hassabis — gave it one of the deepest research benches in the field.
The Gemini family is Google's frontier line, spanning ultra-efficient Flash tiers to flagship reasoning models, alongside Veo video, Imagen image, and Gemma open-weight models. Distribution is Google's structural advantage: Gemini ships inside Search (AI Overviews and AI Mode), Android, Workspace, and Chrome, while Google Cloud's Vertex AI serves enterprises.
Google designs its own Tensor Processing Units, now in their seventh generation, and is the only frontier lab training flagship models entirely on in-house silicon. That stack has become an export: Anthropic contracted for up to a million TPUs, and Broadcom co-designs the chips. Alphabet is also a major Anthropic investor, and its 2024 licensing deal with Character.AI brought transformer co-inventor Noam Shazeer back to lead Gemini work.
Google sells model access we track in the pricing index. API pricing & models →
The Compute · Convergence
How Google’s AI empire converged
Nine years of distributed invention, hesitation and panic — resolving, by 2026, into a single operating line running through Gemini and Google DeepMind to one person.
The timeline, in full
21 eventsEvery event in the sequence, in order, with the source it comes from.
2017· RESEARCH
Transformer
Google researchers publish “Attention Is All You Need,” introducing the architecture that underlies nearly all modern AI models.
SOURCE — Vaswani et al., NeurIPS 20172018· PEOPLE / POWER
Transformer authors begin leaving
Several of the paper’s authors depart Google over the following years to found or join other AI labs.
SOURCE — Company announcements2020· RESEARCH
AlphaFold
DeepMind’s AlphaFold solves the 50-year-old protein folding problem, a landmark for AI in science.
SOURCE — DeepMind, Nature 20202021· PRODUCT
LaMDA unveiled
Google introduces its conversational language model LaMDA, a precursor to its later chatbot products.
SOURCE — Google I/O 20212022· RESEARCH
PaLM & Chinchilla
Google Research and DeepMind separately publish large-scale language models, previewing capabilities behind Gemini.
SOURCE — Google Research, DeepMind2022· PEOPLE / POWER
Page & Brin re-engage
Alphabet’s founders return to a more active role as generative AI accelerates inside the company.
SOURCE — Reports, 2022Nov 2022· PRODUCT
ChatGPT shock
OpenAI’s ChatGPT launches to the public, catching Google off guard and triggering an internal “code red.”
SOURCE — OpenAIFeb 2023· PRODUCT
Bard’s error
Google’s Bard chatbot makes a factual error in its first public demo, wiping billions off Alphabet’s market value.
SOURCE — Reuters, Feb 2023Apr 2023· PEOPLE / POWER
Brain and DeepMind merge
Pichai merges Google Brain and DeepMind into a single unit, Google DeepMind. Demis Hassabis becomes CEO; Jeff Dean becomes Chief Scientist.
SOURCE — Google announcement, Apr 2023Dec 2023· PRODUCT
Gemini launches
Google ships Gemini, its first natively multimodal model family, positioned as its answer to GPT-4.
SOURCE — Google, Dec 2023Apr 2024· PEOPLE / POWER
Research teams move in
More Google Research model-building groups are folded into Google DeepMind.
SOURCE — Google, Apr 2024Oct 2024· PRODUCT
Gemini app team merges
The consumer Gemini app team moves into Google DeepMind, consolidating research and product under one roof.
SOURCE — Google, Oct 2024Oct 2024· RESEARCH
Nobel Prize
Demis Hassabis and John Jumper win the Nobel Prize in Chemistry for AlphaFold.
SOURCE — Nobel Foundation, 20242023–25· PEOPLE / POWER
Shazeer’s exit, return, exit
Noam Shazeer leaves to found Character.AI, later returns to Google, then departs again.
SOURCE — Reports, 2023–2025Jun 2025· PEOPLE / POWER
Koray named Chief AI Architect
Koray Kavukcuoglu becomes SVP and Chief AI Architect, reporting directly to Sundar Pichai, tasked with speeding model-to-product integration.
SOURCE — Google, Jun 20252025· PEOPLE / POWER
Jumper leaves
John Jumper, co-lead of AlphaFold, departs Google DeepMind.
SOURCE — Reports, 20252026· PRODUCT
Gemini 3.5 Pro delayed
The launch of Gemini 3.5 Pro slips, drawing scrutiny of Google’s release cadence against rivals.
SOURCE — Reports, 20262026· PRODUCT
Nano Banana rollback
Google walks back a Nano Banana feature rollout on Earth, a rare public reversal.
SOURCE — Reports, 2026Aug 2026· PEOPLE / POWER
Hassabis steps back
Demis Hassabis leaves day-to-day operational management, becoming Chair of Google DeepMind and Alphabet Chief Scientist.
SOURCE — Google, Aug 2026Aug 2026· PEOPLE / POWER
Koray assumes operational control
Koray Kavukcuoglu becomes SVP and operational head of Google DeepMind while remaining Chief AI Architect, reporting directly to Sundar Pichai.
SOURCE — Google, Aug 2026Aug 2026· PEOPLE / POWER
Dean, Ghemawat, Vinyals, Le depart
Four of Google’s most senior AI leaders leave to launch Discovery Loop.
SOURCE — Reports, Aug 2026
Positions on the chart are illustrative; the dates, roles and events are the record. Reporting-sourced items are labelled as such.
Key people
8Related companies
4Part of the Tokenando AI Landscape · Explore the landscape → · All companies →