Google Cloud
Last reviewed
Sources
28 citations
Review status
Source-backed
Revision
v2 ยท 2,278 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Last reviewed
Sources
28 citations
Review status
Source-backed
Revision
v2 ยท 2,278 words
Add missing citations, update stale details, or suggest a clearer explanation.
Google Cloud is the cloud computing division of Google and its parent Alphabet, and one of the world's largest suppliers of AI computing, selling custom tensor processing units (TPUs), Nvidia GPU capacity, and access to the Gemini model family through its Vertex AI platform. Led since early 2019 by chief executive Thomas Kurian, it spans the Google Cloud Platform (GCP) infrastructure business, the Google Workspace productivity suite, and a growing security portfolio, and has grown from a distant third in cloud infrastructure into a core engine of Alphabet's AI strategy. Its AI customers now include rival laboratories Anthropic, OpenAI, and Meta. In the first quarter of 2026, Google Cloud reported revenue of $20.0 billion, up 63 percent year over year, with operating income of $6.6 billion and a contracted backlog of $462 billion.[1][2]
Google Cloud sells three broad product lines: GCP, which provides compute, storage, networking, databases, analytics, and machine learning infrastructure; Google Workspace, the subscription suite built around Gmail, Docs, Drive, and Meet; and security products built on Mandiant and, since March 2026, Wiz.[3] Kurian reports to Alphabet and Google CEO Sundar Pichai, and Alphabet discloses Google Cloud as a standalone operating segment.
The division lost money for most of its history, posting operating losses as recently as 2022 before swinging to its first quarterly operating profit in early 2023. Profitability has since scaled rapidly: operating margin reached 32.9 percent in the first quarter of 2026, up from a year-earlier margin in the high teens.[2] Pichai said Google Cloud exited 2025 "at an annual run rate of over $70 billion," driven by demand for AI products.[4]
| Period | Revenue | YoY growth | Operating income |
|---|---|---|---|
| FY 2024 | $43.2 billion | 31% | $6.1 billion |
| FY 2025 | $58.7 billion | 36% | $13.9 billion |
| Q4 2025 | $17.7 billion | 48% | $5.3 billion |
| Q1 2026 | $20.0 billion | 63% | $6.6 billion |
Sources: Alphabet earnings releases.[1][2][4]
The order backlog, mostly multi-year cloud and AI capacity commitments, nearly doubled in a single quarter, from $240 billion at the end of 2025 to $462 billion at the end of March 2026; CFO Anat Ashkenazi said roughly half is expected to convert to revenue within 24 months.[2] Alphabet attributed the jump to enterprise AI demand, fast-growing Gemini Enterprise adoption, and a new line of business: sales of TPU hardware itself.[2] The backlog, larger than Google Cloud's entire annual revenue several times over, implies years of contracted growth if Alphabet can build data centers fast enough to fulfill it.
Sustaining the buildout is expensive. Alphabet's capital expenditures rose from $52.5 billion in 2024 to $91.4 billion in 2025, and 2026 guidance was set at $175 billion to $185 billion in February 2026, then raised to $180 billion to $190 billion in April, overwhelmingly for AI infrastructure serving Google Cloud and DeepMind.[2][4]
Google entered cloud computing in April 2008 with App Engine, a platform for running web applications on Google's infrastructure, and added the Compute Engine virtual machine service in 2012. In November 2015 Google acquired Diane Greene's startup Bebop and put Greene, a co-founder of VMware, in charge of a unified cloud organization; the "Google Cloud" brand followed in 2016.[5] Google revealed its first TPU in 2016 and began renting TPUs to cloud customers in 2018, an early differentiator against Amazon Web Services (AWS) and Microsoft Azure.
Kurian, who spent 22 years at Oracle and ran its product development, was announced as Greene's successor in November 2018 and took over in early 2019, when Google Cloud's annual revenue was $8.9 billion.[5] He pushed aggressively into regulated industries, expanded the salesforce, and acquired security firm Mandiant for about $5.4 billion in 2022. The suite formerly known as G Suite was rebranded Google Workspace in October 2020. In March 2026 Google closed its $32 billion acquisition of cloud security company Wiz, the largest acquisition in Google's history, and folded it into Google Cloud.[3] The generative AI boom that followed ChatGPT transformed the division's trajectory: revenue growth accelerated every year from 2023 through early 2026 as AI infrastructure and platform deals stacked up.[1][4]
Google Cloud's pitch to AI customers rests on what it calls the AI Hypercomputer, an integrated architecture introduced in December 2023 that combines TPUs, Nvidia GPUs, optical circuit switching, high-bandwidth networking, and co-designed software such as JAX and Pathways. Alongside large Nvidia GPU fleets, the company's main differentiator is its TPU line, custom accelerators it has designed since 2015 with Broadcom as a longtime co-development partner.
Recent generations:
| TPU | Status | Key specifications |
|---|---|---|
| Trillium (v6e, 6th gen) | GA December 2024 | 4.7x peak compute per chip vs v5e; 32 GB HBM; used to train Gemini 2.0[6] |
| Ironwood (TPU7x, 7th gen) | Unveiled April 2025; GA November 2025 | 4,614 teraflops FP8 per chip; 192 GB HBM3E at 7.4 TB/s; pods of 9,216 chips delivering 42.5 FP8 exaflops and about 1.77 PB of shared memory; roughly 2x performance per watt vs Trillium[7][8] |
| TPU 8t / TPU 8i (8th gen) | Unveiled April 2026 | 8t (training): up to 9,600 chips and 2 PB shared memory per superpod, claimed 3x Ironwood throughput; 8i (inference): 1,152-chip pods, claimed 80% better performance per dollar[9] |
Google billed Ironwood as its first TPU designed "for the age of inference," and it went on general availability in November 2025 alongside new virtual machines based on Axion, Google's Arm-based server CPU.[7][8] A single Ironwood pod's 42.5 FP8 exaflops far exceeds the 0.36 exaflops of Nvidia's GB300 NVL72 system, the comparison Google used to argue its silicon is competitive at the rack scale.[8]
Two strategic shifts stand out. First, Google began selling TPU hardware for deployment outside its own data centers; Alphabet said TPU hardware sales contributed to the record Q1 2026 backlog, with most of that revenue expected to be recognized in 2027.[2] Second, Broadcom's role expanded: an April 2026 three-way agreement with Anthropic covers next-generation TPU capacity and networking for AI data center racks later this decade.[10]
Vertex AI, launched on May 18, 2021, unified Google Cloud's machine learning tooling into one managed platform; generative AI support became generally available in June 2023 with access to PaLM 2, and an agent-building layer followed in April 2024.[11] Its Model Garden catalog grew to more than 200 models, mixing Google's own Gemini, Imagen, and Veo with open-weight families such as Llama and Gemma and third-party models from Anthropic (Claude) and Mistral AI, a notable contrast with Azure and AWS, which do not carry each other's flagship models.[12] By February 2026, Pichai said Google's first-party models were processing more than 10 billion tokens per minute through direct API use by cloud customers.[4]
The platform layer has been reorganized around agents. In January 2025 Google folded Gemini AI features into Workspace Business and Enterprise plans, discontinuing the separate $20 to $30 per seat add-ons while raising base subscription prices.[13] On October 9, 2025 it launched Gemini Enterprise, described by Kurian as "the new front door for AI in the workplace," absorbing the earlier Agentspace product; paid monthly active users grew 40 percent quarter over quarter in early 2026.[2][14][15] At Cloud Next on April 22, 2026, Google announced the Gemini Enterprise Agent Platform as the successor to Vertex AI, positioning it as a full stack for building and governing AI agents. Kurian framed the moment bluntly: "The era of the pilot is over. The era of the agent is here."[9][16]
Google Cloud has converted the AI capacity crunch into a string of landmark contracts, including from direct competitors of Google DeepMind.
| Customer | Date | Deal |
|---|---|---|
| Anthropic | October 23, 2025 | Expansion giving access to up to 1 million TPUs and well over 1 GW of capacity in 2026; reported value in the tens of billions of dollars[16][17] |
| Anthropic (with Broadcom) | April 6, 2026 | Multiple gigawatts (reported around 3.5 GW) of next-generation TPU capacity coming online from 2027[10][18] |
| OpenAI | Confirmed July 16, 2025 | Google Cloud capacity for ChatGPT and the OpenAI API in the US, Japan, the Netherlands, Norway, and the UK[19][20] |
| Meta | Reported August 22, 2025 | Cloud deal worth more than $10 billion over six years for servers, storage, and networking[21] |
| Meta (TPUs) | Reported February 2026 | Multibillion-dollar multiyear agreement to rent TPU capacity, with separate talks to buy TPUs outright for Meta's own data centers from 2027[22][23] |
| Safe Superintelligence | April 2025 | TPU access, with Google Cloud as the primary compute provider for Ilya Sutskever's lab[24] |
| Midjourney | March 2023 | Training and serving its image models on Google Cloud[25] |
| Apple | Disclosed July 2024 | Apple Intelligence foundation models trained on clusters of Google TPUs, including 8,192 TPU v4 chips for the server model[26] |
The Anthropic relationship is the deepest: Google is one of Anthropic's largest outside investors, and Claude runs on a deliberately multi-platform stack of Google TPUs, Amazon Trainium, and Nvidia GPUs.[16][17] Anthropic CFO Krishna Rao said the company chose to deepen its TPU commitment for "price-performance and efficiency," adding that "Anthropic and Google have a longstanding partnership and this latest expansion will help us continue to grow the compute we need to define the frontier of AI."[16]
The OpenAI agreement, first reported by Reuters in June 2025 after being finalized in May, was widely described as unprecedented given that ChatGPT competes directly with Google Search; it became possible after OpenAI's cloud exclusivity with Microsoft lapsed in January 2025, and Google sits alongside Microsoft, Oracle, and CoreWeave on OpenAI's supplier list.[19][20]
Google Cloud remains the third-largest cloud infrastructure provider but is the fastest growing of the big three. Synergy Research Group put worldwide cloud infrastructure spending at $129 billion in the first quarter of 2026, up 35 percent year over year, with AWS holding 28 percent share, Microsoft 21 percent, and Google 14 percent.[27] In that quarter Google Cloud grew 63 percent, against roughly 40 percent for Azure and 19 percent for AWS.[1][28]
Its strategy is the most vertically integrated of the hyperscalers: in-house silicon (TPUs and Axion), its own frontier models from Google DeepMind (a Gemini 3 launch that Pichai called "a major milestone" for the business), a model marketplace that carries competitors' models, and now TPU hardware sales aimed at Nvidia's data center franchise.[4][9] The bet is that demand keeps outrunning supply: Alphabet has repeatedly told investors that Google Cloud remains capacity constrained, and the $462 billion backlog implies years of contracted growth if it can build data centers fast enough.[2]