The usual shorthand for the cloud market is “the big three.” The phrase suggests three companies running at the same speed. The 2026 numbers say otherwise. Third place is running fastest, and the reason why is the key to understanding this company.
Where it stands, in numbers
In Alphabet’s Q2 2026 results, the Google Cloud segment posted revenue of $24.77 billion. That is 82% growth from $13.62 billion a year earlier, and comfortably above the roughly $22.4 billion the market expected. The more telling figure is profitability: the same quarter brought $8.8 billion in operating income at a 35.6% margin. This was not growth bought with discounts.
Contracts are already running ahead of revenue. Backlog — signed business not yet recognised — stands at $514 billion, up more than $50 billion from the previous quarter. The company expects about half of it to convert to revenue within 24 months.
Market share is a separate story. Estimates differ by firm, but as of 2026 the rough split is AWS 28–30%, Azure 21–25%, Google Cloud 13–15% — still third. Growing fast and holding the market are not the same claim. What is true is that the gap is closing faster than in any recent year.
Where it came from: a company that sells its own infrastructure
Google Cloud’s product line is largely made of things Google built first to run its own services. It started with App Engine in 2008; BigQuery in 2012 and the general availability of Compute Engine in 2013 gave it today’s skeleton. Kubernetes, now the standard for container orchestration, came out of Google’s experience running its internal cluster manager (Borg) and was open-sourced in 2014.
So the company’s strength comes less from marketing than from selling the very machinery built to survive Search and YouTube scale. The weakness comes from the same place. Its enterprise sales and support organisation arrived late. After Thomas Kurian took over in 2019, filling that gap absorbed a great deal of time and headcount.
What sets it apart
① Its own silicon. At Cloud Next in April 2026 it split the 8th-generation TPU into training and inference parts. The training TPU 8t binds up to 9,600 chips and 2 petabytes of shared high-bandwidth memory into a single superpod, claiming three times the throughput of the previous generation (Ironwood). The inference TPU 8i packs 1,152 chips per pod with three times the on-chip SRAM. Axion, its own Arm CPU, follows the same logic. Building the chips instead of buying them means a different cost structure from everyone else.
② Models and platform. The axis is the Gemini model family plus the Gemini Enterprise Agent Platform, announced in April 2026 as the evolution of Vertex AI. As of the earnings call, customers were pushing 16 billion tokens per minute through the API directly (up from 10 billion the previous quarter), and close to 500 cloud customers had processed more than a trillion tokens over the past year.
③ Data. BigQuery is still this company’s headline weapon. Running large-scale analytics per query without standing up servers is a shape competitors copied later.
④ Security. On 11 March 2026, Google closed its $32 billion acquisition of cloud security company Wiz — the largest acquisition in its history. Wiz keeps its brand and continues to support multicloud, meaning it is still sold to organisations that do not use Google Cloud at all.
Using it from Korea
The Seoul region (asia-northeast3) has been running since 2020, so data-residency requirements can be met domestically. Pairing it with the Tokyo and Osaka regions for redundancy is a common design. What you should weigh is the reality that the overwhelming majority of Korean job postings ask for AWS experience. The technical choice and the career choice do not always give the same answer.
If you are meeting it for the first time as a developer
- The free on-ramp is fairly wide. There is an always-free tier and new-signup credits, and BigQuery has its own monthly free query allowance.
- Pricing works differently. Sustained-use discounts that apply automatically when something stays on, and one-to-three-year committed use discounts (CUDs), are the basics. The most expensive habit is leaving an instance running and forgetting it.
- The certification path usually runs Associate Cloud Engineer → Professional Cloud Architect / Data Engineer / ML Engineer.
- Look at the weak spots too. A reputation for retiring products, a smaller ecosystem than AWS (examples, third-party tooling, Korean-language material), and support response times that disappoint depending on the situation are the criticisms that come up again and again in real adoption reviews.
⚠️ The figures here come from Alphabet’s Q2 2026 results and public material as of September 2026. Cloud market share varies between research firms because they measure different things, and pricing, free tiers and product names change often. Check the official pricing pages and per-region product availability directly before making an adoption decision.
Frequently asked questions
Is Google Cloud cheaper than AWS?
It is hard to say it is cheaper across the board. There are, however, bands where it wins: sustained-use discounts applied automatically to long-running instances, committed use discounts, and usage-metered services like BigQuery. The result flips depending on workload shape, so pricing it out against your actual usage pattern is more accurate.
Is there a reason to pick Google Cloud specifically for AI work?
The biggest difference is being able to use its own TPUs and the Gemini models on the same platform. In April 2026 it split the chips into the training TPU 8t and the inference TPU 8i, emphasising inference cost efficiency. That said, if your existing code is tied to a particular GPU ecosystem, calculate the porting cost first.
Is there a region in Korea?
Yes. The Seoul region (asia-northeast3) has been operating since 2020, so domestic data-residency requirements can be met. Not every new service opens in every region at once, though, so check individually whether the service you need is offered in Seoul.
Leave a Reply