Google AI Pro and Colab: Does It Really Include A100 and H100 GPUs?

Google AI plans now include Colab, but the viral version of the story, that Google AI Pro gives you A100 and H100 GPUs and 200 compute units a month, goes further than anything Google has said. What Google confirms is narrower: every paid Google AI tier gets priority access to faster accelerators, and only Ultra gets Premium GPU access and background execution.

This post separates what is announced from what is rumor, shows how Colab's GPU access really works, and gives you a quick way to test your own account.

TL;DR: claimed vs verified

What Google actually announced

Google's developer blog says Colab is now part of Google AI plans. In its words, subscribers get "priority access to faster accelerators and more powerful machines," while Ultra subscribers also get "uninterrupted background execution and Premium GPU access." The post says the benefits are "rolling out over the next few weeks to Google AI subscribers in Colab supported countries."

It adds that "if you already subscribe to Colab there are no changes to your subscription," and that the two can be combined. The post names no GPU model, gives no compute-unit figure, and gives no limit on how long a background run can last.

That blog post is the primary source for the rest of this story.

Where the A100, H100 and 200-unit claims come from

The more specific claims trace to secondary sources. A Taiwanese tech outlet, INSIDE, published on September 27, reported that Google AI Pro subscribers receive 200 Colab compute units a month, enough to power high-end GPUs such as the A100 80GB, and linked it to the Colab command-line tool. It says the news spread in engineer communities. The article also says the outlet drafted it with AI assistance and had it reviewed by an editor, and it cites no Google page for the 200 figure.

Other writers fill the gap with guesses. One European blog describes Premium GPUs as "typically A100 and high-end TPUs," which is an inference, not Google's wording. The wrinkle is that these accounts disagree with each other. The INSIDE story puts A100 within reach of Pro. Google's own post puts "Premium GPU access" behind Ultra. Both can be partly true if A100 is already available to paid Colab users through compute units, since Colab has long sold premium hardware by the unit, but that is a reading, not a statement from Google.

How Colab GPU access works

Colab's official FAQ says the types of GPUs and TPUs available vary over time and that premium hardware on paid plans is subject to availability. It says eligible Google AI plans include a monthly allotment of Colab compute units and access to more powerful accelerators, without giving amounts or models. Storage-only Google One plans and free trials do not get the benefits.

So there are three separate things people blur together:

1. Which GPU you may select. A menu of T4, L4, A100 and H100, depending on plan, region and demand.
2. How fast you get one. Priority access is a queue position, not a quota. Google's post says priority, not unlimited.
3. How long it keeps running. Compute units are spent per hour of a GPU runtime, and an idle runtime still burns them. Background execution is what lets a run continue with the tab closed, and Google gives it only to Ultra.

None of that guarantees an A100 on a given afternoon: Google's FAQ says premium GPUs are subject to availability.

The Colab CLI and why agents care

The context that makes this plan change matter is the Colab command-line tool. Reporting on its June 2026 release, citing Google's developer blog, InfoQ and MarkTechPost, describes an open-source tool under Apache 2.0 that connects a local terminal to remote Colab runtimes. Commands include colab new, colab exec, colab download, colab log, colab repl and colab stop. Reported hardware options are T4, L4, A100 and H100 GPUs and v5e and v6e TPUs. It ships a skill file, COLAB_SKILL.md, so coding agents can drive it, and a demo shows an agent running a QLoRA fine-tune of a small Gemma model.

That is why "can my agent get a GPU from my subscription?" became the question. If you use agents like Claude Code, Codex or Google's Antigravity, a priority queue on a cheap plan is valuable on its own. Our coverage of Antigravity's CLI features and its remote control launch shows where Google is taking that agent workflow, and the guide to agent skills explains why a skill file like COLAB_SKILL.md is the glue.

Why the rumor spread so fast

Three things lined up. Google's announcement was short and named no hardware, which left room for readers to fill in the blanks. The Colab command-line tool had already put A100 and H100 in front of developers in June, so the names were familiar. And a flat "200 units" number is easy to repeat on social media, where it reads like a spec sheet even when no spec sheet exists.

This is the usual pattern with plan changes: the vendor describes tiers in general terms, an intermediary adds numbers, and the numbers get quoted as if the vendor had said them. The fix is boring but reliable. Look for the exact words on Google's own page, then look at what your account displays.

A fact-check magnifier over a claim bubble, illustrating how to verify the Google AI Pro Colab A100 claim

What this means for what you pay

A terminal prompt with a blinking cursor, representing agents calling Colab GPUs from the command line

Google is bundling. Colab joins Antigravity and AI Studio perks inside the same Google AI subscription, a pattern we traced in the Google AI product names glossary and in Google's push for startups.

For a practical decision:

Usage rules around AI subscriptions have been shifting everywhere. For comparison, see how Gemini Notebook introduced flexible limits with a five-hour reset and the timeline of Claude usage limits in 2026.

How to check your own account in two minutes

1. Open a new Colab notebook while signed into the Google account that holds your plan.
2. Go to Runtime, then Change runtime type, and list the hardware options. Note whether A100 or H100 is selectable and whether it is greyed out.
3. Open the resources or compute units panel and read the monthly number. That figure, not an article, is your allotment.
4. Run !nvidia-smi after connecting to see which card you were actually assigned.
5. Stop the runtime when you finish; idle GPU sessions still consume units.

If your account shows 200 units and A100 selectable, the viral claim holds for you. If not, remember Google said the rollout takes weeks and varies by country.

How to read vendor claims about plan perks

A useful habit with any subscription change is to sort statements into three buckets: quoted from the vendor, reported by a third party, and inferred by a writer. In this story the first bucket is small (priority access, Ultra-only Premium GPUs and background execution, unchanged Colab plans), the second holds the 200-unit figure, and the third holds the A100 and H100 labels for Pro. Only the first bucket should drive a purchase. If a number matters to your budget, wait for the plan page or your account dashboard to show it, and re-check after the rollout window closes.

What is still not answered

• Which accelerators count as Premium, and whether H100 is among them.
• How many compute units each Google AI tier carries, if any.
• The maximum length of Ultra's background execution.
• Whether Pro's priority extends to the Colab CLI the same way as to the browser.

Until Google publishes those details, the safe summary is the one in its own post: priority access for paid subscribers, Premium GPUs and background runs for Ultra, nothing taken away from existing Colab subscribers.

Plan details are from Google Developers Blog as of September 22, 2026 and this post was written October 10, 2026. Perks, prices and availability can change.

Related reading

• Antigravity CLI features: sandbox, plugins and subagents
• Google Antigravity remote control launch
• Google AI product names and rebrands, explained
• Gemini Notebook flexible usage limits
• Claude usage limits in 2026: a timeline
• What are agent skills
• Colab FAQ