Google Antigravity falls to Earth under OpenClaw-fueled compute load
- Reference: 1771879777
- News link: https://www.theregister.co.uk/2026/02/23/google_antigravity_compute_burden/
- Source link:
Over the past few weeks, Google has been cutting off customers for terms of service violations, often with no warning. The problem for Google is that when third-party agent wrappers or harnesses rely on Google AI under the hood, the company's software isn't necessarily priced or provisioned for heavy autonomous usage.
Many of the software developers who subscribed to these AI services have expressed surprise in various [1]discussion [2]threads , claiming that they were unaware their actions violated contractual agreements they're unlikely to have actually read.
[3]
Varun Mohan, co-founder of Windsurf and now a DeepMind engineer, said that the account bans followed from malicious usage, though customers argue that that's a mischaracterization of what they were doing.
[4]
[5]
"We've been seeing a massive increase in malicious usage of the Antigravity backend that has tremendously degraded the quality of service for our users," he explained in [6]a social media post on Sunday.
"We needed to find a path to quickly shut off access to these users that are not using the product as intended. We understand that a subset of these users were not aware that this was against our [Terms of Service] and will get a path for them to come back on but we have limited capacity and want to be fair to our actual users."
[7]
Mohan went on to clarify that Google has blocked usage of Antigravity but not other Google services.
"It is not intended to use the Antigravity backend as a proxy for other products and users in these groups have overwhelmed our compute," he explained. "We are going to make sure we bring people back on but needed to act fast to make sure we deliver a good experience for people using the product."
[8]Microsoft execs worry AI will eat entry level coding jobs
[9]Linus Torvalds: Someone 'more competent who isn't afraid of numbers past the teens' will take over Linux one day
[10]Global regulators say AI image tools don't get a free pass on privacy rules
[11]Gemini users say their chat histories have quietly vanished
The Register asked Google to provide examples of malicious usage. We've not heard back.
Dissatisfied customers have expressed skepticism that using Antigravity and Gemini CLI with a third-party wrapper or harness like OpenClaw can legitimately be characterized as malicious.
"Users paid for quota, used quota within limits, got banned," said AI engineer Mohan Prakash in a social media [12]post . "That's not malicious, that's using the product you sold them. The real issue is the [Terms of Service] doesn't explicitly ban OpenClaw integration, so users assumed it was allowed. If you don't want harness integration, return an error like Anthropic does – this usage pattern is not permitted. Banning paying customers without warning is how you lose trust faster than you lose capacity."
[13]
Google's account purge follows [14]similar measures taken by Anthropic to prevent users from engaging in token price arbitrage by connecting subscription accounts, instead of its higher-cost API, to third-party services.
It appears that Google made machine learning model tokens too readily available through free tier accounts and non-specific quotas. As a result, the company wasn't prepared for the compute demand or the cost incurred when developers ran its AI services through third-party software.
This suggests what many [15]have already observed – that AI companies are selling tokens at far below cost to gain market share, in the hope they can outspend the competition until the competition goes away and they can raise prices. ®
Get our [16]Tech Resources
[1] https://discuss.ai.google.dev/t/account-restricted-without-warning-google-ai-ultra-oauth-via-openclaw/122778/4
[2] https://discuss.ai.google.dev/t/250-mo-ultra-subscriber-banned-without-warning-the-openclaw-mass-ban-wave-shows-a-systemic-failure-in-googles-developer-support/123015/23
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aZzcEuQwGnFUsOJROnh2mwAAAAw&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aZzcEuQwGnFUsOJROnh2mwAAAAw&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aZzcEuQwGnFUsOJROnh2mwAAAAw&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://x.com/_mohansolo/status/2025766889205739899?s=20
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aZzcEuQwGnFUsOJROnh2mwAAAAw&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://www.theregister.com/2026/02/23/microsoft_ai_entry_level_russinovich_hanselman/
[9] https://www.theregister.com/2026/02/23/linux_7_0_rc1/
[10] https://www.theregister.com/2026/02/23/privacy_watchdogs_ai_images/
[11] https://www.theregister.com/2026/02/23/gemini_users_say_their_chat/
[12] https://x.com/mohanp_ai/status/2025801195567112203?s=20
[13] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aZzcEuQwGnFUsOJROnh2mwAAAAw&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[14] https://www.theregister.com/2026/02/20/anthropic_clarifies_ban_third_party_claude_access/
[15] https://www.wheresyoured.at/why-everybody-is-losing-money-on-ai/
[16] https://whitepapers.theregister.com/
Re: agreements they're unlikely to have actually read
That was probably the problem !!!
:)
So, what subscription price DO they need per user for it to be profitable?
Supposedly even the $200 ones aren't. I can't get my company to pay for my dev tools sub, you think they'll shell out $300 (make that AU$500) or more per MONTH for "AI"?
Never mind home usage. Who's going to go "rent or AI sub, hmmm? Meaning, it doesn't matter even if there's only one hypester left standing, nobody will pay, never mind can afford, what they need to charge. For the *current* offerings, to say nothing of what they need to pay for their insane data centres and power plants they're waffling on about.
Sure, AI companies currently lose money on every token. The break-even for them is currently probably somewhere between 2000$ and 5000$ per sub per month.
These levels are not a sustainable business case for commercial augmenting use or home use.
The whole play is staying alive long enough, so they can continue to increase the size of models, which they expect to somehow magically cure all shortcomings of current AI and then sell your boss an much enhanced AI that (they say) can actually replace employees instead of augmenting them. Remember OpenAI talking about PhD-Level AI in early 2025? That's what they try.
Should they one day in fact succeed, they will charge 80% of your wages, so your (former) employer saves 20%.
Even if that would work, they would catapult the economy in a recession/depression worse than the 1929, due to an nearly instantaneous collapse of demand for most goods following widespread job losses.
There is IMHO simply no viable way for AI companies to not lose their investment.
Either AI does not work well enough to replace humans and they are toast - or they succeed, create the Great Depression Vol. 2 and are also toast.
So they either fizzle out as their supply of fresh money dries up or the whole economy blows up with a bang... Neither alternative will leave anyone insanely rich.
When the shoe is on the other foot.
Let's see, AI companies acting as "agents" accessing other services on behalf of their clients is a good thing and is the future of the world.
But when other people want to have software act as agents to access AI services on behalf of their clients or users, then that's "malicious activity".
It's fascinating how the whole AI gambit revolves around people not doing to the AI companies what the AI companies say they have a right to do to everyone else.
This is yet another sign that the whole industry as it exists now is built on sand.
Isn't it wonderful?
Yet another business model poking the bubble, scaling down the AI business to something more sustainable.
I mean computer programming is certainly not something those text generators are particularly useful at... particularly not at those prices. $200 is a lot of money when compared to programmer wages.
agreements they're unlikely to have actually read
Surely they had AI summarise it for them?