Google's Antigravity Project Crashes to Earth Amid Soaring Compute Demands

Google’s Antigravity service has been throttled to a near‑standstill after a surge of third‑party‑driven compute demand overwhelmed the platform’s backend. According to The Register, the company began suspending AI Ultra customers who paired the Antigravity agent‑development app with external wrappers such as OpenClaw and OpenCode, citing “malicious usage” that degraded service quality for other users [Theregister]. The bans were applied without warning, targeting accounts that paid the $250‑per‑month subscription fee but allegedly exceeded the intended usage patterns for the Antigravity backend.

Varun Mohan, a former Windsurf co‑founder now working at DeepMind, defended the crackdown as a necessary stopgap. In a Sunday social‑media post he said Google had seen “a massive increase in malicious usage of the Antigravity backend that has tremendously degraded the quality of service for our users” and that the company needed to “quickly shut off access to these users that are not using the product as intended” [Theregister]. Mohan clarified that the suspension applied only to Antigravity, not to Google’s broader suite of AI services, and that the offending users were effectively using the Antigravity API as a proxy to run large‑scale autonomous workloads that the subscription tier was not provisioned to handle.

The controversy centers on whether the third‑party integrations truly constitute a breach of Google’s Terms of Service. Several developers, including AI engineer Mohan Prakash, argue that they operated within their purchased quota and that the ToS does not explicitly forbid OpenClaw‑based harnesses. Prakash posted that “the real issue is the [Terms of Service] doesn’t explicitly ban OpenClaw integration, so users assumed it was allowed,” and warned that “banning paying customers without warning is how you lose trust faster than you lose capacity” [Theregister]. The Register’s request for concrete examples of “malicious” behavior has gone unanswered, leaving the precise nature of the alleged abuse ambiguous.

Google’s response mirrors actions taken by rivals such as Anthropic, which recently blocked token‑price arbitrage by preventing subscription accounts from feeding lower‑cost APIs into third‑party services. Both firms appear to be tightening control over how their generative‑model tokens are consumed, especially as enterprise demand for high‑throughput AI workloads spikes. The Register notes that Google’s free‑tier token availability may have inadvertently encouraged developers to route large volumes of compute through Antigravity, amplifying the strain on the platform’s infrastructure.

Industry observers see the episode as a symptom of a broader scaling challenge. As compute‑intensive applications proliferate, cloud providers must reconcile generous pricing models with the finite capacity of their GPU farms. While Google has not disclosed the exact compute metrics that triggered the shutdown, the rapid suspension of multiple accounts suggests a threshold breach that could jeopardize service‑level guarantees for paying customers. The episode also raises questions about transparency: without clear guidance on permissible integrations, developers risk inadvertent violations that can lead to abrupt service loss.

The fallout may prompt Google to revise its API policies and introduce explicit error codes for disallowed third‑party wrappers, a practice already employed by Anthropic. Until then, developers using Antigravity are left to navigate a murky compliance landscape, balancing the lure of powerful autonomous agents against the risk of being “grounded” by a platform that is scrambling to keep its compute engines from overheating.