Good day!
While on the Ultra AI subscription, I constantly encounter the error “Our servers are experiencing high traffic right now, please try again in a minute.”
Theoretically, shouldn’t this be avoided with an Ultra AI subscription? Or am I missing something? 
Is there any way to resolve this issue?
Trajectory ID: efd3a75b-b86d-4a71-
Error: HTTP 503 Service Unavailable
Sherlog:
TraceID: 0xb2dee47
Headers: {"Alt-Svc":["h3=\":443\"; ma=2592000,h3-29=\":443\"; ma=2592000"],"Content-Length":["429"],"Content-Type":["text/event-stream"],"Date":["Thu, 19 Mar 2026 12:03:36 GMT"],"Server":["ESF"],"Server-Timing":["gfet4t7; dur=1576"],"Vary":["Origin","X-Origin","Referer"],"X-Cloudaicompanion-Trace-Id":["b2dee474f04a82ca"],"X-Content-Type-Options":["nosniff"],"X-Frame-Options":["SAMEORIGIN"],"X-Xss-Protection":["0"]}
{
"error": {
"code": 503,
"details": [
{
"@type": "type.googleapis.com/google.rpc.ErrorInfo",
"domain": "cloudcode-pa.googleapis.com",
"metadata": {
"model": "claude-opus-4-6-thinking"
},
"reason": "MODEL_CAPACITY_EXHAUSTED"
}
],
"message": "No capacity available for model claude-opus-4-6-thinking on the server",
"status": "UNAVAILABLE"
}
}
Hello @evjanich,
We’re finding new ways to share more capacity with the Antigravity community. During windows where we have unused capacity, currently in the late afternoon PST, you’ll notice that your baseline quota stretches further. You’ll get more requests and less downtime within your existing plan.
As global demand shifts, we’ll continue to adjust these windows to pass any available capacity back to you. We’re committed to expanding access and ensuring every developer has the capacity they need to keep building.
Question: are you absolutely sure that this is normal Antigravity performance on an Ultra AI subscription?
Hi Abhijit. The ‘capacity sharing’ narrative might sound good for PR, but from a system architecture and operational standpoint, it’s fundamentally flawed.
First, publicly advising the entire community to push their workloads to ‘late afternoon PST’ is a naive approach to load balancing. Anyone who understands backend systems knows this is a textbook recipe for the Thundering Herd problem. When thousands of devs schedule their agent tasks for that exact window, you aren’t solving the bottleneck; you are just shifting the peak load and essentially self-DDoSing your own servers. The 503 errors will just happen at a different time.
Second, I am a paying Ultra subscriber. Subscribing to the highest tier implies paying for high availability and priority throughput, not paying to play a guessing game for ‘leftover capacity’ during off-peak hours like a free-tier user.
Paying customers expect proper hardware capacity that meets the SLA, or at the very least, a straightforward apology for the current infrastructure shortfall—not corporate jargon about ‘adjusting to global demand.’ When can we expect an actual server capacity upgrade, rather than being told to adjust our own working schedules to accommodate your limitations?
