Gemini’s new token limits are just as bad as Claude’s, and maybe even a little dumber

Google overhauled Gemini pricing and usage limits, cutting AI Ultra from $250 to $200 per month but adding a new $100 tier and tightening compute-based quotas. The changes sparked backlash as users report faster throttling, heavier restrictions on 3.1 Pro access, and worse reliability from 3.5 Flash versus older models. The update is likely negative for user retention and sentiment around Google’s consumer AI offering, though the impact is more product-level than market-wide.

Analysis

This looks less like a pure product tweak and more like Google re-pricing inference capacity to protect margins as usage scales. The second-order issue is that the company is signaling the most valuable AI behavior is not casual consumer engagement but bounded, monetizable workloads; that tends to slow adoption among exactly the heaviest, highest-LTV users who drive word-of-mouth and developer mindshare. In the near term, that can create a negative feedback loop: degraded perceived quality drives migration, while stricter quotas encourage users to internalize switching costs and explore alternatives.

For GOOGL, the bigger risk is not a one-quarter ARPU miss, but brand damage in the premium AI tier just as rivals are competing on reliability rather than feature breadth. If advanced users conclude older models are being intentionally throttled to push them onto cheaper, weaker defaults, churn risk rises over the next 1-3 months in power-user segments: researchers, coders, and enterprise prototypers. That cohort is small in consumer count but outsized in influence and willingness to pay, so even modest defections can pressure sentiment, app rankings, and eventually conversion into paid plans.

The upside case is that this is a temporary optimization move and most mainstream users never hit the limits, meaning headline outrage may exceed revenue impact. But the market will likely focus on the strategic signal: Google may be conceding that frontier AI is expensive enough to require tighter rationing, which implicitly validates competitor narratives around better model quality or better unit economics. That creates a watchpoint over the next several weeks for any evidence of softer engagement, higher refund requests, or an accelerated rollout of enterprise SKUs that can repackage the same constraints into contracts.

RDDT is not a direct beneficiary on the numbers today, but it can absorb spillover attention as users seek peer reviews, workarounds, and model comparisons; that makes the name a sentiment beneficiary rather than a fundamentals story. More broadly, the article supports a tradeable divergence between companies monetizing AI demand and those bearing inference cost, especially if consumer frustration persists long enough for alternative platforms and local-model tools to gain share.

Key Decisions for Investors

• Short GOOGL tactically for 2-6 weeks into post-I/O backlash digestion; use downside to monetize a sentiment overshoot, with the core risk being rapid clarification that limits are mostly immaterial for mainstream users.
• Buy GOOGL put spreads 1-2 months out rather than outright puts; the setup is more about multiple compression from AI trust concerns than a large earnings revision, so defined-risk bearish convexity is preferable.
• Pair trade: long RDDT / short GOOGL for 1-2 months if social chatter around model switching accelerates; thesis is that Reddit captures troubleshooting and comparison traffic while Google absorbs reputational damage.
• Fade the knee-jerk underperformance in GOOGL only after evidence of stable app usage or enterprise adoption emerges; if engagement holds through the next 4-6 weeks, the market may have over-discounted the impact.
• Watch competitor read-throughs in CHAT/LLM proxies and enterprise AI tooling names for 30-90 days; if Google’s throttling drives migration, those beneficiaries should outperform before the effect is visible in GOOGL fundamentals.