Google's entire suite of apps has been "contaminated" by the new model.

Google's Gemini 3.5 Flash is being criticized for poor output quality, high token consumption, and uneven benchmark performance, despite low sticker pricing of $1.5 per million input tokens and $9 per million output tokens. In an Artificial Analysis suite, Flash cost $1,552 to complete tasks versus $282 for Gemini 3 Flash and about $870 for Gemini 3.1 Pro, while long-context performance fell sharply from 77.3% at 128k to 26.6% at 1M on MRCR v2. Google Cloud revenue rose 63% year over year to $20 billion and backlog doubled to $462 billion, but the article argues Google still lacks a convincing flagship AI model and that Gemini product issues are degrading the user experience across Search and Workspace.

Analysis

The market is starting to separate Google’s model stack into two businesses: a consumer-facing intelligence product and a compute-intensive inference engine. The negative read-through is not just to Gemini’s brand, but to unit economics; if the model needs far more turns and longer contexts to complete work, margin leverage migrates away from software and toward the infrastructure layer. That dynamic is favorable for the picks-and-shovels names in the AI supply chain, while Google’s own ad/search franchise risks higher operating friction if AI features remain unstable or token-hungry.

The second-order issue is competitive positioning in enterprise. A fast-but-weak model can still win as a sub-agent, but only if the planner layer is clearly superior; until 3.5 Pro is proven, Google is effectively asking customers to pay for an unfinished architecture. That creates a window for rival frontier models to capture higher-value reasoning workloads, while Google becomes the default execution layer inside its own ecosystem—a lower-ARPU, lower-control role.

The hardware angle is more important than the model disappointment suggests. If cloud demand is constrained by available compute rather than customer interest, then near-term upside accrues to TPU capacity, networking, and power delivery rather than to the model brand itself. That’s bullish for Broadcom on custom silicon content and for Google’s infrastructure monetization, but it also means capex intensity likely stays elevated longer, which can cap multiple expansion in GOOGL until investors see cleaner software monetization.

Key Decisions for Investors

• Short GOOGL into the 3.5 Pro release window over the next 2-6 weeks; risk/reward favors a tactical fade if product quality remains uneven and AI search friction continues to pressure sentiment.
• Go long AVGO on any post-news weakness for a 1-3 month horizon; custom silicon demand and TPU scaling should be the cleaner monetization path than Google software, with downside cushioned by diversified AI exposure.
• Pair trade: long AVGO / short GOOGL for 1-2 quarters; thesis is that infrastructure spend remains real even if model adoption disappoints, while GOOGL bears execution and brand risk.
• Keep META on a relative-value short vs AVGO rather than outright short; it is less exposed to Google’s product issues, but the broader AI-capex cycle still favors hardware vendors over app-layer monetization.
• Use GOOGL call spreads only after evidence of 3.5 Pro stabilization; otherwise the setup is asymmetric to the downside because the current market is paying for a narrative reset that has not yet happened.