Competitive Brief: Call Sheet
Researched: AWS · for Google Cloud reps
Focus: AI/ML infrastructure
Today's angle
On September 22, 2026, AWS made OpenAI's GPT-6 Sol and GPT-6 Luna generally available on Amazon Bedrock, extending the GPT-6 family (alongside Astra) with two models that support up to 1M tokens of context and ship at lower prices than their GPT-5.6 predecessors. This lands the same week AWS also added Claude Opus 5.5 to Bedrock.
Top 2 plays
PLAY 01
We win the largest-scale training and inference workloads on silicon depth.
For AI-native startups, frontier labs and research-heavy enterprises, Google's 8th-gen TPUs (8t/8i) and single-fabric scale to 1M+ chips are a genuine moat: customers run on the same hardware that trains Gemini. AWS Trainium is competitive on cost-per-token but supply-locked and scales across multiple clusters, not one fabric.
"You can run on the exact infrastructure that trains Gemini (one fabric, a million-plus chips) and you can actually get the capacity, today."
PLAY 02
Google ships frontier model gains faster than buyers can finish evaluating the last one.
On September 2, 2026, Google released Gemini 3.8 Flash, its third Flash-tier model in six weeks after 3.7 Flash landed just three weeks earlier. 3.8 Flash holds the exact same price as 3.7 Flash, $0.75 per million input tokens and $3.75 per million output tokens through the end of 2026, while its DeepSWE v1.1 long-horizon coding score climbs to 73.7% from 65.3%. It went live at once across the Gemini app, Gemini Enterprise, Google AI Studio, the Gemini API, Android Studio, Google Antigravity and AI Mode in Google Search, not a staged rollout.
Google has shipped three Flash upgrades in six weeks at the same price, each one better at coding than the last, so your model keeps improving without your bill moving.
Objection handling
"Google Cloud had that massive 2025 outage. Why trust critical AI to them?"
Counter: Don't minimize it: acknowledge, point to the concrete process changes and pivot the reliability conversation to forward architecture, not past incidents.
"Google kills products. You just retired Vertex AI. Why bet our ML platform on that?"
Counter: Convert the deprecation fear into contractual commitments rather than denying the pattern; that's what earns credibility.
"Even Pichai says Google Cloud is compute-constrained. We'll be stuck in a queue."
Counter: Turn the constraint into proof of demand, and lock in committed capacity now, because AWS is just as supply-limited.
"We want Claude, and that means AWS Bedrock."
Counter: Neutralize the "Claude = AWS" reflex early; make the conversation about which infrastructure runs Claude best, not which cloud has it.
"AWS is the market leader. That's the safe pick."
Counter: Concede the current-share point, then redirect to a 3 to 5 year trajectory framing where the growth gap and AI differentiation favor Google.
Agent Scout · every claim verified against its source · 67 claims tracked