In partnership with
Join Fin, Anthropic, and Clay at Pioneer on October 7th
Pioneer, the summit where CX leaders redefine what’s possible, is on October 7th.
Join leaders from Fin, Anthropic, and Clay for an insightful conversation on the state of AI transformation.
You’ll discover how some of the most innovative minds in CX have transformed their organizations, learn how they think about CX, and hear how they're planning for what's next.
Join the conversation in San Francisco, or tune in virtually.
ShanghAI - The $90/Mo perfect outbound stack
Hey Lobsters, welcome to this new edition of ShanghAI.
A post went viral on LinkedIn this week where a founder flexed a "lean marketing stack" that took their startup to amazing results this month.
It looked slick on the surface: HeyReach for LinkedIn, Instantly for cold email, Claude for inboxes, Apify for scraping, BlogSEO for search, Wispr Flow for dictation, and Loom for video recaps.
Let's do the actual math that Western SaaS influencers never show on their feed.
HeyReach is $99/mo. Instantly runs $97/mo. Claude Team seats and API calls burn another $120/mo. Apify takes $49/mo. BlogSEO takes $49/mo. Add Wispr, Loom, and ChatGPT Plus, and this "lean" startup is silently bleeding over $550 every single month just to run basic scraping, drafting, and outbound messaging.
Western founders love bragging about autonomous AI systems that are really just eight separate subscription markups taped together with Zapier.
If your system breaks every time an API provider tweaks their seat pricing or updates their tier limits, you haven't built an operational moat; you’ve built a high-maintenance cash drain.
Inside the Chinese tech ecosystem, growth teams run 10x that volume at an 80% discount. They don't buy expensive on-premise H100 clusters, nor do they pay closed-source SaaS markups.
Instead, they pair lightweight open-source orchestration with high-efficiency open-weight models and lean proxy routing.
Let’s break down the realistic architecture to slash this overhead from $550+ down to under $100 a month.
Implementation & Advisory Note: Looking to deploy this architecture internally? ShanghAI provides enterprise advisory, token audits, and turnkey private cloud deployments for DeepSeek, Qwen, and ByteDance stacks with zero data egress and GDPR compliance. Explore LeanAI Deployment Services.
OUR PARTNER
Become an AI-ready CMO
You already understand how the models work. AI-Ready CMO shows you what they do for GTM. Peter and Torsten, two CMOs who built marketing at high-growth scaleups, write a daily briefing on AI for marketing: which developments change pipeline, positioning and team structure, and which are noise. It's free, five days a week, and read by 20,000 senior marketing leaders. Everything they know is also open in the hub: workshops, courses, prompt library, tool reviews and playbooks.
DISCOVER
Token Arbitrage & SaaS Wrapper Trap
The core tasks of an outbound engine are simple:
Extract lead text,
Enrich company signals,
Classify intent,
Draft personalized responses.
Paying premium SaaS vendors for this workflow is an arbitrage against your own balance sheet.
When you break down the token economics, Claude 3.5 Sonnet charges $3.00 per million input tokens and $15.00 per million output tokens.
In contrast, calling frontier open-weight models like DeepSeek-V3 via direct serverless endpoints runs at roughly $0.27 to $0.28 per million input tokens and $0.42 to $1.10 per million output tokens: a fraction of the cost for identical output on extraction, classification, and short-form copy.
On non-coding tasks: general reasoning, comprehension, and structured JSON output, DeepSeek-V3 matches Anthropic models point for point. Paying a 1,400% markup to have Claude draft cold emails delivers zero measurable conversion lift.
The SaaS tax isn't just the AI, though, it is also the wrapper markups. Tools like BlogSEO and inbox assistants charge $50–$100/month for simple wrapper logic you can execute on a $10 VPS instance running self-hosted n8n or Dify.
ONLY FOR SUBSCRIBERS
Will AI Recommend Your Business?
See how ChatGPT, Gemini, Claude, and Perplexity understand your business.
USE
Slashing costs without burning accounts
To build an open growth engine that actually survives in production, you have to trade the $550 SaaS tax for a lean, defensible ~$90/month infrastructure:
The Orchestration Core ($10/mo): Spin up a dedicated VPS (Hetzner or DigitalOcean) hosting n8n or Dify. This eliminates Zapier and serves as your central command for webhooks, parsing, and cron triggers.
The Intelligence Layer ($5/mo): Route all extraction, scoring, and drafting calls to serverless DeepSeek-V3 or Qwen-2.5-72B endpoints. Generating 10,000 personalized emails and lead evaluations costs less than a cup of coffee in raw API usage.
Scraping & Signal Extraction ($25/mo): Replace third-party scrapers with containerized Playwright scripts. The critical key: route them through residential rotating proxies ($15–$25/mo) so you extract clean HTML signals without hitting Cloudflare blocks or 403 errors.
Deliverability & Inboxes ($50/mo): Never rely on naive custom SMTP relays that land straight in spam. Instead, allocate your saved budget where it matters: purchase 3 secondary domains, set up 6–10 Google inboxes, and connect them to a dedicated warmup/rotation pool.
By swapping inflated SaaS margins for open-weight intelligence and lean infrastructure, your monthly software overhead drops from $550+ down to ~$90/month, giving you an automated outbound machine that protects your deliverability, cuts your burn, and locks in your margins.
GIFT
The full replacement blueprint
I mapped out the entire end-to-end replacement stack into a unified orchestration blueprint.
It includes the n8n webhook routing schemas, the DeepSeek intent classification prompts, and the proxy-rotation configurations.
Free for subscribers, accessible in one click.
Until next week,
The ShanghAI Guy
