LinkedIn and 3rd parties use essential and non-essential cookies to provide, secure, analyze and improve our Services, and to show you relevant ads (including professional and job ads) on and off LinkedIn. Learn more in our Cookie Policy.
Select Accept to consent or Reject to decline non-essential cookies for this use. You can update your choices at any time in your settings.
Merge is the connective infrastructure for production AI.
Its unified platform lets AI access business data, take action across SaaS systems, and control how models run in production.
Instead of building integrations, agent infrastructure, and model management in-house, teams integrate with Merge once.
Thousands of companies, from AI-native product teams to Fortune 500 engineering orgs, trust Merge to get AI from pilot to production faster.
You build the AI. Merge handles the infrastructure.
Private equity firms care about margins, not model names. Here's how one of our customers is cutting AI costs without cutting quality.
Our team is working with a local two-step distributor to improve how it uses AI, routing customer support requests across multiple LLMs based on quality, speed, and cost.
Merge Gateway gives the team more control, reduces reliance on any one model, and creates a playbook that can scale across a private equity portfolio.
Learn more or request a demo:
We're sponsoring the London AI Cost Hack next week! 🇬🇧
Hosted by Kaashif Hymabaccus, the challenge is simple: take an AI workflow and make it as cheap as possible without breaking the evals. Bring a team or come solo, hack away, and submit your solution as a public repo.
🗓️ Thursday, July 30th
📍 All Saints Church
🕠 5:30–9:00 PM
We're bringing Merge Gateway credits and swag to help you squeeze every last cent out of your model calls.
RSVP in Comments!
Going slow was the edge.
Shensi Ding spent 9 months researching before writing a line of code. Then she built Merge , the infrastructure frontier AI labs now run on.
Her bar for founding: only do it if you're willing to eat glass.
On Not Another Podcast, she makes the case for the boring, disciplined path in the most hyped market in tech.
NEW: Merge launches Embedded Routing Stack, and has a fascinating sales motion story now expanding beyond enterprise.
Shensi Ding (Co-Founder & CEO of Merge) joins GTMnow to break it all down.
Merge customers include companies like OpenAI, Mastercard, JP Morgan, and Amex.
Those are enterprise logos.
Most companies go PLG then expand to enterprise. Merge did it reverse: enterprise first, PLG later.
They now have 400+ enterprise customers and 20,000+ self-serve orgs, running side by side.
Key takeaways:
1️⃣ Merge ran the GTM playbook in reverse: enterprise first, PLG second.
Three years ago, Merge made the go up-market decision to go up-market while enterprise was greenfield and build the security and depth that make you hard to replace. It worked. Most revenue now comes from the 400+ enterprise logos like Mastercard, JP Morgan, and Amex. Only now is Merge adding a self-serve layer back on top, because AI buying got consumerist and developers need to know you, not just the C-suite.
2️⃣ Tokenmaxxing is so 2025 - 2026 is about intentional spend.
Last year everyone defaulted to the newest, biggest model because the cost was subsidized and experimentation was the point. Now teams are trained, spend adds up real quick, and open source has caught up. The unlock is smart routing: send more simple tasks to Haiku, not Opus 4.8, and save the expensive models for the tasks that need them.
3️⃣ Where a model is built and where your data lives are two different things.
The stigma around Chinese open source models conflates origin with hosting. A model built in China can run on a US host (Baseten, Bedrock, Fireworks) with US servers, and customers can block routing to China entirely. Shensi works directly with ByteDance, Alibaba (Qwen), and Kimi, with data retention terms and US fallback hosts built in.
4️⃣ Forward deployed engineers are the enterprise cheat code.
Merge's product fit 99% of enterprise use cases, but the last 1% blocked adoption. Forward deployed engineers closing that gap grew usage 5 to 10x almost immediately, then the pattern repeated across accounts.
5️⃣ Treat marketing like an engineering function and productize it.
When product started shipping faster than marketing could keep up, Merge broke marketing into repeatable, automated actions: changelogs pulled from GitHub, launch images and videos auto-generated in the brand voice. Even field dinners went from 1-2 a quarter to 3-5 a month!
Full interview:
- YouTube: https://lnkd.in/gE2vxh26
- Apple: https://lnkd.in/gbZub4J6
- Spotify: https://lnkd.in/g37Z3Psd
--
💚 Thank you to our incredible sponsor for supporting the podcast:
Primer - the targeting layer that lands your B2B ads on the right accounts across Meta, Google, and Reddit.
💡 GTMnow shares how the best in tech build, scale and invest.
With all this talk about AI Gateway routing, we ran the numbers on Merge so you don't have to.
Fixed on Claude Opus 4.8: $8.17 per task.
Intelligent routing: $2.87.
That's a 65% CUT in model spend across 120 trials on every prompt type we tested.
The overhead for routing? 90-650ms at the median. Barely a blip next to actual generation time.
The real question isn't "does routing save money," it's "what's it costing you to get there."
Merge Gateway:
- +1,146% month-over-month growth from May to June
- +3,000% weekly growth from the first full week to the last full week
- 66% of our user base was generated in the last 4 weeks
- Signups over the last 4 full weeks were 8x higher than the average during the first 4 weeks
We benchmarked intelligent routing on Merge Gateway against fixed flagship models.
Over 120 paired trials, intelligent routing model spend was 65% lower ($8.17 to $2.87), and correctness held at 239 of 240.
DeepSWE coding scores don't usually jump twelve points between minor model versions.
Gemini 3.6, Google's newest Flash model, just did that: 37% to 49% on DeepSWE, 78.4% to 83% on OSWorld computer use, and output pricing cut to $7.50 per million tokens.
It's live on Merge Gateway. Point your existing requests at it and go.
