Breaking the Bottleneck | Issue 109
[2/16/2026] Industrial Software Switching Costs, MaintainX on GTM Adoption, Gemini Deep Think 3 and more!
Breaking the Bottleneck is a weekly-ish newsletter and interview series about manufacturing and physical AI. If you want to chat, feel free to reach out at aditya@machinafactory.org or on LinkedIn here!
One Big Thing… Industrial Software Switching Costs
The recent wave of legacy-provider AI launches, amid a broader SaaS meltdown, has me thinking about the switching costs for large enterprises adopting new AI products. The marginal cost of software development is declining, which, counterintuitively, helps incumbents. These platforms can further entrench themselves by shipping AI features on top of existing installations. For enterprise buyers, it's infinitely easier to flip on a vendor's native AI capability than to navigate platform approvals, architecture reviews, and data governance exercises for a new vendor. This is the "front door" risk in reverse: the incumbents and their ecosystem of SIs and consultants are the front door for most large industrial enterprises, and they know it. Unless a challenger fundamentally rethinks the workflow, not just the interface, displacing an ERP or MES that's already wired into a company's operations is extremely difficult. It gets even harder when incumbents expose MCP servers that let enterprises build custom AI workflows on top of existing systems, removing the need for a challenger entirely. AI will enable bespoke customization with low switching costs in the SMB market, where systems of record are lighter and less embedded. But in larger industrial enterprises? I find it hard to see a world where large systems of record get replaced by upstarts, even ones that solve a killer use case, because solving a use case and replacing a system of record are two very different problems. The system of record might get pushed down the stack over time (see the OpenAI Frontier below), reduced to something closer to middleware, but if built correctly one that serves its own agentic execution layer.
Content I Enjoyed Last Week 🗞️🔬 📚
Some Interesting Reads:
The CTO of MaintainX offers an interesting perspective on bottoms-up adoption for AI transformation, arguing that companies should start with “the wrench rather than the dashboard” by building software and, importantly, UX/interaction mechanisms that fit technician workflows, voice interfaces being a prime example. When tools align with how frontline workers actually operate, adoption soars, enabling actionable metrics on breakdown frequency and repair times. Combining operator input with sensor data creates a complete operational picture that solves pressing day-to-day problems through predictive alerts, producing cleaner patterns for organization-wide use. [Fast Company]
An awesome deep dive report on humanoid hardware, with everything from the components to the landscape of startups, key suppliers, and geopolitics. [Sourish Jasti, Zoey Tang, Intel Chen, Vishnu Mano]
A Cato Institute brief on tariffs’ impact on US Manufacturing. [Cato]
More robots making robots [Unitree]
Intrinsic and Open Robotics launched an open competition for developers and roboticists. [Intrinsic]
Interesting analysis by NAM on the cost of freight congestion to US Manufacturers [NAM]
Infosys released its Manufacturing Tech Index, and surprise, everyone’s adopting AI. [Infosys]
An interesting deep dive into Chinese glassmaker Fuyao’s success at its Ohio factory as a lens into ways manufacturers might skirt Trump’s tariffs, and then the corresponding impact on US Manufacturing. [WSJ]
GPT‑5.2 derives a new result in theoretical physics [OpenAI]
Ouster acquired Stereo Labs, creating a unified sensing and perception platform. [Ouster]
McKinsey, with a breakdown of areas/opportunities for improvement across the Defense Industrial Base [McKinsey]
Products & Releases:
Google announced a major upgrade to Gemini 3 Deep Think, demonstrating its promise for physical-world applications. Check out the applications to semiconductor materials and physical prototyping below.
Vention introduced GRIIP, a unified pipeline from perception to motion by integrating its proprietary models with NVIDIA Isaac open models. [Vention]
Alibaba just released RynnBrain, it’s an embodied foundation model.
Emmi AI announced Noether, an open-source deep learning framework for engineering AI. [Emmi AI]
Oracle becomes another system of record, releasing its own AI Agents for manufacturing. [Manufacturing Digital]
Tweets & Blogs
An Interview with Ryan Kelly from Manufacturing Tech
Manufacturing “Coming Back” + AI from Grumpy Chinese Guy
Finance & Transactions 💵
Apptronik raised $520M from AT&T Ventures, John Deere, Mercedes, and others
Robotics / Physical AI • Growth • USA
Gather AI raised $40M from Smith Point Capital
Robotics / Physical AI • Series B • USA
Trener raised $32M from Engine Ventures and IAG
Robotics / Physical AI • Series A • USA
Didero raised $30M from Chemistry and Headline
Manufacturing, Procurement • Series A • USA
Planned Downtime 🧑🔧
Rivian R2 First Drive
Breaking the Bottleneck is brought to you with support from our partners.
Jiga - Is a platform that connects hardware teams with vetted manufacturers for custom parts. Instead of black-box instant quotes, you see exactly who’s making your parts and can talk to them directly.
Industry 4.0 Club - Leading cross-functional group accelerating the adoption of Industry 4.0, delivering better experiences to consumers, better profits to manufacturers, and better jobs to factory workers.
OMNI VC - Leading manufacturing tech VC, investing in Manufacturing Tech startups at the earliest possible stage.