# Runtime: Open season for AI routing; How TFMs tackle structured data; GitHub DDoS'd itself > Source: > Published: 2026-08-20 15:00:50+00:00 [Runtime](/tag/runtime/) *Welcome to Runtime! Today: Stripe's acquisition of OpenRouter shows how important AI gateways have become to enterprise tech, tabular foundation models could help businesses unlock their structured data, and GitHub explains what went wrong on Monday.* *Please forward this email to a friend or colleague! If it was forwarded to you, *__sign up here__* to get Runtime for free every week, or*__ level up here__*.* ### First up: Find your route "Tokenmaxxing" will be one of those 2026 buzzwords that will evoke wry chuckles in a few years, given how quickly enterprise tech embraced and abandoned the idea that burning AI tokens would translate into a competitive advantage. Companies want to use coding agents and build apps around AI models, but as open-weight AI models have approached frontier-model performance at a fraction of the cost, they want to pick and choose the right tools for the right job. OpenRouter is just three years old, but it quickly convinced 10 million developers that its AI gateway software was the right tool to help them assign certain workloads to certain AI models. On Wednesday Stripe announced that it plans to acquire OpenRouter for a reported $7.5 billion, one of the largest AI infrastructure exits since ChatGPT dropped. "OpenRouter has built a platform that helps businesses dynamically evaluate each request, routing it to the optimal model based on task complexity, price, speed, and reliability, and is already used by the likes of Nvidia, Zoom, and Lovable," Stripe[ said in its announcement](https://stripe.com/newsroom/news/stripe-agrees-to-acquire-openrouter?ref=thestack.technology). While the frontier model companies appear to be [, it's still very expensive to use the flagship models from either company for targeted workloads that don't necessarily need to access that much horsepower to return a result.](https://www.thestack.technology/anthropic-follows-openai-with-frontier-model-price-cuts/) __kicking off a bit of a price war__However, while OpenRouter might have impeccable street cred with developers at this point, Stripe is entering a crowded space as established enterprise software vendors turn their attention to the AI gateway as a control point for enterprise AI in general. The infrastructure layers that help companies run enterprise AI are starting to get a lot of attention, similar to how a whole ecosystem of software companies built on top of the cloud providers emerged a decade ago, and most companies have yet to build their first AI agent. ### The rest of The Stack **Ramp up**: FinOps and AI routing look like they're going to be joined at the hip for the forseeable future, which means it's not just enterprise software companies that want a piece of that business. Ramp, which helps companies track their spending, launched its own AI model routing service Wednesday at router.com. **Totally tabular**: LLMs have been a revelation in the enterprise because they helped companies make sense of their unstructured data, but so much of their data remains locked away in spreadsheets and structured databases that LLMs struggle to interpret. Tabular foundation models could be the answer, and although it's very early SAP's $1.2 billion acquisition of Prior Labs indicates that a market could be developing for these tools. **Try, try again**: A cascading series of errors took down much of GitHub's core services on Monday, the company acknowledged in a root-cause analysis of the incident released on Tuesday. A traffic surge on a Monday morning overwhelmed an Istio sidecar, which missed its signal to scale up and caused several downstream errors exacerbated by "optimistic retry logic" from internal systems causing an even greater surge in traffic. **August vacation**: OpenAI responded to the embarrassing Hugging Face incident by shutting down reinforcement learning training on unreleased models for two weeks until it could improve its security policies and monitoring tools, the company said Tuesday. Training on one future model remains on hold as it evaluates the new policies, and the new monitoring services will further reduce its margins with a 20% increase in computing power required to detect and prevent further incidents. **Sound up**: In this week's One to Watch, Noah Bovenizer checked in with HappyRobot, which just raised $143 million in Series C funding for its voice AI models. "We can go to a customer, and show these fantastic human-level voice agents that can not only talk but also coordinate through WhatsApp and Teams, and look into a database and remember a memory snippet like, 'your router broke down yesterday, I'm going to call you back in an hour,'" said co-founder and CEO Pablo Palafox. ### Stack ranking: Enterprise moves **Ahmed Shama** is [ the new CEO of CloudWerx](https://financialpost.com/pmn/business-wire-news-releases-pmn/cloudwerx-names-ahmed-shama-as-chief-executive-officer-to-lead-next-phase-of-growth?ref=thestack.technology), joining the Google Cloud service provider after leadership roles at T-Mobile, Google, and Slalom Consulting. **Jason Forget** is [ the new president and chief revenue officer at Resolve.ai](https://resolve.ai/news/jason-forget-president-chief-revenue-officer?ref=thestack.technology), joining the AI agent management company after serving in a similar role at Cockroach Labs. **Mat Velloso** and **Vinh Le** are [ the new chief product officer and chief financial officer](https://www.docker.com/press-release/docker-strengthens-leadership-team-with-appointments-of-mat-velloso-as-cpo-and-vinh-le-as-cfo/?ref=thestack.technology), respectively, at Docker. **Patti Degnan** is [ the new chief information security officer at Glow](https://www.glow.io/blogs/glow-appoints-patti-degnan-as-chief-information-security-officer?ref=thestack.technology), joining the endpoint AI company after serving as an operating partner with Andreessen Horowitz and CISO of Notion. ### We're also reading: **"**__We turned off Pub/Sub and nobody noticed,__**" a blog post from Patrick Hamann and Mike Fisher of Incident.io**, describing how it removed a single point of dependency on Google's messaging broker and added some resiliency. *Thanks for reading — see you Saturday!*