IBM Powers up; Deepgram talks code

Welcome to Runtime! Today on Product Saturday: IBM goes to 11, Deepgram launches voice coding, and the quote of the week.

(Was this email forwarded to you? Sign up here to get Runtime each week.)

Ship it

Scale up?: Most applications written over the last couple of decades were designed to run on chips from Intel or AMD inside relatively cheap servers from hardware vendors or cloud providers. But IBM has continued to develop its Power processor over that time for companies that put a premium on uptime and performance, and this week it released Power11.

"Power11 is designed to be the most resilient server in the history of the IBM Power platform, with 99.9999% of uptime," IBM said in a press release that every Power customer should hang on their wall after highlighting that number. Power servers tend to be used in regulated industries where data has to be stored and processed on premises, such as banking or health care.

Sharing is caring: As companies training AI models start to run out of easy sources of data, they can either turn to synthetic data sources like Writer or pool their data together with partners working on similar models, but the idea of sharing ownership of that data is a non-starter inside many tech organizations. The Allen Institute for AI introduced a new data-sharing framework this week that could allow companies to work together on model training while retaining full control over their corporate data.

"FlexOlmo enables data owners to contribute to the shared model without directly sharing their raw data," and "data contributors retain control over when their data contributions are active in the model and can deactivate them at any time," AI2 said in a blog post. The technology remains a work in progress, but early tests point to "10.1% better model performance than earlier approaches to merging neural networks," according to SiliconAngle.

Rollin' in the deep: Generative AI is having an outsized impact on the process of software development by allowing developers to create code with natural-language commands. On Friday Deepgram launched Saga, which was designed to enable vibe coding by voice.

Saga is "a universal voice interface that lets you control your dev workflow with natural speech," the company said in a blog post. Speech-to-text development tools have been around for a while, but Deepgram argued that Saga takes the idea a step further by integrating directly into tools that developers are already using, like IDEs and code editors.

Running hot: Companies interested in building generative AI models need to make a crucial decision pretty early on in the process: Where are they going to run? Vendors are falling over themselves in hopes of convincing companies to put their models inside their pretty agentic gardens, but Clarifai introduced a product this week that takes a different approach.

AI Runners is "essentially ngrok for AI models, letting you build on your current setup and keep your models exactly where you want them, yet still get all the power and robustness of Clarifai's API for your biggest agentic AI ideas," Clarifai's Alfredo Ramos said in a press release. Businesses experimenting with proof-of-concept models would be able to test demand for their models without moving them wholesale to a third-party infrastructure platform, which can set off the lock-in night terrors.

Business time: Fifteen years ago Rackspace had high hopes that OpenStack could become an open-source alternative to AWS, but that outcome was probably never in the cards. However, OpenStack remains alive and kicking in 2025, and this week Rackspace introduced a private cloud service based around the project in hopes of capitalizing on the chaos at Broadcom/VMware.

Rackspace OpenStack Business is "especially well-suited for performance-sensitive applications that require guaranteed system resources, as well as regulated industries that demand single-tenant environments to meet strict compliance requirements," the company said in a press release. It's an interesting time for hybrid cloud strategies as businesses have shown increased interest in running AI workloads on premises while maintaining their investments in public cloud infrastructure.

Stat of the week

Software organizations have gotten the message: 90% of developers and engineering managers recently surveyed by Jellyfish have adopted AI coding tools in some fashion. However, the survey also indicates that those organizations are still looking for results, with just 25% reporting an "increase in developer velocity and productivity via AI coding."

Quote of the week

"I think it may discover new technologies as soon as later this year. I would be shocked if it has not done so next year. So I would expect Grok to, yeah, literally discover new technologies that are actually useful no later than next year, and maybe end of this year. And it might discover new physics next year, and within two years, I’d say almost certainly.” — Elon Musk, fantasizing about the impact Grok 4 could have on the world while presumably somewhere around Barstow, on the edge of the desert.

The Runtime roundup

OpenAI's bid to acquire Windsurf for $3 billion is kaput and CEO Varun Mohan has agreed to join Google DeepMind, according to The Verge, after OpenAI and Microsoft appeared unable to agree on whether Windsurf's IP would have been accessible to Microsoft as part of their partnership deal.

CISA warned government agencies and private companies that a new vulnerability in Citrix's NetScaler product is being actively exploited, urging them to patch sooner rather than later.

Thanks for reading — see you Tuesday!