AWS Graviton goes to 5; Nova Forges ahead

Welcome to Runtime! Today on Product Saturday: A selection of interesting product launches from re:Invent 2025, and the quote of the week.

(Please forward this email to a friend or colleague! And if it was forwarded to you, sign up here to get Runtime each week.)

Ship it

Five on it: You can never have too many chips in Las Vegas, and this week AWS unveiled the next-generation of its Graviton family of Arm-based CPUs. Graviton5 promises "up to 25% better compute performance than the previous generation while maintaining leading energy efficiency," AWS said in a blog post.

Over the last two years, more than half of the new CPU workloads started on AWS are using Graviton, said Nafea Bshara, vice president at AWS and co-founder of AWS chip division Annapurna Labs, in an interview at re:Invent with Runtime. "There's nothing that gains customer trust to move to your platform and commit to your platform than having credible execution," he said, noting that AWS has rolled out new Graviton designs at a steady clip since the first one was introduced in 2018.

Servers for serverless: The definition of "serverless" computing has been stretched in so many directions since AWS Lambda debuted in 2014 as a new way of building applications based around functions, which ran more or less on their own without the user having to spec out the cloud instances needed for the job. Developers using Lambda like the simplicity of that approach, but long-running Lambda workloads can get real expensive, real fast, and some applications need specialized hardware.

This week AWS introduced Lambda Managed Services, which allows Lambda users to choose when, where, and how they'd like to run their applications on the EC2 compute service without having to actually provision those instances the traditional way. "There will be some large customers for whom managed instances will be a substantial benefit; the new enhancement reduces but does not eliminate the price premium for using the serverless model," according to Devclass.

Forging ahead: Amazon's Nova family of models hasn't exactly set the AI world on fire, but it thinks companies might be more willing to give those models a shot if they can train them against their own data. This week AWS introduced Nova Forge, which allows companies to build their own models based around Nova without having to train a big large-language model of their own from scratch.

"Nova Forge customers can start their development from early model checkpoints, blend their datasets with Amazon Nova-curated training data, and host their custom models securely on AWS," the company said in a blog post. CNBC reported that it will cost $100,000 to build a custom model through Nova Forge and customers will have to deploy that model on Amazon Bedrock, but that's a fraction of the cost it would take to train a leading-edge AI model on their own.

Marketplace of ideas: Vercel's approach to building web applications has drawn a lot of attention over the past couple of years, and most of its services run on AWS. This week Vercel announced that customers will be able to provision several AWS databases for their apps directly through its marketplace without having to set that up directly through the AWS accounts.

As of next week, Vercel developers will be able to add Aurora PostgreSQL, Amazon DynamoDB, and Aurora DSQL to their apps. "Vercel handles credentials and environment variables automatically, so you select your database, connect it to your project, and start writing queries," the company said in a blog post.

More containment, please: MCP servers have been hailed as an important step toward making AI agents easier to build, but security concerns have confined them to mostly internal use, according to research from The Pragmatic Engineer. This week at re:Invent Netskope announced that it added new MCP security features to its Netskope One platform, which companies use to manage cloud security.

"With the new platform capabilities, Netskope One can protect MCP-enabled AI interactions by providing full visibility into MCP tool use, enforcing least-privilege access, securing sensitive data, and ensuring compliance," the company said in a press release. Just as the rise of cloud computing and containers made companies realize that older security techniques didn't quite work in the new world, agents and standards like MCP will likely require new tools to operate safely.

If you value independent enterprise tech journalism from big events like AWS re:Invent, please consider becoming a monthly supporter of Runtime. For $10 a month, you'll help us continue our mission to bring reliable and actionable coverage of this vital sector of the economy and gain access to supporter-only features currently in the works, such as an exclusive discussion and events forum.

Stat of the week

The scramble to find GPUs and sources of energy to run next-generation AI data centers gets most of the attention these days, but those data centers wouldn't be nearly as effective without fast and reliable networking connections linking those buildings across the world. According to CEO Matt Garman, AWS increased its networking capacity by 50% in the last 12 months to encompass 9 million kilometers of cabling around the planet, which is "enough optical cabling to reach from the earth to the moon and back over 11 times," he said in his Tuesday keynote.

Quote of the week

"It’s time for those different voices of AWS to be in front of you. This is my decision.” — Amazon CTO Werner Vogels, riding off into the sunset after delivering his 14th and last keynote address at AWS re:Invent 2025.

The Runtime roundup

Cloudflare apologized for a widespread but relatively short outage Friday morning that occurred after attempting to deal with the nasty React Server Components vulnerability disclosed earlier this week.

Salesforce CEO Marc Benioff suggested the company would consider changing its name to "Agentforce," which … sure, fine, whatever.

Thanks for reading — see you Tuesday!