Today on Product Saturday: OpenAI unveils GPT-4.5, creating huge expectations for GPT-5, AWS joins its cloud competitors with a custom quantum chip, and the quote of the week.
Why Salesforce is going to have to wait a lot longer than it expected for the agentic AI revolution, Nvidia, meanwhile, still can't make enough GPUs, and the latest enterprise moves.
Today: Pinecone's second-generation serverless infrastructure for its managed vector database gets an upgrade, Microsoft's data-center buildout hits a snag, and the latest funding rounds in enterprise tech.
Today on Product Saturday: OpenAI unveils GPT-4.5, creating huge expectations for GPT-5, AWS joins its cloud competitors with a custom quantum chip, and the quote of the week.
Welcome to Runtime! Today on Product Saturday: OpenAI unveils GPT-4.5, creating huge expectations for GPT-5, AWS joins its cloud competitors with a custom quantum chip, and the quote of the week.
(Was this email forwarded to you? Sign up here to get Runtime each week.)
Orion's belt: As promised earlier this month, on Thursday OpenAI rolled out GPT-4.5, which the company said would be its last model that doesn't use the chain-of-thought process, commonly described as "reasoning." OpenAI had once hoped that its Orion release would be a bigger leap over the groundbreaking GPT-4 release in March 2023, but it acknowledged that the new model doesn't perform as well as several of its releases over the last year.
Ars Technica pretty much nailed it: GPT-4.5 "is big, expensive, and slow, providing marginally better performance than GPT-4o at 30x the cost for input and 15x the cost for output." And it doesn't even sound like it will be around that long; "we’re evaluating whether to continue serving it in the API long-term as we balance supporting current capabilities with building future models," OpenAI said in a blog post.
Land of lakes: Observability software is a powerful tool for diagnosing problems with application infrastructure, but sorting through the reams of data that it collects is not fun. Cribl launched a new product Wednesday called Cribl Lakehouse that was designed specifically to store and manage telemetry data.
"Cribl Lakehouse is built to scale effortlessly, provide instant insights from any dataset, and empower teams to modernize telemetry data management while reducing costs and complexity," the company said in a press release. Data lakes and lakehouses have been growing for several years as companies look for open and cheaper ways to store and refine data for AI applications, and anything that makes it easier to analyze telemetry data should be welcome.
Still costs a lot: AWS has been a little more discreet with respect to its quantum computing research compared to its peers, but on Thursday it became the last of the Big Three cloud infrastructure providers to introduce a quantum computing chip in recent months. On Thursday It unveiled Ocelot, a "prototype quantum computing chip" based around the "cat qubit" that was "designed to test the effectiveness of AWS’s quantum error correction architecture," the company said in a press release.
One of the many, many roadblocks along the way to practical quantum computers is the fact that it's extremely hard to stabilize qubits, and unstable qubits make a lot of errors. AWS hopes that "quantum chips built according to the Ocelot architecture could cost as little as one-fifth of current approaches, due to the drastically reduced number of resources required for error correction."
Network the premises: Right now the cloud infrastructure providers and GPU cloud upstarts hold the keys to AI computing resources, but as generative AI matures some of the bigger customers might decide they can save money doing it themselves. Cisco and Nvidia unveiled a partnership this week that might make on-premises AI a little easier to do.
The companies have co-designed a new networking device based on custom chips from both companies and "by enabling interoperability between both companies’ networking architectures, the two companies are prioritizing customers’ needs for simplified, full-stack solutions," Cisco said in a press release. According to The Next Platform, the partnership "is about creating a Cisco-flavored Spectrum-X offering, which both Nvidia and Cisco both believe is necessary to make a more familiar and palatable AI cluster offering to customers used to having Cisco as their networking vendor."
Buzz my agent: It was PagerDuty's turn this week to unveil an agentic AI strategy for its incident management and response software, which helps companies prevent outages and recovery as quickly as possible from the ones that do happen. "PagerDuty is charting a path towards a future where autonomous AI agents will be able to resolve well-understood incidents even if they require multiple steps across disparate tools," the company said in a press release, and what could possibly go wrong with that plan.
The "Spring 2025" release of PagerDuty will come with three agents; a site reliability engineer, an operations analyst, and a scheduler. Those agents "will work with responders and autonomously resolve issues, empowering organizations to efficiently redeploy their resources towards higher-value work," the company said.
On top of everything else CISOs have to deal with to keep their own companies secure, problems caused by their software supply chains — or "third-party risk" — are surging. According to new data from cyber insurance company Resilience, third parties were involved in 23% of claims awarded in 2024, up from 0% in 2023.
"People used to talk about web scale as being big. But the ratio between how much stuff you put on the web versus how many emails you've got, how many text messages you wrote, how many pictures you have on your phone, how many meetings you've taken and how many sentences you've spoken and heard, how many documents you signed and saved … all of that data sits somewhere, and now it's actionable." — Pinecone founder and CEO Edo Liberty, outlining the bull case for vector databases.
Microsoft announced Friday that it would finally retire Skype, the once-groundbreaking communications software platform it bought in 2011 for $8.5 billion and started phasing out years ago in favor of Microsoft Teams.
Intel delayed plans to open a long-planned chip factory in Ohio until at least 2030, and the odds of that facility ever actually opening will hinge on whoever winds up getting its manufacturing business in what appears to be a looming firesale.
Thanks for reading — see you Tuesday!
Why Salesforce is going to have to wait a lot longer than it expected for the agentic AI revolution, Nvidia, meanwhile, still can't make enough GPUs, and the latest enterprise moves.
Today: Pinecone's second-generation serverless infrastructure for its managed vector database gets an upgrade, Microsoft's data-center buildout hits a snag, and the latest funding rounds in enterprise tech.
Today on Product Saturday: ST Microelectronics and AWS collaborated on a new data center chip, VAST Data's core product now supports block storage and Kafka streaming, and the quote of the week.
Today: the quantum computing hype train leaves the station once again, this time with Microsoft in the driver's seat, multiple reports outline DOGE's control over computing infrastructure formerly run by official government agencies, and the latest enterprise moves.
