OpenAI wants those enterprise tech dollars

Welcome to Runtime! Today: there were already 10,000 vendors trying to help companies build AI agents and enterprise-grade coding assistants, and now there are 10,001, Oracle's cloud AI growth has come at a price, and the latest funding rounds in enterprise tech.

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

Time to build building machines

While it doesn't appear that 2025 is going to be the year enterprises truly embrace agentic AI, as was promised last year, it certainly has been an active year for companies that are building tools for building agents. For all its wackiness, OpenAI remains at the center of the generative AI discussion, and it would like to play a bigger role in the enterprise.

Yesterday at its annual Demo Day, OpenAI unveiled a new agent-building tool called AgentKit as well as a version of its Codex coding assistant designed to give IT a little more control. “You should expect a huge focus from us on really leaning into enterprise,” CEO Sam Altman said in a press conference after the presentation, according to Reuters.

• AgentKit comes with several components, but the big one is Agent Builder, a no-code development tool that "provides a visual canvas for composing logic with drag-and-drop nodes, connecting tools, and configuring custom guardrails," the company said in a blog post.
• OpenAI also released a Connector Registry as part of AgentKit that it said makes it easier for developers to connect their agents to popular services such as Dropbox, Google Drive, and Microsoft Teams as well as custom MCP servers.
• Generative AI apps don't really work unless developers establish strong connections between the chat interface and those data sources, as Mastercard's George Maddaloni relayed last month.
• However, as The New Stack pointed out, to consider AgentKit "a fully enterprise-ready system (something more akin to OutSystems or Mendix) [it] would also have to include in-depth data governance, auditing and security tools."

Companies that want to build AI agents certainly have their choice of tools from basically every enterprise software vendor with a pulse, but OpenAI does not have the enterprise chops that those vendors and rival Anthropic have developed over time. However, it's clear that the company has a strong presence among fast-growing startups and younger developers that may one day be CTOs.

• After it dropped GPT-5 in early August, usage of its Codex coding tool — which is now generally available — grew 10x, the company said in a blog post.
• During Monday's event OpenAI also released a Slack integration for Codex, a software-development kit for using the GPT-5-Codex model in other software pipeline tools, and enterprise-oriented admin features that provide additional security controls and analytics tools around its flagship coding assistant.
• And it gave "Tokens of Appreciation" to dozens of developers who used up to 1 trillion tokens through its API, which is an older enterprise tech marketing gimmick, but it checks out.

The $500 billion question is whether OpenAI can build a real, meaningful enterprise business over the next few years, which might be its only hope of generating enough revenue to cover its immense costs. It certainly is trying to build the infrastructure to service that kind of demand, adding a huge deal with AMD on Monday to a hardware spending spree "that altogether could easily top $1 trillion," according to Bloomberg.

• Building a rapport with enterprise customers is not the kind of thing you can do with flashy demos alone; it takes years of trust and a commitment to solving the edge cases to bet on the kind of platform vendor that OpenAI clearly wants to become.
• It might be useful to consider the history of AWS when handicapping OpenAI's chances in the enterprise; very few enterprises were willing to bet on AWS during the first three or four years it offered cloud services, but startups and developers working on experimental projects inside big companies loved those new tools and eventually won over the bosses.
• But AWS had one advantage OpenAI lacks: "We faced no like-minded competition for seven years. I think the big established enterprise software companies did not see Amazon as a credible enterprise software company, so we had this incredible runway," said founder and chairman Jeff Bezos at a conference years ago, according to Fortune.
• This time around, OpenAI is competing with everyone, even its closest partner.

Any business that survived the digital revolution is very familiar with how quickly networking requirements can shift. But even with the crash course most got as the internet, mobile computing, and the cloud emerged, few are prepared for the speed with which emerging technologies today are challenging existing environments.

Check out our latest post, "Why future-proofing the workplace for AI starts at the network," sponsored by Cisco.

Margin call

Oracle's stock fell more than 7% in midday trading Tuesday after The Information reported that the gross margins on its cloud GPU infrastructure service are extremely thin, which suggests the company has a long road ahead of it to pay back the massive amounts of debt it is taking out to fund its expansion. The report appeared to be a reality check for investors that poured money into the company's stock after it announced a huge deal with OpenAI last month.

During the first quarter of its 2026 fiscal year, Oracle's gross margins for its Nvidia GPU business were just 14%, which according to the report means it actually lost about $100 million selling access to Nvidia's new Blackwell chips, once you factor in labor and power costs. Those costs hit Oracle harder than other hyperscalers that own their own data centers because Oracle prefers to lease.

It's also practically giving away cloud GPU instances according to Guggenheim Securities’ John DiFucci, who told The Information "they are pricing their offering very aggressively." Oracle was also known for aggressive discounts when it was trying to win cloud infrastructure business away from the Big Three during the pre-ChatGPT era, but Nvidia's GPUs cost a lot more than Intel's CPUs.

Enterprise funding

Nscale raised $433 million in Pre-Series C SAFE funding just one week after it hauled in a $1.1 billion Series B round to fuel its neocloud ambitions.

Temporal scored $105 million in secondary funding to build out its software development platform, which is based around workflows.

Supabase landed $100 million in Series E funding, which values the managed Postgres back-end company at $5 billion.

DualEntry raised $90 million in Series A funding as it attempts to take on the massive ERP market with a "native AI" service.

Phaidra scored $50 million in Series B funding for its data-center control systems, which use AI agents to manage power and cooling resources.

Dash0 landed $35 million in Series A funding to build out an observability service based around AI agents.

The Runtime roundup

Shares of Dell rose 3.5% Tuesday after the company raised its guidance for the year, saying "customers are hungry for AI and the compute, storage and networking we provide to deploy intelligence at scale."

Deloitte signed a huge deal to use Anthropic's AI across 470,000 employees one week after it agreed to reimburse the government of Australia for $290,000 after it turned in a report with "multiple citations to non-existent academic reports" that appeared to be AI-generated.

Thanks for reading — see you Thursday!