Vibe Code Brief, October 28: Anthropic Files for a $2 Trillion IPO

Four headlines today that each move a different piece of the board: public markets, enterprise distribution, engine licensing, and how frontier AI research gets released. Here is what happened and what it means if you design or build things for a living.

Anthropic filed for an IPO, and the numbers are staggering

Anthropic, the company behind Claude, filed its public S-1 on Monday, October 26, according to NBC News. The filing shows an $80 billion net loss as the company spends unprecedented amounts on compute, and Fortune reports plans to spend up to $567 billion on infrastructure, including its own data centers. The offering, expected mid-November, could value Anthropic above $2 trillion and rival the largest tech debuts in history. The filing itself even warns that AI could "pose catastrophic or existential risks to humanity."

Image: Anthropic, via Anthropic

Why it matters to designers and builders: the model provider a huge chunk of you build on every day is about to answer to public markets. Quarterly earnings pressure will shape Claude's pricing, feature velocity and enterprise focus, which eventually shapes your stack costs and your roadmap. The $80 billion loss also says the "AI margins are easy" era is not arriving on schedule.

What to do: keep building, but stop marrying one model. Switching costs that feel like a weekend today may feel like a quarter by next year.

Oracle is putting ChatGPT and Codex inside its apps

At Oracle AI World in Las Vegas on Tuesday, Oracle announced it is embedding OpenAI's ChatGPT and Codex into its application stack to make its databases more actionable. Mahesh Thiagarajan, Oracle's president of Cloud Infrastructure, put the rollout plainly: ChatGPT "allows anybody to build an application, deploy it, manage it." Oracle Cloud Infrastructure is also adding Google TPUs (a first for Google's custom chips on another hyperscaler), OpenAI's Jalapeño chip, and Cerebras hardware, with Sam Altman joining Oracle co-CEO Clay Magouyrk on stage.

Image: Oracle, via Oracle Newsroom

Why it matters: vibe coding tooling is landing inside the boring, critical software enterprises already run. The same IT departments that blocked browser extensions will soon have ChatGPT inside their finance and healthcare workloads. Distribution, not capability, is about to be the story in enterprise.

What to do: if your client list includes big companies, start asking what is in their Oracle stack. There is a wave of design work coming for AI-flavored data apps nobody has designed yet.

Epic is dropping its AI training ban in the Unreal Engine EULA

Epic Games CEO Tim Sweeney confirmed on X that Epic is working to remove the EULA restriction that currently blocks AI training on outputs of Unreal Engine and MetaHuman, while developers still get to decide the terms for their own games. Times of AI reports that Epic engineers already use Claude Code and Codex across Fortnite, UEFN and Unreal Engine 6, including a DirectX 12 rendering layer rewrite that went from an estimated three to six weeks of engineering time to about a week.

Image: Unreal Engine via Wikimedia Commons, public domain

Why it matters to 3D and game designers: your engine maker is now publicly all in on AI-assisted development and is rewriting the licensing rules around training data to match. The tools you use daily are being reshaped by this posture, and the artwork pipeline you rely on is part of the conversation whether you opted in or not.

What to do: read the EULA change when it lands before you assume anything about what can or cannot be trained on your outputs.

OpenAI dropped three major proofs, and the transparency fight is the real show

A day after filing its S-1 competitor move, OpenAI was back in the news for research: Scientific American reports the company dropped proofs touching the Kakeya conjecture in every dimension, the existence half of the Yang-Mills problem, and a refutation of the Exponential Time Hypothesis, one of the bedrock assumptions in computer science. Two of the results landed next to closely related work by an Anthropic-assisted team, and OpenAI's own expert advisory panel (AGMAI) is publicly criticizing the release format: proprietary model, no public prompts or chain of thought, papers still missing.

Image: OpenAI via Wikimedia Commons, public domain

Why it matters: the capability frontier is one headline, but the release-transparency fight is the better story for our community. When the biggest AI labs start shipping work without process, the "show your work" standard that designers defend in critique and code review is quietly being litigated at the highest level. The amateurs of process are now billion-dollar labs.

What to do: watch it, form your own view on how AI research should be published, and steal the opposite instinct for your own work: document your process.

Our take

The pattern across all four stories is that AI capability is being institutionalized: priced by public markets, distributed by Oracle, licensed by Epic, published (opaquely) by OpenAI. The frontier labs are fighting over compute and IPOs, which mostly means the tools you use get cheaper, faster and more tangled in enterprise paperwork. That is good news for people who ship. Your taste, your judgment and your process transparency remain the only moats nobody is raising $567 billion to copy.

Building something from today's news? Base44 is the fastest way I know to prototype it. Want deliberate practice instead? Today's prompt is waiting at app.newform.community/daily-vibe-code.

Related on NewForm

• A Vibe-Coded Site Got Hacked 6 Days After Launch. It Was Not the AI's Fault
• OpenAI Just Put Up to $500 a Month of API Credits Inside Your ChatGPT Subscription
• AI Can Clone Your App in a Weekend. So What Is Your Moat Now?