# AI infrastructure stocks — X 热门讨论 (2026-09-16 00:16 UTC)

## @CryptoBusy (CryptoBusy) · 09-15 22:38 · ♥50 ↻10 💬10 This is the RWA narrative I want to see.

Not just stocks moving onchain, but real cash flows from AI compute, and digital infrastructure becoming accessible to both institutions and retail.

The USD.infra Vault launches September 22!

Definitely watching @dawninternet closely. > 引用 @dawninternet: The USD.infra Vault is coming.

DAWN is bringing the cash flows of digital infrastructure on-chain as yield for institutions and retail.

We built the network. Now we’re launching the RWA protocol for AI compute, connectivity, and edge infrastructure.

Sept. 22, 2 PM UTC. 🔒 https://t.co/FeiPR2gJPU https://x.com/CryptoBusy/status/2099991289320308822

## @AIonBase_ (AI on Base) · 09-15 20:30 · ♥41 ↻6 💬11 Top AI Projects on @base

The AI stack on base covers private inference, agent memory, model access, paid agent work, forecasting, robotics, and autonomous worlds.

Here are the top projects:

■ @basemateagent - B20 launch agent for creating tokens paired with tokenized stocks through iMessage.

■ @AskSurplus - Inference marketplace for discounted model access and enterprise AI capacity.

■ @KorProtocol - KORUS lets agents create and license music with Base MCP and x402 payments.

■ @OpenGradient - Private inference network with 651K+ inferences and 4.81B+ processed tokens.

■ @usedotai - Private AI tools for browsing, coding, images, video, and MCP workflows.

■ @gitlawb - Open-source agent tooling, including Zero, a Go-based agent harness.

■ @officialbunnyos - Persistent game world where AI agents compete, trade, coordinate, and create behavior data.

■ @reppo - Evaluation markets for AI agents, datasets, and robotics systems.

■ @QuotientHQ - AI forecasting tools for prediction markets, trading, and live event probabilities.

■ @BlockRunAI - USDC gateway for agents to access AI models, data, compute, voice, and blockchain RPCs.

■ @sibylcap - Agent memory infrastructure with staking, research, and builder programs.

■ @aeonframework - Background agents for security scanning, code fixes, monitoring, and long-running tasks.

■ @myrad_hq - Personal AI terminal for recommendations based on individual taste and preferences.

Which project should get the next spotlight? https://x.com/AIonBase_/status/2099958936237416475

## @MelvinInvests (Melvin) · 09-15 18:56 · ♥30 ↻6 💬6 OpenAI just ended Anthropic’s two year reign on OpenRouter and that will be a major demand signal for Oracle.

The chart measures wallet share, which means the percentage of OpenRouter spending directed toward each company’s models rather than the percentage of total users.

OpenAI’s share rose from approximately 20% at the beginning of 2026 to more than 50% during the week of September 7, marking its first weekly lead over Anthropic since February 2024.

The shift appears to be driven by OpenAI’s latest model family rather than renewed demand for its older models, with Astra alone reaching a 19% share while several GPT 5.6 variants added further spending.

Also, it's important to note that OpenRouter represents only one distribution platform, so this chart does not measure total enterprise demand, direct API usage or ChatGPT subscriptions.

The broader implication is that OpenAI’s model momentum is recovering at the same time that inference demand is becoming an increasingly important part of the AI infrastructure cycle.

Training a frontier model requires an enormous amount of compute during a concentrated period, while serving that model to millions of users creates recurring demand every time someone sends a prompt, generates code or runs an AI agent.

If OpenAI continues taking wallet share, it will need substantially more inference capacity in addition to the infrastructure required to train its next generation of models.

This is where the chart becomes bullish for Oracle.

Oracle and OpenAI are developing an additional 4.5 gigawatts of capacity, bringing their total planned infrastructure under development to more than 5 gigawatts with the ability to support over two million chips.

The wider partnership between the two companies exceeds $300 billion over five years, while the operating Abilene campus is already running early OpenAI training and inference workloads on Oracle Cloud Infrastructure.

Oracle has also started distributing OpenAI models and Codex through its marketplace, allowing OCI customers to purchase access through their existing cloud contracts and Oracle Universal Credits.

This gives Oracle exposure to OpenAI from two directions, because Oracle can earn infrastructure revenue from powering the models while also helping distribute those models to enterprise customers.

Bullish on Oracle because more OpenAI usage means more inference, more infrastructure and ultimately more demand flowing through Oracle.

Follow @MelvinInvests for more AI infrastructure insights and if you want to see exactly what stocks I'm buying to play this buildout, you can join Milk Road Pro for just $1 using the link below.

https://t.co/Hkywss4Ugh https://x.com/MelvinInvests/status/2099935430846431489

## @ashbay22 (Ashbay) · 09-15 17:38 · ♥30 ↻7 💬0 What Prism APIs Unlock for Developers https://x.com/ashbay22/status/2099915776522346707