One-Line Summary

Capital and compute deals tighten the AI stack: an AMD-first lab raises big money, OpenAI dangles $2M tokens for YC equity, and agentic tools land in mainstream design and desktop workflows.

Big Tech

Musk–OpenAI verdict spotlights antitrust questions in AI

A jury rules for OpenAI on statute-of-limitations grounds in Elon Musk’s case, and a legal analysis argues the fight previews how antitrust will shape AI by examining dependence on big tech capital and clouds. ¹

OpenAI leans on Microsoft, which has invested roughly $13 billion and delivers models via Azure; Amazon has committed up to $8 billion to Anthropic and is its primary cloud partner, while Google has made multibillion-dollar investments and support. ¹

Regulators signal concern: the Federal Trade Commission has opened inquiries into AI partnerships and flagged risks from equity stakes, exclusive cloud terms, and governance rights, while moves like Microsoft’s hiring of Inflection AI’s leadership and deals involving Adept and Character.AI stoke “killer acquisition” debates. ¹

Industry & Biz

Zyphra raises $500M at $5B+ to back AMD-first AI

Zyphra is a San Francisco AI lab that says it trains and runs its open-weight models entirely on AMD chips, and it is raising $500 million in a Series B that includes AMD’s participation at a valuation of at least $5 billion, per Forbes’ sources. ²

Beyond releasing about a dozen open-weight models — from brainwave-decoding research to Zaya reasoning models — Zyphra operates Zyphra Cloud, a “neocloud” positioned to host advanced open-weight models on AMD hardware; it has told investors it will provide compute to labs constrained for capacity, naming Anthropic, Meta, and OpenAI. ²

Investor interest is shifting toward U.S.-based alternatives to DeepSeek, with rival Reflection AI reportedly discussing a $25 billion valuation. Compute remains the bottleneck across chips, power, data center space, and supply chains; labs are diversifying, with OpenAI exploring custom silicon and AMD, and Anthropic using Amazon Trainium and Google TPUs while planning to work with AMD. ²

If AMD-first training and inference prove cost and supply-chain advantages at scale, more teams could access powerful models without queueing for scarce Nvidia processors — a practical lever on budgets and roadmaps. ²

OpenAI offers $2M in tokens to every YC startup for equity

OpenAI CEO Sam Altman offers $2 million worth of OpenAI tokens to each of roughly 169 startups in Y Combinator’s current class in exchange for equity via an uncapped SAFE that converts at the next priced round, typically Series A. ³

Supporters see a way to cover steep AI infrastructure bills without cash; critics warn about platform risk and uncertain dilution, with investor Jason Calacanis cautioning that big platforms might study and copy startup ideas, while others note falling inference costs could make OpenAI’s outlay cheap relative to the equity it gets. ³

The trade-off is concrete: YC already takes 7% for a $500,000 investment, so founders must estimate token use and weigh equity cost at conversion against the benefit of building on OpenAI instead of competitors like Anthropic’s Claude Code. ³

New Tools

IrisGo: a desktop agent that learns and automates your workflows

IrisGo is a PC companion that watches you do a task once and then automates it, aiming to anticipate routine actions; it raised a $2.8 million seed led by Andrew Ng’s AI Fund and launched beta apps for macOS and Windows. ⁴

In demos it repeats multi-step tasks like online ordering, ships with skills for email drafting, invoices, reports, and summarization, includes a coding assistant, and processes much of your data on-device with encrypted cloud processing only when authorized; the startup is backed by Nvidia and Google and has a preinstall deal with Acer. ⁴

Figma adds an AI agent to the collaborative canvas

Figma introduces an AI assistant inside its design canvas so you can prompt it to generate new designs, edit existing ones, and automate iterations — even running multiple agents simultaneously. ⁵

Figma says the agent understands design context via fine-tuned models, launches first in Figma Design, and aims to bring design and code closer; it reports Q1 2026 revenue of $333.4 million, up 46% year over year. ⁵

What This Means for You

AMD-first infrastructure and “neoclouds” point to more options to run and host advanced models as labs and vendors diversify beyond a single chip supplier — a potential relief valve for cost and capacity constraints on AI features. ²

Credits-for-equity offers like OpenAI’s YC deal show how infrastructure providers finance early adoption; if you’re building on a platform’s tokens, map actual usage and the equity you’d give up at conversion against cash alternatives. ³

Agentic tools are moving into everyday work: designers can try Figma’s on-canvas prompts for quick iterations, while operations and support teams can pilot IrisGo on repetitive desktop tasks with on-device privacy defaults. ⁵

Regulators are scrutinizing big tech entanglements in AI; procurement and legal teams should watch contract terms that entrench one cloud or model provider and build portability into your stack. ¹

Action Items

1. Test Figma’s AI assistant on a live file: If available in your workspace, ask it to generate three design variations and auto-iterate a component to gauge quality and speed.
2. Pilot IrisGo on two repetitive chores: Install the beta on macOS or Windows and record tasks like invoice downloads or weekly report assembly to measure accuracy and time saved.
3. Request a one-page compute comparison: Ask your infra or data partner to estimate costs and latency of AMD-powered options versus your current setup for one representative workload.
4. Price the equity cost of tokens: If you’re in an accelerator or early-stage, model token consumption and estimate dilution at your expected Series A valuation before accepting credits-for-equity.
5. Add portability clauses to AI vendor contracts: Ensure your next AI agreement includes data export, model-switch rights, and termination assistance to avoid lock-in.