OpenAI secures $110 billion as Google rolls out Nano Banana 2

OpenAI announced $110 billion in new investment commitments on February 27, 2026. Amazon, NVIDIA, and SoftBank are backing the round, which values OpenAI at $730 billion before the new investment.¹ Amazon and OpenAI are also expanding their multiyear AWS agreement by $100 billion over eight years. The agreement includes plans to use Amazon’s Trainium chips.²

AWS will be the exclusive third-party cloud distribution provider for OpenAI Frontier, a platform for building and managing enterprise AI agents.² OpenAI also plans 3 gigawatts of inference capacity and 2 gigawatts of training capacity on NVIDIA Vera Rubin systems.¹ The funding round remains open to additional investors.¹ Amazon’s $50 billion commitment will not arrive all at once: $15 billion is due initially, with another $35 billion contingent on specified conditions or an OpenAI IPO.³

Google announced Nano Banana 2 on February 26. The image generation and editing model has the technical name Gemini 3.1 Flash Image. It is rolling out in the Gemini app, Google Search, Google Ads, and Flow. Developers can access it in preview through the Gemini API in AI Studio and Vertex AI.¹²

Google positions the model as combining Pro-level capabilities with Flash-like speed. It also highlights improved instruction following and the use of current web information in image creation.¹³ In the Gemini app, Nano Banana 2 replaces the previous Pro model across Fast, Thinking, and Pro modes. API access gives companies another option for image workflows, though the available source text does not specify per-image prices.¹²

Perplexity introduced Computer on February 25. The service is designed to handle projects across persistent workflows, from research and design to coding and deployment. It launched on the web for Perplexity Max subscribers. Perplexity describes it as a system that assigns subtasks to specialized agents and coordinates 19 models. Execution takes place in sandboxed environments.¹²³

The product extends Perplexity beyond search into multistep work. Rather than relying on one chatbot, Computer is intended to organize tasks across models and bring their outputs together in a project.¹⁴ For companies, cost and access controls matter: Perplexity offers usage-based pricing, spending caps, and optional model selection for sub-agents.⁵

Samsung announced Perplexity as a preloaded AI assistant for the Galaxy S26 series on February 26.¹² The integration goes beyond a standalone app: Perplexity says it can access Samsung Notes, Calendar, Gallery, Clock, and Reminders to read and update information in those apps.¹ Perplexity APIs also power features in Bixby and Samsung Internet.³ The service is therefore built into phone functions rather than offered only as another search app.¹³

Samsung is continuing to offer multiple assistants on its new phones, naming Bixby and Gemini alongside Perplexity.² That gives users another option for requests and workflows spanning supported apps without removing Gemini from the lineup.¹² Using Perplexity requires a network connection and a Perplexity account.⁴

OpenAI and Figma introduced a connection between Codex and Figma on February 26, 2026. Users can generate Figma designs from their work in Codex and, in the other direction, use Figma designs as a basis for implementation in code. Figma’s MCP server connects the tools. Both companies presented the new workflow as available starting on the announcement date.¹²

The integration aims to make it easier to move between design and development, rather than treating finished designs only as handoffs to programmers.¹² Figma also connects its design tools to other AI coding assistants.³ Teams will still need to check whether generated designs and code meet their requirements. The announcements do not specify pricing or detailed access restrictions.¹²

Since late February, Claude Code has been able to save useful information from project work automatically and reuse it in later sessions. That can include project context, architecture notes, and solutions to recurring bugs. The feature is enabled by default in Claude Code. It writes notes to local Markdown files, with MEMORY.md serving as the automatically maintained memory file.¹²

This memory has a different role from CLAUDE.md, which remains the place for instructions maintained by users. The automatic notes are specific to a project; they are not a shared knowledge base for a team. For longer-running work, that means Claude Code can draw on earlier findings without users restating the same context every session. Teams should still put rules they want to maintain deliberately in their own instruction files.²