PLUS: NVIDIA's first CPU for agents and Hershey's $2B AI bet
Happy reading
Research lab Odyssey just revealed a major step toward shared digital worlds. Its new Agora-1 system acts as a multi-agent world model, allowing both humans and AIs to interact within a single, consistent simulation.
The technology moves AI training from isolated, single-agent sandboxes to dynamic, shared digital worlds. Could these complex environments, where AIs can learn from each other through cooperation and competition, be the key to accelerating the development of more advanced social behaviors?
In today’s Next in AI:
Odyssey’s multi-agent world model
NVIDIA’s first CPU for agentic AI
Hershey's $2B agentic AI strategy
Nebius's $44B rise in AI cloud
The Multi-Agent World Model
Next in AI: AI research lab Odyssey just unveiled Agora-1, a multi-agent world model that lets multiple humans and AIs interact in a shared simulation. The system acts as a fully learned game engine, demonstrated with a stunning four-player deathmatch in the classic game GoldenEye.
Explained:
Agora-1's design cleverly separates world simulation from visual rendering, allowing it to maintain a consistent shared state for all participants and generate unique viewpoints for each player in real-time.
Unlike previous approaches that struggled when agents lost sight of each other, Agora-1 ensures the world remains coherent, solving a major hurdle for multi-agent systems and enabling more complex interactions.
The technology extends far beyond gaming, creating new possibilities for collaborative robotics and providing rich environments for agents to learn through open-ended competition and cooperation.
Why It Matters: This breakthrough marks a shift from single-agent AI sandboxes to dynamic, shared digital worlds. These environments will serve as powerful training grounds, allowing AIs to co-evolve and learn complex social behaviors at an accelerated pace.
NVIDIA's New Agent Brain
Next in AI: NVIDIA just shipped Vera, its first CPU purpose-built for agentic AI. The initial systems are now in the hands of leading labs like OpenAI, Anthropic, and Oracle Cloud Infrastructure, signaling a major hardware shift for autonomous agents.
Explained:
AI agents rely heavily on CPUs for tasks like using tools and retrieving data, creating a need for hardware designed for these concurrent workloads.
Vera packs 88 custom Olympus cores and 1.2 TB/s of memory bandwidth, which helps agent sandboxes run 50% faster and enterprise data queries execute up to 3x faster than traditional CPUs.
Oracle Cloud Infrastructure is the first cloud provider to go all-in, planning to deploy hundreds of thousands of Vera CPUs starting in 2026 to power enterprise-scale agentic AI.
Why It Matters: This move shows that the hardware powering AI is evolving beyond GPUs to address the specific, CPU-intensive demands of autonomous agents. Purpose-built chips like Vera could significantly accelerate the development and deployment of more capable agents across the enterprise.
Hershey's $2B AI Bet
Next in AI: The Hershey Company is deploying agentic AI to analyze its massive $2 billion marketing budget. The move aims to transform its slow, historical analysis into a near real-time system for making smarter spending decisions.
Explained:
Hershey is tackling a classic analytics problem: its old method delivered 2024 performance data midway through 2025, which is too late for 2026 planning.
The new system uses a multi-agent AI from Mutinex, powered by models like Claude and Gemini, while the platform Tracer first cleans and standardizes all the marketing data.
The company expects this shift to boost media-driven revenue by 4-5%, a significant application of AI for financial optimization that the tech community is discussing.
Why It Matters:
This move signals a significant shift for AI in the enterprise, moving beyond creative tasks and into core financial strategy. It showcases how companies can apply AI to directly influence multi-billion-dollar investment decisions.
The $44B AI Cloud Contender
Next in AI: Nebius, a company spun out of Russia's Yandex, has emerged as a heavyweight in the AI cloud race, securing massive deals with Meta and Microsoft worth over $44 billion combined.
Explained:
The company's Q1 revenue skyrocketed to $399 million, a massive jump from $50.9 million just a year ago, fueled by intense AI cloud demand.
Nvidia recently injected $2 billion into Nebius, a strategic move that provides closer access to the latest AI hardware and solidifies its position against competitors like CoreWeave.
Beyond financial deals, Nebius is aggressively expanding its physical footprint, breaking ground on a new 400-acre data center campus in Missouri to meet growing capacity needs in the U.S.
Why It Matters: Nebius's rapid rise creates a powerful new option for AI companies, challenging the established cloud hierarchy and potentially driving down compute costs. This intense competition in AI infrastructure is critical for fueling the next wave of model development and deployment across the industry.
AI Pulse
Anthropic will brief a global finance watchdog on its Mythos model, whose powerful cyber flaw-finding abilities have raised concerns about AI's potential impact on financial stability.
Linus Torvalds warned that the Linux security mailing list has become “almost entirely unmanageable” due to a flood of duplicate, low-quality bug reports generated by AI tools.
Albertsons deployed an in-house AI tool that uses computer vision to inspect produce like strawberries and grapes, aiming to improve quality control and consistency across its 2,000+ stores.
Cloudflare shared its findings from testing Anthropic’s Mythos model under Project Glasswing, detailing how it created a sophisticated "harness" to manage the AI for vulnerability discovery and overcome the signal-to-noise problem in its own code.