Thursday, May 7, 2026

What happened: DeepSeek, the Chinese AI lab, is close to completing a funding round that values the company at approximately $45 billion, with China's state chip fund leading the investment.

Key details:

• DeepSeek is approaching a funding round valuation of roughly $45 billion
• China's state chip fund is leading the round
• This funding represents major support for the Chinese AI ecosystem

Why it matters: A $45 billion valuation for a Chinese AI company signals significant state-level support for AI development and chip independence from Western suppliers. This follows increased geopolitical tensions around semiconductor access and reflects China's strategic priority on AI self-sufficiency. The involvement of the state chip fund suggests government backing beyond typical venture capital, potentially enabling DeepSeek to compete more aggressively with US-based AI labs despite US export controls on advanced chips.

Practical takeaway: Watch DeepSeek's model releases closely—with state backing and a $45B valuation, the lab has substantial resources to compete on both frontier models and open-weight alternatives, potentially reshaping global AI competition.

What happened: Anthropic secured the full computing capacity of SpaceX's Colossus-1 data center and committed to spending $200 billion on Google Cloud over five years, while simultaneously launching new capabilities for Claude Managed Agents including a "Dreaming" feature for autonomous learning.

Key details:

• Anthropic is taking over more than 300 megawatts and over 220,000 NVIDIA GPUs from SpaceX's Colossus-1 data center, expected to come online within a month
• The deal represents approximately $5 billion per year and reflects 8000% annualized ARR growth
• Anthropic has committed to spending roughly $200 billion on Google Cloud over the next five years—more than 40 percent of Google's entire cloud backlog
• New "Dreaming" feature for Claude Managed Agents is an asynchronous process that reviews past agent sessions, cleans up duplicate or outdated memory entries, and distills new insights
• Outcomes and Multiagent Orchestration are now both in public beta alongside Dreaming
• Anthropic is doubling rate limits for Claude Code and significantly raising API limits for Opus models

Why it matters: These moves position Anthropic as a major computational force capable of scaling Claude inference and training at unprecedented scale, while agent features like Dreaming enable systems to continuously improve without human intervention—critical for long-running autonomous workflows. The $200B Google Cloud commitment reveals the enormous financial commitment required to power frontier AI, with Anthropic and OpenAI now accounting for roughly half of $2 trillion in committed cloud revenue across Amazon, Microsoft, Google, and Oracle.

Practical takeaway: If you're building on Claude or planning to rely on its API, expect significant capacity improvements and new agent orchestration capabilities in the coming months.

What happened: OpenAI removed the $50,000 minimum budget requirement for ChatGPT ads, opening self-serve advertising to small businesses and signaling aggressive monetization of its user base.

Key details:

• The $50,000 minimum budget requirement for ChatGPT ads is gone
• Advertisers in the US can now book ads on their own through a self-serve platform
• OpenAI is targeting $2.5 billion in ad revenue this year
• The move represents a shift toward a full self-serve ad platform model

Why it matters: By democratizing ChatGPT advertising access, OpenAI is building a new revenue stream that doesn't rely on cloud compute sales or API licensing. Reaching $2.5 billion in ad revenue would represent a major source of cash flow at scale, but also signals OpenAI's confidence in ChatGPT's user engagement. This diversification of revenue models is critical as the company approaches a potential IPO and needs to demonstrate multiple profitable business lines.

Practical takeaway: If you operate a small business, monitor ChatGPT ads as a potential channel—the removal of minimum spend requirements makes it accessible for testing, though effectiveness will depend on ChatGPT's targeting and user intent.

What happened: Google updated its AI Search features to include "a preview of perspectives" from firsthand sources like Reddit and other web forums, improving source attribution and user trust in AI-generated summaries.

Key details:

• Google is updating AI Search features (AI Overviews) to include perspectives from firsthand sources
• The update specifically highlights content from Reddit and other web forums and social media platforms
• The change is designed to make it easier for users to find information from sources they know and trust
• Search results now link user queries with online conversations around similar topics

Why it matters: This update represents Google's attempt to address trust concerns about AI-generated summaries by grounding them in community-driven content. By prominently featuring Reddit and other forums, Google acknowledges that users value peer perspectives and real-world experiences alongside AI-generated overviews. This also signals shifting user behavior and expectations around search—people increasingly want to see what real humans are discussing, not just AI summaries.

Practical takeaway: If you're publishing community content or user discussions online, expect increased visibility in Google Search results as the search engine prioritizes firsthand perspectives in AI-powered overviews.

What happened: Google released multi-token prediction optimizations for its Gemma 4 model achieving 3x speedup, while the field continues advancing on context window sizes and training methodologies.

Key details:

• Google released multi-token prediction drafters for its Gemma 4 open model family
• The optimization speeds up text generation by up to three times
• A small auxiliary model suggests several tokens at once while the main model checks them in a single pass
• Recent releases include models with 12 million token context windows
• vLLM released improvements focused on correctness in reinforcement learning workflows

Why it matters: Multi-token prediction represents a practical efficiency gain that doesn't require model retraining—auxiliary models can accelerate existing models without architectural changes. Combined with expanding context windows and RL refinements, these improvements make open models increasingly competitive with closed-source alternatives on both speed and capability. For inference-heavy applications, these optimizations directly reduce latency and computational cost.

Practical takeaway: If you're using Gemma 4 for generation tasks, update to versions with multi-token prediction support to get immediate 2-3x latency improvements without model changes.

What happened: OpenAI developed MRC, an open-source network protocol created in partnership with AMD, Broadcom, Intel, Microsoft, and NVIDIA, designed to eliminate bottlenecks in large-scale AI supercomputer infrastructure.

Key details:

• MRC is an open-source network protocol that sends data across hundreds of paths simultaneously between GPUs
• Instead of the traditional three or four switch layers, MRC requires only two layers to connect over 100,000 GPUs
• The protocol reduces both power consumption and infrastructure costs
• MRC is already running on OpenAI's Stargate supercomputer
• The protocol was developed collaboratively with AMD, Broadcom, Intel, Microsoft, and NVIDIA

Why it matters: AI supercomputers face a critical bottleneck in networking—data movement between hundreds of thousands of GPUs can become a limiting factor for training and inference speed. By requiring fewer switch layers while supporting more GPUs, MRC enables more efficient resource utilization and lower operational costs for frontier model training and deployment. Open-sourcing the protocol signals a shift toward industry standardization for AI infrastructure challenges.

Practical takeaway: Watch how MRC adoption progresses across cloud providers and data centers, as this networking architecture will likely become a standard requirement for competitive large-scale AI inference infrastructure.