Google Cloud Debuts New AI Chips | Bloomberg Tech 4/22/2026

Key Concepts

• TPU (Tensor Processing Unit): Google’s custom-designed application-specific integrated circuits (ASICs) optimized for AI training and inference.
• Inference vs. Training: Training involves teaching AI models using massive datasets; inference is the process of the model executing tasks (e.g., chatbots, agentic AI).
• Agentic AI: AI systems capable of performing complex, multi-step tasks autonomously over extended periods.
• Data Flywheel: A self-reinforcing cycle where product usage generates data, which is then used to improve AI models, leading to better products and more usage.
• Rule of 40: A metric for software companies where the sum of revenue growth rate and profit margin should exceed 40%.
• AIOS (AI Operating System): Software infrastructure layer that manages data, compute, and storage for AI workloads.

1. Google Cloud’s AI Infrastructure Expansion

Google Cloud announced a new generation of TPUs, specifically the TPU 8T (for training) and TPU 8i (for inference).

• Strategic Advantage: Mandep Singh (Bloomberg Intelligence) notes that Google’s vertical integration—owning the silicon, the cloud, and the models (Gemini)—positions them as the "lowest cost token provider."
• Partnerships: Google expanded its ecosystem through deals with Oracle (natural language data interaction), Nvidia (AI factories), Salesforce, Crowdstrike, and Broadcom.
• Market Position: Google is aggressively capturing market share in AI workloads, with cloud growth significantly outpacing competitors like Azure.

2. Rivian’s R2 Launch and Strategic Pivot

Rivian CEO RJ Scaringe confirmed the start of production for the R2 SUV at the Normal, Illinois plant, despite recent tornado damage to the facility.

• Technical Specs: The R2 features in-house silicon (800 TOPS per chip, two chips per vehicle) and integrates LiDAR, radar, and cameras.
• Autonomous Roadmap: Rivian is transitioning toward Level 2 (hands-off, eyes-on) and eventually Level 3 (hands-off, eyes-off) autonomy.
• Business Model: Rivian is pursuing a dual-revenue strategy: selling vehicles and licensing its software/compute stack to other OEMs (e.g., the $5.8 billion deal with Volkswagen).

3. Anthropic’s "Mythos" Security Breach

A small group of unauthorized users gained access to Anthropic’s new AI model, Mythos, which the company had restricted due to its potential for enabling dangerous cyberattacks.

• Methodology: The breach was not a sophisticated hack but rather "sleuthing." Users guessed the URL structure based on previous models and leveraged credentials obtained through a third-party contractor.
• Implications: This highlights a critical vulnerability in AI security protocols, raising concerns about how frontier models are protected when shared with third-party partners.

4. SpaceX and Cursor: The Race for Coding AI

SpaceX has secured the right to acquire the AI coding startup Cursor for $60 billion, with a $10 billion breakup fee.

• Strategic Rationale: The deal is a response to the intense competition for compute resources. Cursor’s agentic coding model is highly valued, and this acquisition (or the threat of it) secures the necessary infrastructure for Elon Musk’s AI ambitions (XAI).

5. Vast Data’s $30 Billion Valuation

Vast Data raised $1 billion in a Series F round, tripling its valuation.

• Function: The company acts as the "middle layer" of the AI stack, providing storage and data management (AIOS) for massive GPU clusters.
• Growth: CEO Renan Hak highlighted the company’s efficiency, citing a "Rule of 228" (a play on the Rule of 40), and indicated that an IPO is being considered for later this year.

6. KPMG Survey: The "Great Skill Reset"

A KPMG survey of 2,100 C-suite executives revealed that AI investment is a top priority despite economic uncertainty.

• Key Findings: While 92% of firms have an AI strategy, only 8% report a clear ROI.
• Shift in Focus: Companies are moving from "pilot mode" to enterprise-wide deployment. The focus is shifting toward upskilling existing employees rather than just hiring, a phenomenon termed the "Great Skill Reset."

Synthesis and Conclusion

The current tech landscape is defined by a massive "compute scramble." Companies are vertically integrating—designing their own chips (Google, Rivian), securing massive data infrastructure (Vast Data), and acquiring specialized AI talent (SpaceX/Cursor)—to gain a competitive edge. While investment levels are at record highs, the industry is currently grappling with a "value gap," where the focus is shifting from experimental pilots to tangible, agentic AI workflows. Security remains a significant bottleneck, as evidenced by the Anthropic breach, suggesting that as AI capabilities grow, the governance and protection of these models must evolve at an equal pace.