Nvidia reported its first-quarter fiscal 2027 earnings, with revenue reaching $81.6 billion, up 85% year-over-year, and net income of $58.3 billion, a 211% increase. Data center business revenue was $75.2 billion, up 92%, including $14.8 billion in networking revenue, which surged 199% year-over-year. The company has split its data center segment into 'hyperscale' and 'ACIE' (AI Cloud, Industrial, and Enterprise), signaling that AI infrastructure spending is expanding beyond large cloud providers to enterprises, industrial, and telecom sectors. Nvidia expects next quarter's revenue to reach approximately $91 billion and plans to launch its Vera CPU platform in the second half of 2026. The company projects global annual AI infrastructure spending will reach $3–4 trillion by 2030.

Anthropic forecasts Q2 2026 revenue of $10.9 billion, a more than 130% increase from the previous quarter, with an expected quarterly operating profit of approximately $559 million—its first profit since inception. Growth is driven by strong performance from Claude Code, improved computational efficiency, and enterprise customer expansion. SpaceX’s S-1 filing reveals Anthropic pays around $1.25 billion monthly to rent compute capacity at the Colossus data center, with a contract running through May 2029 and a total three-year value of about $450 billion, covering over 220,000 NVIDIA GPUs. Analysts note that part of this quarter’s profitability stems from discounted rates offered by SpaceX for the first two months, raising questions about sustained profitability going forward.

Alibaba’s Tongyi Qianwen team has released Qwen3.7-Max, a proprietary flagship model, scoring 56.6 on the Artificial Analysis Intelligence Index, ranking fifth globally and first among domestic models, with performance close to GPT-5.4. Designed specifically for agent scenarios, the model supports a 1-million-token context window and 64K output, enabling continuous autonomous operation for 35 hours and completing 1,158 tool calls with a 10x geometric mean speedup. API pricing is set at $2.5 per million input tokens and $7.5 per million output tokens, significantly undercutting competitors. Unlike previous open-source versions, Qwen3.7-Max is available only via API, sparking developer concerns over data control and lack of local deployment options.

Paul Nakasone, OpenAI board member and former NSA director, visited Japan on May 21 and announced the company is considering offering a cybersecurity-enhanced AI model, GPT-5.5 Cyber, to Japanese government agencies and select enterprises. Focused on areas like system vulnerability detection, the model aims to counter cybersecurity threats posed by China’s accelerating AI development. Nakasone stated he had met with Japanese policymakers and also revealed OpenAI is in discussions with the European Union on similar collaborations. Japan’s three major banks are also exploring potential cooperation with OpenAI in AI-driven cybersecurity. This move marks a strategic effort by OpenAI to strengthen cybersecurity partnerships with allied nations.

Hark, an AI lab founded by Brett Adcock, announced a $700 million Series A funding round, achieving a post-money valuation of $6 billion. The round was led by Parkway Venture Capital, with participation from Nvidia, AMD Ventures, ARK Invest, and Salesforce Ventures. Hark is developing agent-based AI systems intended to serve as a universal interface across the digital world, planning to release its first multimodal AI model this summer, followed by dedicated hardware devices. The team currently numbers around 70 and operates a data center equipped with Nvidia B200 GPUs. Former Apple product executive Abidur Chowdhury serves as Design Director, emphasizing the product is designed for general users, not just developers.

Tencent Hunyuan has open-sourced its new translation model series Hy-MT2, offering 1.8B, 7B, and 30B-A3B variants supporting translation across 33 languages. The 7B and 30B-A3B models achieve open-source SOTA on general benchmarks like FLORES-200, outperforming mainstream commercial translation models in vertical domains such as finance and politics. The 30B-A3B model introduces a MoE architecture for the first time, with 300 billion total parameters but only 30 billion activated, balancing performance and efficiency. A 1.25-bit ultra-quantized version based on Tencent’s proprietary Sherry framework requires only 440MB storage and runs 1.5x faster than the 4-bit version on Apple A15 chips, enabling offline translation. A companion mini-program, 'Tencent Hy Translate,' has also been launched.

22-year-old entrepreneur Kye Gomez reverse-engineered the core architecture of Anthropic’s closed-source model Claude Mythos using first-principles reasoning, naming it OpenMythos and releasing it fully open-source. The architecture employs Recurrent Deep Transformer (RDT), allowing up to 16 cycles of inference using the same set of weights for efficient iterative thinking. It consists of three components: Prelude, Recurrent Block, and Coda, combining MoE for breadth and recurrence for depth, while incorporating DeepSeek-V2’s multi-latent-variable attention mechanism to significantly reduce KV cache memory consumption. Experiments show a 770M-parameter model achieves performance comparable to a 1.3B standard Transformer, doubling parameter efficiency. This breakthrough challenges the technical moat of closed-source models.

Meta announced layoffs affecting approximately 8,000 employees (about 10% of its workforce) while reassigning around 7,000 others to AI teams, reallocating resources to focus aggressively on AI development. This reduction occurs despite record-high revenue and profits. Internal recordings reveal Meta is implementing a 'Model Capability Program' that collects employee keyboard and mouse activity data for AI training, with Zuckerberg viewing elite engineers’ behavioral data as high-quality training material. This large-scale organizational restructuring reflects tech giants’ shifting strategic priorities amid the AI wave and underscores AI’s profound impact on employment structures.

Kore.ai has launched the Artemis AI agent platform, enabling AI to autonomously design, build, test, and deploy enterprise-grade AI agents, reducing traditional development timelines from months to days. Key technologies include Agent Blueprint Language (ABL), a YAML-based standard for defining and governing agents; AI Architect Arch, which converts natural language business requirements into production-ready code; and a 'dual-brain architecture' that combines LLM reasoning with deterministic business rule execution, enhancing safety for regulated industries like finance and healthcare. The platform supports 175 AI models and over 400 enterprise integrations, emphasizing vendor neutrality. Kore.ai has raised $223 million and serves over 500 Fortune Global 2000 companies.

ChromaDB has disclosed a critical security vulnerability with a CVSS score of 10.0 (CVE-2026-45829), allowing unauthenticated attackers to achieve remote code execution via malicious Hugging Face models. The flaw stems from the Python FastAPI server loading user-controlled embedding model configurations before authentication, leveraging the trust_remote_code=True mechanism in the Python ecosystem to automatically download and execute remote malicious code. Introduced in version 1.0.0, the vulnerability remains unpatched as of version 1.5.8. Globally, 73% of publicly exposed ChromaDB instances run affected versions, facing severe risks.