The Stanford Institute for Human-Centered Artificial Intelligence (HAI) released its 2026 annual AI Index report, revealing several key findings: global generative AI adoption reached 53% within three years, surpassing the early internet's diffusion speed; enterprise AI adoption rose to 88%, with 79% routinely using generative AI. The performance gap between U.S. and Chinese AI models has largely disappeared, with China leading in research papers, patents, and robot deployments, while U.S. private AI investment hit $285.9 billion—23 times that of China. Employment among U.S. software developers aged 22–25 declined nearly 20%. AI's environmental costs remain high, with Grok-4's training emissions equivalent to those of 17,000 cars over a year. Only 31% of Americans trust AI regulation, compared to 84% in China.

According to an internal OpenAI memo dated April 12, revenue lead Denise Dresser stated the company is actively advancing its partnership with Amazon to expand in the enterprise market and reduce reliance on Microsoft. Many enterprise customers have already built AI applications on AWS Bedrock, with demand described as "remarkable" since the collaboration was announced at the end of February. Under the agreement, Amazon will invest up to $50 billion in OpenAI, far exceeding Microsoft’s previous investment of over $13 billion. This move is seen as a critical strategy to counter Anthropic’s rise in the enterprise AI market. OpenAI achieved a valuation exceeding $850 billion in late March financing, while Anthropic reached $380 billion, with both companies potentially going public this year.

AI legal tech company Harvey announced a new round of funding totaling $200 million, achieving a $11 billion valuation, led by GIC and Sequoia, with participation from Andreessen Horowitz, Coatue, and Kleiner Perkins. The company has raised over $1 billion cumulatively. More than 25,000 custom AI agents are now operating on the Harvey platform, serving over 100,000 lawyers across 1,300 organizations in more than 60 countries, including most AmLaw 100 law firms, over 500 corporate legal departments, and 50 asset management firms. The CEO stated that AI is transitioning from an assistive tool to a core system in legal work, enabling automation of high-complexity tasks such as mergers and due diligence.

Moonshot AI's Kimi K2 is now available on Nscale inference endpoints. The model features 1 trillion parameters, uses a Mixture-of-Experts architecture, has native visual processing capabilities, and supports a 256K context window. It performs strongly across multiple benchmarks: SWE-Bench Verified reaches 76.8%, BrowseComp achieves 60.6% (outperforming Claude Opus and Gemini 3 Pro), and GPQA-Diamond hits 87.6%. Developers can deploy via API with pricing set at $0.45 per million input tokens and $2.20 per million output tokens, supporting fully managed inference, on-demand scaling, and data isolation.

Google Cloud has launched a new tool called QueryData, designed to improve the accuracy of AI agents querying databases, claiming near-perfect natural language to SQL translation. Enterprises must first define a 'context' containing database schema, table relationships, and business semantics, which is then iteratively validated for query precision using the Context Engineering Assistant in Gemini CLI and the Evalbench framework. It supports AlloyDB, CloudSQL for MySQL/PostgreSQL, and Spanner, and can be integrated into BigQuery data agents or custom agent systems. Analysts note this approach focuses on secure data-layer access, differentiating it from AWS and Microsoft’s connector-centric strategies, though it incurs higher upfront design and maintenance overhead.

Chinese AI startup StepFun (Jieyue Xingchen) plans to dismantle its offshore corporate structure to pave the way for a domestic listing. Founded in April 2023 in Shanghai by former Microsoft executive Jiang Daxin, StepFun focuses on developing general-purpose foundational large models and is considered one of China’s leading AI startups. Given its primary backing by state capital, the company has decided to adopt a domestically compliant corporate structure to advance its initial public offering process. This reflects a broader trend among Chinese AI firms adjusting their financing and listing strategies.

On April 13, 2026, 16 professional academic institutions—including the China Association for Artificial Intelligence, China Computer Society, and World Robotics Cooperation Organization—jointly issued the 'Global AI Governance Initiative for Scientific and Technological Associations' under the coordination of the China Association for Science and Technology. Centered on the principles of 'benefiting people's livelihood' and 'safe and controlled,' the initiative requires AI to remain under human control and respects all nations’ equal rights to participate in governance. It calls for establishing cross-sectoral collaborative mechanisms, promoting mutual learning among governance systems, enhancing public literacy through science communication, and urging global scientific societies to uphold professional ethics in advancing 'intelligent good'.

Beijing released the 'Measures to Support High-Quality Development of Innovative Pharmaceuticals (2026)', introducing 32 upgraded policies. AI-driven drug discovery is elevated from a 'supporting tool' to a 'core driver,' with support for building no fewer than 15 types of AI prediction models and organoid models, and investments in pharmaceutical AI foundation models and intelligent computing infrastructure. For surgical robots, eligible robotic-assisted procedures will be included in medical insurance billing, promoting hospital adoption across laparoscopic, orthopedic, and neurosurgical categories, with R&D and usage outcomes incorporated into public hospital performance evaluations. Additionally, 674 nationally negotiated drugs are directly added to hospital formularies, and Beijing’s inclusive health insurance will cover 159 specialty drugs in 2026.

AWS announced several major updates: Amazon Bedrock now offers a preview of Claude Mythos, Anthropic’s most advanced cybersecurity AI model capable of identifying software vulnerabilities and analyzing large codebases, currently available only to whitelisted organizations. Simultaneously, AWS Agent Registry enters preview, providing centralized AI agent discovery and governance with semantic search and approval workflows to help enterprises manage growing AI agent deployments. Additionally, Amazon S3 Files enables S3 buckets to be accessed as file systems, and Amazon Braket now supports Rigetti’s 108-qubit Cepheus QPU.

Global AI compute supply is experiencing severe shortages, signaling the end of the free AI era. Spot prices for NVIDIA’s latest chips have surged nearly 50% in weeks, with providers like CoreWeave raising prices over 20% and demanding long-term contracts. OpenAI canceled a video project due to compute constraints, while Anthropic frequently faces system outages. Token throughput has more than doubled within months, but data center construction lags, unable to meet rising demand. Despite Anthropic’s revenue jumping from $9 billion to $30 billion, infrastructure delays have become a key bottleneck for AI development. The industry is shifting from software-led to resource-intensive models, forcing enterprises to prioritize AI usage and optimize architectures.