OpenAI launched the GPT-5.3-Codex programming model just 15 minutes after Anthropic released Claude Opus 4.8. The model excels in benchmarks such as SWE-Bench Pro, TerminalBench 2.0, and OSWorld, requiring over half fewer tokens and significantly improving efficiency. GPT-5.3-Codex also features self-acceleration capabilities, enabling it to participate in debugging and optimizing its own training process. Meanwhile, OpenAI introduced Frontier, an enterprise-grade Agent platform, and partnered with Ginkgo to reduce protein synthesis costs by 40% using GPT-5. This marks a further escalation in the competition between OpenAI and Anthropic in the AI programming domain.

StepFun has released Step 3.7 Flash, an open-source 198-billion-parameter sparse Mixture-of-Experts (MoE) vision-language model under the Apache 2.0 license. It activates only about 11 billion parameters during runtime and supports a 256K context window. Optimized for agent workflows, the model features native multimodal understanding, with tool-calling reliability exceeding 98% on the τ²-bench test. The newly introduced Advisor Mode achieves 97% of Claude Opus 4.6's performance on SWE-Bench Verified, with a single-task cost of just $0.19. Cross-framework coding performance variance has narrowed from 43–73% to 64.5–71.5%, enhancing stability. The model is now open-sourced on platforms like Hugging Face and can run locally on devices such as Mac Studio M4 Max and DGX Spark.

SpaceX has reduced its IPO target valuation from over $2 trillion to at least $1.8 trillion following investor feedback, aiming to raise up to $75 billion. Roadshows are scheduled to begin on June 4, with listing planned for June 12 on Nasdaq under the ticker SPCX. The company disclosed 2025 revenue of $18.67 billion and a net loss of $4.94 billion, with Starlink generating $11.4 billion in revenue and $4.42 billion in operating profit—its only profitable segment. The AI division incurred an operating loss of $6.36 billion. Capital expenditures reached $10.1 billion in Q1 2026, of which $7.72 billion was allocated to AI infrastructure. Goldman Sachs leads a consortium of 23 banks in underwriting, with Elon Musk holding 42% equity and 85% voting rights.

On May 28, CNN filed a lawsuit against Perplexity AI in the U.S. District Court for the Southern District of New York, accusing the company of unauthorized copying and redistribution of over 17,000 CNN articles, videos, and images for its commercial AI products. The complaint alleges that Perplexity ignored robots.txt blocking directives, used stealth crawlers with disguised IPs to bypass technical protections, and continued using CNN content after failed negotiations. CNN provided multiple examples where Perplexity’s outputs closely resemble original content, including behind-the-paywall articles. Perplexity is also accused of falsely claiming users could access premium CNN content via subscription. This case follows major legal challenges by The New York Times and Dow Jones against AI data scraping practices.

Dell’s stock surged 30% in a single day following its Q1 earnings report, surpassing market expectations, primarily driven by AI server revenue reaching $60 billion. Analysts note that the market previously underestimated the trend of enterprises building in-house AI infrastructure. Dell has collaborated with NVIDIA in promoting the 'enterprise AI factory' concept, enabling companies to deploy AI computing power on-premises, with over 5,000 customers already adopting the solution. NVIDIA’s latest earnings also separately highlighted growth in the enterprise segment. HP, Lenovo, and other server vendors are performing strongly as well, indicating broad diffusion of AI hardware demand and a growing trend toward localized AI deployment.

Lightmatter officially listed on the Hong Kong Stock Exchange on May 30, with an issue price of HK$183.2 per share and pre-market trading exceeding HK$830, marking a gain of over 350%. The company focuses on optoelectronic hybrid computing technology, leveraging optical computing and optical interconnects to overcome power, bandwidth, and latency bottlenecks inherent in traditional electronic computing. In 2025, revenue reached RMB 106 million, with an 88.3% market share in optical interconnects and optical computing contributing 20.8% of total revenue. Approximately 70% of raised funds will be allocated to R&D, focusing on silicon photonics chips and optoelectronic transmission technologies. Backed by Alibaba, CMCC Capital, GIC, and Hillhouse, the IPO aims to advance self-reliance and control in silicon photonics technology.

xAI has released grok-build-0.1, a cost-optimized agent model designed for programming tasks, currently available in public beta via API. The model has been integrated into development platforms including OpenRouter, Vercel AI Gateway, and Cursor. Positioned as a high-value intelligent programming agent, it competes with OpenAI Codex and Anthropic Claude Code. Early user testing indicates it can generate images, though video generation and reading X posts remain unavailable. Its programming capabilities still lag behind Codex and Cursor. This move marks xAI’s official entry into the AI programming tools arena.

Gartner forecasts global AI spending will reach $2.59 trillion in 2026, a 47% year-on-year increase. AI infrastructure will be the largest spending category, accounting for over 45% of total expenditure, with AI-optimized server spending expected to triple over the next five years. AI model spending will see a short-term growth of 110% in 2026. Currently, AI spending is primarily driven by tech companies and hyperscalers, while enterprises have not yet fully tapped their investment potential and tend to adopt incremental AI projects. At the same time, more companies are re-evaluating AI ROI and imposing usage restrictions, highlighting cost control as an emerging priority.

BYD has launched the Xuanji A3, China’s first self-developed automotive-grade 4nm intelligent driving chip. The chip adopts a dedicated NPU architecture instead of a general-purpose GPU approach, hardwiring common AI operators such as matrix multiplication and convolution, reducing power consumption per unit compute by 20% and doubling compute utilization. BYD now operates five wafer fabs and maintains a 7,000-member chip R&D team, offering a product portfolio covering 13 categories and 567 types of automotive-grade chips, serving 46 external car brands. The chip natively supports Transformer large models, and its low-latency, high-efficiency characteristics are also suitable for robotics, drones, and other Physical AI applications.