2026.05.08 AI News Briefing — Microsoft: "Korea's AI adoption is rising fastest"; Anthropic ARR overtakes OpenAI; 'physical AI' closes out COEX

In its Q1 2026 'Global AI Diffusion Report', Microsoft singled out Korea, along with Japan and Thailand, as the countries with the 'largest gains'. At the same time, Anthropic's annualized revenue (ARR) of about $30 billion overtook OpenAI (about $24 billion), and Oracle is cutting 20,000–30,000 jobs while redirecting $8–10 billion into AI infrastructure. 'AI EXPO KOREA 2026' at COEX closes its final day with an 'AI Summit'.

---

1. Microsoft Global AI Diffusion Report — Korea, Thailand, and Japan post the biggest gains; UAE leads at 70.1%

• Source: Microsoft On the Issues / Global AI Diffusion Report | 2026.05.07
• What happened: Microsoft released its Q1 2026 'Global AI Diffusion Report'. During the quarter, AI usage among the world's working-age population rose 1.5 percentage points from 16.3% to 17.8%, and the number of countries above 30% grew to 26. The UAE leads at 70.1%, and the US climbed from 24th to 21st at 31.3%. Microsoft noted that 'as Asian-language model performance improves, Korea, Thailand, and Japan show the largest gains'. The same report found global Git pushes up 78% year on year in Q1, and US software developer employment reaching a record 2.2 million in 2025, up 8.5%.
• Why it matters: It is global validation that Korea has positioned itself simultaneously as 'a country that builds AI fast' and 'a country that uses AI fast'. The statistics line up exactly with the government's 'sovereign AI' policy, the industry ministry's 'AI Factory' call, and the momentum of 'AI EXPO KOREA' at COEX.

2. Anthropic ARR of about $30 billion overtakes OpenAI — tripled in four months

• Source: devFlokers / TLDL / Air Street Press | 2026.05.03–07
• What happened: A string of reports found that Anthropic's annualized revenue (ARR) of about $30 billion has overtaken OpenAI (about $24 billion) for the first time. Anthropic grew its ARR roughly threefold in four months from about $9 billion in January 2026, and now has more than 1,000 enterprise customers spending over $1 million a year. Based on Ramp data, 70% of 'companies adopting AI for the first time' choose Anthropic, far ahead of OpenAI (26%). Meanwhile, Anthropic secured an additional $40 billion from Google and $5 billion from Amazon (including a $100 billion AWS compute package), and is reported to have signed chip deals worth hundreds of billions of dollars with Google and Broadcom.
• Why it matters: The 'AI Big Three' lineup has effectively been reshuffled. As the center of gravity shifts from the 'chatbot race' to 'agentic workflow revenue', Korean companies' criteria for adopting LLMs are also changing from 'performance' to 'operational stability and governance of agents'.

3. Oracle cuts 20,000–30,000 jobs while redirecting $8–10 billion into AI infrastructure

• Source: Crescendo AI News | 2026.05
• What happened: Oracle announced it will redirect $8–10 billion into AI infrastructure through layoffs of 20,000–30,000 people. In the same period, JPMorgan Chase officially reclassified AI investment from 'experimental R&D' to 'core infrastructure', specifying a 2026 technology budget of about $19.8 billion and 2,000 dedicated AI staff. By the bank's internal estimate, AI creates about $2.5 billion in annual value through efficiency gains and revenue growth, and its models scan more than $10 trillion in transactions every day. A Science study from Harvard Medical School and BIDMC was also published, finding that an OpenAI reasoning model beat veteran physicians on diagnosis and management accuracy in a Boston emergency department.
• Why it matters: Global Big Tech, finance, and healthcare are all shifting their stance to 'AI is infrastructure, not an experiment'. Korean conglomerates and financial institutions need the same redefinition of their IT budget categories and staffing.

4. AI EXPO KOREA 2026 closes — 'AI Summit' and the declaration of physical, 'action-oriented AI'

• Source: ZDNet Korea / Industry Journal / AI Times / KMJ | 2026.05.06–08
• What happened: The 9th AI EXPO KOREA 2026 at COEX closes Friday the 8th. The final day features the 'AI Summit', where Son Byung-hee, research director at physical AI company Maum AI, addresses the shift in industry structure with the message 'AI is no longer a technology that answers, but one that works'. AIWORKX gave live demos of four industry-specific 'task-completing' agent solutions, 'AxDC (Action × Do · Complete)'; Enerzai showcased an on-device voice AI solution based on ultra-low-bit quantization; and A3Security put its deepfake detector 'TruthLens' and manufacturing AI 'AIMES' front and center. Day two of the main conference features DIGITIMES chairman Colley Hwang's analysis of the global supply chain and tech hegemony, and Qualcomm's outlook session on on-device and edge AI.
• Why it matters: It shows Korea's AI industry evolving a step beyond 'chatbot and LLM demos' to 'task-completing, action-oriented agents and on-device AI'. The first week of May is likely to be recorded as a turning point for the Korean AI industry.

5. SpaceX holds a $60 billion option to acquire AI coding IDE Cursor — exercisable after its summer IPO

• Source: Air Street Press / State of AI: May 2026 | 2026.05
• What happened: SpaceX is reported to have secured a $60 billion acquisition option ahead of Cursor's $2 billion funding round. An option for a $10 billion AI partnership deal was offered as well, with any acquisition deferred until after SpaceX's summer IPO. In the same period, OpenAI acquired Hiro (a personal finance agent), its seventh acquisition of 2026. Also notable: Canada's Cohere and Germany's Aleph Alpha merged with the blessing of both governments, positioning themselves as a 'sovereign AI' alternative. China formally blocked Meta's $2 billion acquisition of domestic agent startup Manus.
• Why it matters: Big Tech is turning into 'holding companies' across AI infrastructure, coding, and agents. Korean AI startups' M&A bargaining power will rise temporarily, but the risk of dependence on a single infrastructure grows with it.

6. Mandiant 'M-Trends 2026' — "exploits arrive before patches", the era of AI-assisted attacks

• Source: The Hacker News / Mandiant / Edgescan | 2026.05.04
• What happened: In its 'M-Trends 2026' report, Mandiant declared that 'time-to-exploit has effectively gone negative'. 28.3% of CVEs are used in real attacks within 24 hours of disclosure, and exploits arriving before patches is becoming routine. The average time to patch is 74 days, and even enterprises with more than 1,000 employees never fix 45% of their vulnerabilities. The 2025 'Shai-Hulud' attack on the npm ecosystem infected more than 500 packages and led to widespread damage, including the theft of $8.5 million from Trust Wallet. The analysis is that the surge in SWE-bench scores for ChatGPT, Claude, and Gemini works just as well for attackers.
• Why it matters: The data confirms that 'AI makes attackers, not just defenders, four times faster'. Korean companies' security operations (SecOps) strategy needs to shift its center of gravity from a 'speed race' to 'structural immunity'. It is the same trend as the UK AISI's 'cyber doubling every four months' assessment and the Five Eyes agentic AI guidance.

---

Today's insight

• Korea the 'user nation' accelerates alongside Korea the 'builder nation': The Microsoft report is the first global data to confirm Korea as 'one of the countries where AI usage is rising fastest'. The government's sovereign AI investment, the industry ministry's AI Factory program, and the 'task-completing agent' demos at COEX are all faces of the same trend. From now on, the metric for Korean companies' AI strategy is not 'adoption' but 'intensity of use'.
• The center of the AI revenue curve is moving to 'agents': Anthropic's ARR overtaking OpenAI is not a mere ranking; it is a signal that 'agents, not chatbots, make money'. Korean SI and SaaS companies should now be tracking the unit price, NRR, and failure rate of 'task-completing agents' rather than running 'chatbot adoption' PoCs.
• AI security becomes a fight over 'structure', not 'speed': The Mandiant report, the AISI assessment, and the Five Eyes guidance all point to the same conclusion. In content and customer governance systems such as AEM, DAM, and CRM, 'agent activity logging plus permission scope limits plus human review steps' will soon be standard requirements.