AI News Archive: June 8, 2026 — Part 2
Sourced from 500+ daily AI sources, scored by relevance.
- AMD is committing £2 billion to power U.K. AI supercomputers and research
The chipmaker's five-year pledge includes new AI supercomputers at Cambridge and partnerships with Imperial College London
- Oriole Networks deploys photonic AI network in first commercial rollout with AMD
U.K. photonic networking startup Oriole Networks Ltd. today announced that it will deploy what it describes as the world’s first large-scale artificial intelligence system running on a pure photonic network, marking the first commercial use of its technology. The deployment, built in collaboration with Advanced Micro Devices Inc., forms part of the U.K. Advanced Research […] The post Oriole Networks deploys photonic AI network in first commercial rollout with AMD appeared first on SiliconANGLE .
- Trump memo on AI aims to avoid repeat of Anthropic debacle
Trump memo on AI aims to avoid repeat of Anthropic debacle Breaking Defense
Score: 64🌐 MovesJun 8, 2026https://breakingdefense.com/2026/06/trump-memo-on-ai-aims-to-avoid-repeat-of-anthropic-debacle/ - Advancing U.S.–UK Cooperation to Secure Frontier Artificial Intelligence
This interim report identifies a significant opportunity for collaboration between the United States and the United Kingdom and proposes a framework for coordinating efforts to secure frontier artificial intelligence development.
- VinDynamics and Skild AI form strategic partnership to advance humanoid ROBOTICS
VinDynamics and Skild AI form strategic partnership to advance humanoid ROBOTICS USA Today
- South Korea names first female prime minister in decades to lead AI push
South Korea names first female prime minister in decades to lead AI push The Japan Times
Score: 64🌐 MovesJun 8, 2026https://www.japantimes.co.jp/news/2026/06/08/asia-pacific/south-korea-female-pm-ai/ - Starmer positions government as active partner in Britain’s AI future
Starmer positions government as active partner in Britain’s AI future Computing UK
Score: 62🌐 MovesJun 8, 2026https://www.computing.co.uk/news/2026/government/starmer-positions-govt-active-partner-ai-future - Skild AI partners with Vietnamese firm to explore integrating AI into humanoid robots
The partnership will focus on validating humanoid robotics systems and integrating Skild's AI model into VinDynamics' platforms. The deal comes months after a partnership with Nvidia.
- Here's How to Try Out a Robotaxi Ride in London Using the Uber App
Here's How to Try Out a Robotaxi Ride in London Using the Uber App PCMag Australia
Score: 62🌐 MovesJun 8, 2026https://au.pcmag.com/ai/118128/heres-how-to-try-out-a-robotaxi-ride-in-london-using-the-uber-app - Apple found a way to sharply cut token use
“Sonata” proposes a "reliable proxy for thinking necessity"
Score: 62🌐 MovesJun 8, 2026https://www.thestack.technology/apple-found-a-way-to-sharply-cut-token-use-2/ - 'The time has come to reclaim what was stolen from us': Bernie Sanders wants the American public to own 50% stake in AI companies
Under this proposal, US AI firms could have to cough up a one-time 50% tax (in shares) into an AI Sovereign Wealth Fund.
- Unitree Files for IPO, Aiming to Become 'First Embodied Intelligence Stock'
Unitree Technology, a globally leading high-performance general-purpose robotics company, is set to have its initial public offering reviewed by the Shanghai St...
- Inside Altman’s pitch on government AI stake
OpenAI’s Sam Altman last year offered to cede equity in his firm to the Trump administration.
Score: 62🌐 MovesJun 8, 2026https://www.semafor.com/article/06/08/2026/inside-altmans-pitch-on-government-ai-stake - Anthropic Sounds Alarm on Recursive Self-Improvement
Anthropic Sounds Alarm on Recursive Self-Improvement The Information
Score: 62🌐 MovesJun 8, 2026https://www.theinformation.com/newsletters/ai-agenda/anthropic-sounds-alarm-recursive-self-improvement - Samsung Elec's chip chief says he discussed next-generation foundry with Nvidia CEO
Samsung Elec's chip chief says he discussed next-generation foundry with Nvidia CEO Reuters
- Uber and Wayve to Launch London’s First AI Robotaxis
The companies are also expanding into other major cities, including Tokyo.
Score: 62🌐 MovesJun 8, 2026https://aibusiness.com/generative-ai/uber-and-wayve-launch-london-s-first-ai-robotaxis - “앤트로픽 AI 서비스, 지나치게 비싸” MS AI 수장 비판
“남 탓하는 것 아닌가?”라는 반응이 나올 법하다. 마이크로소프트(MS)의 AI 부문 수장 무스타파 술레이만이 경쟁사 AI 서비스의 가격이 지나치게 비싸다고 비판한 시점이, 자사 깃허브 코파일럿의 사용량 기반 과금 체계가 개발자들에게 본격적인 비용 부담을 주기 시작한 때와 맞물렸기 때문이다. MS AI 최고경영자(CEO)인 무스타파 술레이만은 블룸버그와의 인터뷰 에서 “앤트로픽은 매우 비싸며, 많은 사람이 대안을 절실히 찾고 있다고 생각한다”라고 말했다. 현재 AI 업계에서는 서비스 비용 문제가 주요 관심사로 떠오르고 있다. 기업 내 다양한 부서에서 AI 기술 활용이 확대되고 있지만, 상당수 기업은 의미 있는 투자 대비 수익률(ROI)을 입증하는 데 어려움을 겪고 있기 때문이다. 이 같은 상황에서 MS는 이번 주 연례 개발자 행사인 빌드(Build)에서 새로운 AI 모델 7종을 공개하며 비용 경쟁력을 적극 부각했다. 회사는 보다 저렴한 AI 모델이 제공될 경우 더 많은 기업이 AI 프로젝트의 사업성을 확보할 수 있을 것으로 기대하고 있다. 시장조사업체 가트너는 2025년 보고서에서 현재 추진 중인 많은 AI 프로젝트가 2027년까지 중단될 수 있다고 전망했다. 업계에서는 보다 저렴한 AI 구축 방식이 이러한 문제를 해결할 수 있는 대안이 될 것으로 보고 있다. MS는 자사 AI 개발 도구가 앤트로픽 제품보다 더 높은 비용 효율성을 제공한다고 판단하는 것으로 보인다. 실제로 지난달에는 MS가 오는 6월 반기 계약 종료 시점에 맞춰 대부분의 클로드 코드(Claude Code) 라이선스를 해지하고, 엔지니어들을 자사 코파일럿 도구로 전환할 계획이라는 보도가 나온 바 있다 . dl-ciokorea@foundryco.com
- Perplexity Search as Code Lets AI Models Write Their Own Search Pipelines
Perplexity Search as Code lets AI models write their own search pipelines
Score: 60🌐 MovesJun 8, 2026https://opentools.ai/news/perplexity-search-as-code-ai-models-write-search-pipelines - Scoop: White House, Hill relaunch effort to block state AI laws
The White House is negotiating a federal preemption of some state AI laws in exchange for its support of key tech policy priorities from the Hill, Axios has learned. Why it matters: States are increasingly passing stronger AI laws and the Trump administration is feeling the heat to get something done. The talks are aiming to pair one of the tech industry's top priorities — overriding state AI laws — with legislation aimed at protecting kids online and combating deepfakes. Behind the scenes: Sen. Marsha Blackburn (R-Tenn.) is leading the negotiations, per a spokesperson, which include the Kids Online Safety Act and other tech-related measures. "Senator Blackburn is spearheading the negotiation with the White House to finalize legislative text of an AI preemption package that includes protections for kids, creators, and communities through the Senate version of KOSA, the NO FAKES Act, and age verification requirements," a Blackburn spokesperson said. The spokesperson said that the package is not "blanket pre-emption of all laws regulating AI or child safety." "The White House continues to proactively engage across government and industry," a White House official said. The big picture: The maneuvering between Congress and the White House shows that a bipartisan proposal from Reps. Jay Obernolte (R-Calif.) and Lori Trahan (D-Mass.) isn't the likely vehicle for AI policy in this Congress. That bill would preempt state AI laws for three years, formally establish the Center for AI Standards and Innovation and require certain developers to address risks prior to releasing models. Catch up quick: The last time the Trump administration tried to preempt the states, Republicans were inundated with pushback from advocacy groups and state lawmakers across the country. Blackburn's support, which the White House did not previously have, would be key for passage. The Obernolte-Trahan bill was also met with pushback from groups saying states should be free to regulate. Zoom out: Two sources told Axios the White House is also holding a meeting this week with AI companies to dig into what benchmarking should look like for the recent AI and cyber executive order. The revival of the preemption fight comes on the heels of Trump signing an AI and cyber executive order last week that includes voluntary pre-deployment testing of frontier models. The bottom line: This flurry of AI action will ultimately be hard to pull off as August recess in an election year nears.
Score: 60🌐 MovesJun 8, 2026https://www.axios.com/2026/06/08/white-house-hill-relaunch-effort-block-state-ai-laws - Intel lands chipmaking deal with Google, report says
Intel Foundry is trying to become the second-largest global contract chipmaker. Investors sent shares up more than 12% after the report.
Score: 60🌐 MovesJun 8, 2026https://www.bizjournals.com/sanjose/news/2026/06/08/intel-stock-boost-google-tensor-tpu.html?ana=brss_6150 - Google to Pay SpaceX $30B for AI Compute
The agreement is SpaceX's second major AI deal with a competitor.
Score: 60💰 MoneyJun 8, 2026https://aibusiness.com/generative-ai/google-pay-spacex-30-billion-ai-compute - Confidential submission of draft S-1 to the SEC
OpenAI confirms a confidential S-1 submission to the SEC and has not yet determined timing for further action.
- Waymo expands Arizona footprint: Alphabet subsidiary acquires massive northwest Valley property
The autonomous vehicle company plans to deploy its next-generation Ojai robotaxis in Phoenix. This expansion comes as the firm refines freeway operations.
Score: 58🌐 MovesJun 8, 2026https://www.bizjournals.com/sanjose/news/2026/06/08/waymo-arizona-proving-grounds-deal.html?ana=brss_6150 - 깃허브마저 사용량 기반 청구 시대로…코파일럿 기능 확장 나서
깃허브가 AI 코딩 도구 코파일럿의 활용 범위를 IDE 밖으로 확대한다. 새 데스크톱 애플리케이션과 협업 작업 공간인 ‘캔버스(Canvas)’를 선보이며, 코파일럿을 에이전트 네이티브(agent-native) 소프트웨어 개발의 컨트롤 타워로 자리매김시키겠다는 전략이다. 깃허브는 이번 주 열린 마이크로소프트(MS) 연례 개발자 행사 ‘빌드(Build)’에서 데스크톱 애플리케이션을 공개했다. 회사는 블로그를 통해 이 애플리케이션이 개발자가 소프트웨어 개발 전 과정에서 AI 에이전트와 협업할 수 있는 전용 환경을 제공하도록 설계됐다고 설명했다. 기존처럼 편집기 내부에서 코드 생성 작업에만 AI를 활용하는 수준을 넘어선다는 의미다. 애플리케이션에는 ‘캔버스’라는 협업 작업 공간도 포함됐다. 개발자는 이 공간에서 AI와 함께 아이디어를 구상하고, 요구사항을 구체화하며, 프로젝트 계획을 수립하고, 반복적인 개선 작업을 수행할 수 있다. 또한 새로운 ‘에이전트 머지(Agent Merge)’와 코드 리뷰 기능도 탑재됐다. 이를 통해 개발자는 여러 AI 에이전트의 작업을 결합해 특정 목표를 달성하도록 코파일럿을 자동화하거나, 사전에 설정한 기준에 따라 자율적으로 코드 검토를 수행하도록 할 수 있다. 시장조사업체 HFS리서치(HFS Research)의 CEO 필 퍼슈트 는 “이러한 신규 기능은 개발자의 작업 전환(context switching)을 줄이고 엔지니어링 효율성을 높이는 동시에 소프트웨어 제공 주기를 단축하는 데 기여할 수 있다”고 평가했다. 사용량 기반 과금 전환, 정당한 변화인가 하지만 최근 몇 주 동안 개발자 커뮤니티의 관심은 신규 기능보다 다른 이슈에 쏠려 있었다. 바로 깃허브가 지난 4월 발표하고 이번 주부터 적용한 코파일럿의 사용량 기반 과금 체계 전환이다. 이 같은 변화는 깃허브 커뮤니티 포럼 에서 거센 비판에 직면했다. 일부 사용자는 이를 “미끼 상품 판매 후 조건 변경(bait and switch)”이라고 비난했고, 일부는 환불을 요구하거나 구독 해지 계획을 밝히기도 했다. 그러나 업계 분석가들은 적어도 깃허브의 관점에서 보면 이번 가격 정책 변화가 필요하고 정당한 조치였다고 평가했다. 브로드컴(Broadcom)의 수석 신뢰성 엔지니어 아드바이트 파텔 은 “이번 가격 정책 변화는 현재의 제품이 아니라 깃허브가 지향하는 미래 방향에 근거한 것”이라며 “샌드박스 환경에서 여러 에이전트를 병렬로 운영하고, 캔버스 검토와 에이전트 머지를 지속적 통합(CI) 프로세스와 연계하는 작업은 IDE 플러그인보다 클라우드 컴퓨팅에 훨씬 가깝다. 컴퓨팅 자원을 정액 좌석 라이선스로 가격 책정할 수는 없기 때문에 사용량 기반 과금은 구조적으로 올바른 선택”이라고 설명했다. HFS리서치의 퍼슈트는 개발자와 CIO가 코파일럿을 단순한 코딩 보조 도구가 아닌 소프트웨어 개발 에이전트와 워크플로우를 조율하는 플랫폼으로 바라봐야 한다고 강조했다. 퍼슈트는 “이 변화는 투자 대비 효과(ROI)를 평가하는 방식을 크게 바꾼다”라며 “CIO는 더 이상 코파일럿을 좌석 라이선스 기반 생산성 도구로 볼 것이 아니라 AI 기반 소프트웨어 개발 플랫폼으로 평가해야 한다”고 말했다. 이어 “평가 지표 역시 ‘생성된 코드 라인 수’에서 벗어나 배포 속도, 코드 품질, 결함 감소, 엔지니어링 효율성 등 보다 폭넓은 운영 성과 중심으로 전환돼야 한다”고 설명했다. AI 에이전트가 기업 전반으로 확산되고 더 복잡하면서도 연산 집약적인 소프트웨어 개발 업무를 수행하기 시작하면서 가격 정책을 재검토하는 기업은 깃허브가 처음은 아니다. 지난 1년 동안 클로드 코드(Claude Code), 리플릿(Replit), 커서(Cursor), 키로(Kiro) 등 플랫폼도 사용자 반발에도 불구하고 가격 체계를 지속적으로 조정해 왔다. 증가하는 인프라 비용과 제한된 GPU 공급, 그리고 갈수록 고도화되는 AI 모델 및 에이전트 운영 비용이 주요 배경으로 꼽힌다. CIO의 과제는 ROI 입증 IT 컨설팅 기업 카네리카(Kanerika)의 최고분석책임자(CAO) 아밋 찬닥 은 이러한 공통된 압박 요인 때문에 개발자와 CIO가 깃허브의 과금 방식 자체보다 실제 비즈니스 가치 창출 여부에 더 집중해야 한다고 조언했다. 찬닥은 “깃허브가 발표한 신규 기능은 생산성을 크게 높이는 촉매제가 될 수 있지만, 반대로 이에 상응하는 비즈니스 가치를 제공하지 못한 채 사용량만 늘릴 가능성도 있다”라며 “도입 이전에 생산성 기준선을 설정하지 않으면 기업은 비용만 증가하고 그 비용이 실제 성과로 이어졌는지 확인하지 못하는 상황에 처할 수 있다”고 지적했다. 퍼슈트는 가격 체계가 변화하는 만큼 개발자와 CIO가 거버넌스, 모니터링, 재무 통제에 더욱 집중해야 한다고 강조했다. 그는 “거버넌스 문제는 매우 현실적인 과제”라며 “자율형 에이전트는 지속적으로 추론하고 테스트하며 수정 작업을 수행하는 동시에 여러 시스템과 상호작용한다. 이는 기존 SaaS 도구보다 훨씬 예측하기 어려운 사용 패턴을 만들어낼 수 있다”고 설명했다. 반면 파텔은 신규 기능이 아직 기술 프리뷰 단계라는 점을 고려할 때 사용자와 의사결정권자가 보다 신중한 태도를 취해야 한다고 조언했다. 파텔은 “고객은 아직 실제 운영 환경에서 검증되지 않은 가치에 대해 변동 요금을 지불하도록 요구받고 있다”라며 “새로운 기능이 더 높은 비용 지출을 정당화한다고 섣불리 가정해서는 안 된다”고 말했다. 이어 “90일간 파일럿 프로젝트를 진행한 뒤 투자 금액 대비 병합된 풀리퀘스트(PR) 수가 얼마나 늘어났는지 측정하고 데이터에 따라 판단해야 한다”라며 “그 비율이 개선됐다면 현재 가격은 합리적이다. 하지만 그렇지 않다면 기업은 실제 성과가 아니라 미래에 대한 기대에 비용을 지불하고 있는 것”이라고 덧붙였다. dl-ciokorea@foundryco.com
- A Security Raises $37 Million for Autonomous Offensive Security Platform
The company founded by Yossi Torati, Omer Gull, and Yuval Itzchakov has emerged from stealth mode. The post A Security Raises $37 Million for Autonomous Offensive Security Platform appeared first on SecurityWeek .
Score: 58🌐 MovesJun 8, 2026https://www.securityweek.com/a-security-raises-37-million-for-autonomous-offensive-security-platform/ - The new Siri won’t be available on iPhones in the EU, due to Digital Markets Act
The brand new iOS 27 Siri AI experiences will not be coming to users in the European Union at the same time as everyone else. In a company press release that directly blames the Digital Markets Act, Apple says iPhone and iPad users in the EU will have to wait longer for Siri AI. However, Siri AI will be launching on Mac, Apple Vision Pro and Apple Watch in the EU, as those platforms are not subject to the same gatekeeper requirements … more…
- Verizon CEO: AI will take over ‘a large percentage’ of customer service
Artificial intelligence “will dramatically improve our ability to satisfy customers,” Dan Schulman said.
Score: 58🌐 MovesJun 8, 2026https://www.hrdive.com/news/verizon-ceo-ai-take-over-a-large-percentage-of-customer-service/822219/ - VICTORY: Meta Strips Facial Recognition Code From Smart Glasses App After Public Outcry
Just days after a damning WIRED report exposed that Meta had quietly embedded facial recognition technology (FRT) code into millions of phones, the tech giant has quietly acquiesced in demands to reverse course. Last week, researchers identified code in Meta AI, a companion app for its line of smart glasses, that could convert images of faces into unique biometric signatures to identify strangers in public. EFF’s Threat Lab verified these findings through static analysis, and reminded consumers to think twice before buying or using Meta’s surveillance glasses. Just as quietly as Meta embedded this code, the app’s June 5th app update appears to have quietly removed all those features and systems. Gone is the face-recognition technology, the code meant to trigger “Person recognized” alerts, and the machine learning models and databases designed to detect, digitize, and store the biometric signatures of people users engage with. When WIRED broke the news last week, Meta’s executives immediately went on the defensive . Yet, their actions speak louder than their tweets: less than 48 hours after the public caught wind of their plans, Meta quietly launched an update to scrub nearly all traces of the FRT system from their app. But this quiet deletion of code does not equal a permanent change of heart. Meta previously used face recognition, and stopped only after it faced the legal and financial consequences . Now the company has refused to answer WIRED’s inquiries on whether it plans to bring the NameTag system back in the future, or what they did with any data they may have already collected during internal testing. There are billions of reasons not to turn Meta’s customers into a distributed surveillance machine. This whiplash behavior proves exactly why we cannot rely on the "good will" of Big Tech to protect our digital rights. We need robust, enforceable consumer privacy laws, complete with a private right of action that allows everyday people to sue companies that violate their biometric privacy. While we won this round, Meta's FRT ambitions probably aren't going away. EFF will keep watching.
- Why Google's SpaceX deal signals the rise of the AI compute landlord
Google's $920-million-a-month SpaceX agreement shows how AI compute is turning into a scarce, rent-generating infrastructure asset
- UK PM Starmer backs sovereign computing with $533 million AI chip investment
"We will use the power of public procurement to support British ingenuity," Starmer told the London Tech Week conference in a speech on Monday.
- Google is ordering millions of AI chips from Intel as TSMC strains to meet demand
The order comes as TSMC struggles to keep up with AI chip demand, prompting major tech companies to seek alternative manufacturers
- Alphabet taps Intel to make three million in-house chips, The Information reports
Alphabet taps Intel to make three million in-house chips, The Information reports Reuters
- Hackers likely hijacked over 20,000 Instagram accounts with Meta’s AI chatbot
Meta blames a bug on an exploit that allowed hackers to ask its AI support bot to link a victim’s account with their own email.
Score: 58🌐 MovesJun 8, 2026https://www.theverge.com/tech/945658/meta-ai-support-chatbot-exploit-instagram-accounts - Waymo bought Apple’s self-driving car proving ground for $220M
Waymo has acquired a massive 5,500-acre proving ground in Arizona owned by Route 14 Investment Partners LLC, a Delaware shell company associated with Apple, according to documents filed with Maricopa County.
Score: 58🌐 MovesJun 8, 2026https://techcrunch.com/2026/06/08/waymo-bought-apples-self-driving-car-proving-ground-for-220m/ - TCS bags multi-million Euro AI transformation contract from Canada Life
TCS bags multi-million Euro AI transformation contract from Canada Life Techcircle
Score: 58🌐 MovesJun 8, 2026https://www.techcircle.in/2026/06/08/tcs-bags-multi-million-euro-ai-transformation-contract-from-canada-life - Microsoft’s MAI Models: Big Tech Is Done Being OpenAI’s Customer
Microsoft spent billions to fund OpenAI, and now it’s quietly building the tools to replace them. Microsoft spent roughly $13 billion funding the company that arguably started the modern AI race. It gave OpenAI its cloud infrastructure, its enterprise distribution, and the credibility that comes with having one of the world’s most valuable corporations as your main backer. For years, every time someone used GitHub Copilot, Microsoft 365 Copilot, or Bing Chat, they were effectively renting intelligence from OpenAI. Photo by Simon Ray on Unsplash That era is over. At Build 2026 in San Francisco, Microsoft unveiled seven in-house AI models under a brand called MAI, and the message was unmistakable. CEO Satya Nadella called it a shift from “consuming a frontier model to fully participating at the frontier.” That one sentence quietly rewrites four years of corporate history. The Partnership That Made AI Mainstream (And Why It Couldn’t Last) To understand why the MAI launch matters, you have to understand just how lopsided the original arrangement was. Microsoft did not just invest in OpenAI. It became OpenAI’s engine. Azure was the exclusive cloud, Microsoft products were the primary distribution layer, and the relationship came with contractual constraints that limited how aggressively Microsoft could compete with OpenAI on its own model stack. Essentially, Microsoft had poured billions into building a supplier it could not easily replace or compete against. For a company with $37.5 billion in quarterly AI capital expenditure and a $25 billion AI revenue target for fiscal 2026, having a single-model dependency starts sounding less like a partnership and more like a structural risk. As one analysis put it: when your flagship AI product sits inside Microsoft 365, “single supplier” starts sounding like a vulnerability you have to explain on earnings calls. The relationship started shifting in late 2025. OpenAI restructured as a for-profit entity called OpenAI Group PBC, overseen by a non-profit parent called the OpenAI Foundation. The reset gave both sides room to breathe and, more importantly, room to compete. Then in April 2026, the restructuring went further. Microsoft and OpenAI announced a revised agreement that ended Azure exclusivity, scrapped the controversial AGI clause, and reset revenue sharing through 2030. The money flow also flipped in a significant way. Microsoft will no longer pay OpenAI a revenue share for distributing OpenAI models on Azure. OpenAI, for its part, retains a commitment to spend at least $250 billion on Azure services by 2032. Read that again. The customer became the landlord. Enter MAI: Seven Models, One Message The Build 2026 announcement was Microsoft’s most public declaration of AI independence yet. The flagship model, MAI-Thinking-1, is a reasoning model with 35 billion active parameters and a 256,000-token context window. Microsoft said it was trained from scratch with no distillation from other AI companies’ models, a detail meant to appeal to enterprise customers concerned about data lineage. That last part is not a minor footnote. Enterprise customers in healthcare, finance, and government have serious compliance requirements around where their AI was trained and on what data. The fact that MAI-Thinking-1 was trained on commercially licensed data with no refining from OpenAI or any other third-party model is a genuine competitive differentiator, not just marketing language. The benchmarks are also worth taking seriously. MAI-Thinking-1 scores 97.0% on AIME 2025 and 94.5% on AIME 2026, both benchmarks that test mathematical and multi-step scientific reasoning. On SWE-Bench Pro, Microsoft says the model matches Claude Opus 4.6 on coding tasks. For a company that was not even in the foundation model race 18 months ago, those numbers represent a remarkably fast ascent. In blind testing conducted by independent raters, MAI-Thinking-1 was preferred over Anthropic’s Claude Sonnet 4.6. Microsoft did not just ship a functional model. It shipped one that beats its competitors in the tests that enterprise buyers actually care about, at a fraction of the cost. The Coding Model Is Where It Gets Really Interesting If MAI-Thinking-1 is the strategic statement, MAI-Code-1-Flash is the tactical weapon, and it tells a more interesting story about how Microsoft thinks about model development. MAI-Code-1-Flash is a 5-billion-parameter coding model now rolling out inside GitHub Copilot. Five billion parameters is small. For context, frontier models are measured in hundreds of billions. But Microsoft did something clever with how it trained this model. Most coding models are trained on code datasets and then evaluated against Copilot-style workflows. MAI-Code-1-Flash was trained inside GitHub Copilot’s production harness, meaning the training distribution matches the exact patterns of real developer interactions, not academic code datasets. The result is a model that is not trying to be generally intelligent. It is trying to be exceptionally good at the specific thing developers do all day inside VS Code. And the benchmark results reflect that specialization. It posts a 16-point lead over Anthropic’s Claude Haiku 4.5 on SWE-Bench Pro, with higher pass rates on SWE-Bench Verified, SWE-Bench Multilingual, and Terminal Bench 2 as well. There is also a cost efficiency angle that matters enormously at enterprise scale. Microsoft says MAI-Code-1-Flash solves harder problems with up to 60% fewer tokens. When you are running a model inside a product used by millions of developers daily, 60% token reduction is not a small thing. It is the difference between a profitable product line and one that bleeds money on inference costs. The Full Stack Play MAI-Thinking-1 and MAI-Code-1-Flash get the most attention, but the broader MAI lineup reveals something important about Microsoft’s ambitions. The six other models in the MAI family cover a range of tasks. MAI-Image-2.5 and a flash variant handle both text-to-image and image-to-image generation and are already live in PowerPoint, with a rollout underway in OneDrive. MAI Transcribe 1.5 supports 43 languages, while MAI-Voice-2 and its flash variant add more than 15 new languages and new voice options. This is not a research lab releasing models for prestige. This is an enterprise software company instrumenting every surface of its product portfolio with its own intelligence layer. PowerPoint is getting MAI image generation. OneDrive is getting it. GitHub Copilot is running MAI-Code-1-Flash. This is vertical integration at a scale that no pure-play AI lab can match. The strategic picture that emerges is of a company preparing for a future in which AI capability is not rented from a partner but generated internally, at scale, across every layer of the stack. What Happens to OpenAI Now? Here is where the analysis gets complicated, because despite everything above, Microsoft and OpenAI are still deeply entangled. OpenAI still represents an estimated 45% of Microsoft’s cloud backlog, making the relationship deeply intertwined even as both parties diversify. OpenAI is simultaneously a partner powering Copilot products, a tenant spending $250 billion on Azure through 2032, and a competitor Microsoft is training itself to no longer depend on. Dan Ives at Wedbush Securities described it as “the most complex corporate partnership in technology history. Microsoft needs OpenAI for Azure revenue today, but they’re simultaneously building the capability to compete with OpenAI tomorrow.” That description is accurate, but it undersells the asymmetry that is developing. OpenAI needs Microsoft’s cloud. Microsoft no longer exclusively needs OpenAI’s models. The leverage is shifting. OpenAI is also facing pressure from the other direction. Amazon invested up to $50 billion and expanded its AWS agreement with OpenAI by $100 billion. Google has its own frontier models. Anthropic has Claude. The competitive moat that once made OpenAI the obvious default for every enterprise AI decision is narrowing fast. The AI Supply Chain Is Being Rebuilt What Microsoft is doing with MAI is part of a broader pattern playing out across the industry. Big Tech companies that initially moved fast by relying on external AI providers are now building in-house, for reasons that go beyond cost. Control over data lineage. Compliance guarantees. The ability to customize at the infrastructure level. The freedom to compete without contractual restrictions. These are not nice-to-have features. For the largest enterprises in the world, they are increasingly requirements. Microsoft had already started shipping in-house models before Build 2026, but the seven MAI models announced at Build are the most ambitious release yet: a full multimodal family spanning reasoning, code, image generation, transcription, and voice. Microsoft also has custom silicon in play. The Maia 200 AI accelerator chip is being developed specifically to support MAI inference, which means Microsoft is working toward owning the hardware layer too. At that point, the dependency on NVIDIA, on OpenAI, on external model providers, all of it shrinks to something manageable rather than existential. Mustafa Suleyman described Microsoft’s position at Build 2026 as a “best-of-both environment, where we’re free to pursue our own superintelligence and also work closely with them.” That framing is diplomatically careful, but the direction is clear. What This Actually Means for Developers and Enterprises If you use GitHub Copilot, MAI-Code-1-Flash is already rolling out to your model picker in VS Code, available across Free, Pro, Pro+, and Max tiers. You do not need to configure anything. Microsoft is quietly swapping what runs underneath. For enterprise buyers evaluating AI infrastructure, the MAI launch changes the conversation. The pitch used to be: “Use Azure to get access to OpenAI’s models.” The new pitch is: “Use Azure to get access to the best models, including ones with cleaner data lineage, lower inference costs, and compliance guarantees that OpenAI cannot offer.” That is a different value proposition, and a stronger one for regulated industries. For developers building on AI APIs, the competitive landscape is now more fragmented but also more interesting. Microsoft’s models are available on Azure AI Foundry and select third-party providers. You are no longer limited to choosing between OpenAI, Anthropic, and Google. A fourth credible option has arrived, and it comes with Microsoft’s enterprise distribution attached. The Bottom Line The narrative around Microsoft and OpenAI has always been framed as a partnership story. The $13 billion investment, the ChatGPT integration, the Copilot rollout, all of it positioned Microsoft as OpenAI’s most important ally. What Build 2026 revealed is that the partnership is becoming something more complicated: a coexistence between two companies that still need each other today, but are actively building the capability to need each other less tomorrow. Microsoft is not abandoning OpenAI. It is hedging. And when a company with Microsoft’s resources and distribution decides to hedge by building its own frontier AI lab from scratch, the implications ripple across the entire AI supply chain. The companies that assumed “enterprise AI” meant “buy from OpenAI, deploy on Azure” are going to need to reconsider that equation. Microsoft just proved that the equation has a new variable, and it has Microsoft’s name on it. Thanks for reading. If you found this useful, consider following for more analysis on AI strategy, Big Tech, and where the technology industry is actually heading. Microsoft’s MAI Models: Big Tech Is Done Being OpenAI’s Customer was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.
- Deputy PM to unveil AI labs to drag legal sector out of ‘analogue’ age
The UK government is launching a new AI growth project aimed squarely at dragging the UK’s billion-pound legal sector out of its “analogue” age. Ahead of London Tech Week, City AM can reveal that Deputy Prime Minister David Lammy will unveil ‘AI growth labs’, a regulatory safe space designed to cut through bureaucratic red tape [...]
Score: 58🌐 MovesJun 8, 2026https://www.cityam.com/deputy-pm-to-unveil-ai-labs-to-drag-legal-sector-out-of-analogue-age/ - Snowflake and 1Password tackle the growing challenge of securing AI agents at scale
As AI agents gain access to sensitive enterprise data, AI agent security is becoming a top priority for organizations. The challenge is no longer just protecting systems, but ensuring autonomous agents can be trusted, governed and controlled at scale. For the past two decades, enterprise security has focused primarily on protecting applications and infrastructure. Today, […] The post Snowflake and 1Password tackle the growing challenge of securing AI agents at scale appeared first on SiliconANGLE .
- Apple's Craig Federighi on Siri AI: 'We see Siri not as a separate chatbot, an unintegrated place you go and chitchat, but rather as an integral but conversational tool'
Apple's Craig Federighi on Siri AI: 'We see Siri not as a separate chatbot, an unintegrated place you go and chitchat, but rather as an integral but conversational tool' Tom's Guide
- Palladyne AI partners with IAI to bring Israeli Harpy, Harop drones to US
Palladyne AI partners with IAI to bring Israeli Harpy, Harop drones to US Breaking Defense
- AI cracked an Erdős math problem. Now experts want guardrails
The result is correct but challenges core norms of mathematics: checking proofs, crediting ideas and keeping research open to everyone.
- Amazon launches AI image generator to narrow search queries
Users describing a product in the e-commerce giant’s app can now choose from a variety of AI-generated images to shop similar items.
Score: 56🌐 MovesJun 8, 2026https://www.retaildive.com/news/amazon-launches-ai-image-generator-search-bar-queries/822231/ - Nvidia touts its contribution to UK sovereign AI plans
Nvidia touts its contribution to UK sovereign AI plans IT Pro
- Tech Rivals Unite To Stop AI-Designed Bioweapons
AI leaders are worried that powerful AI models could be used to design highly contagious viruses that could be ordered online from gene-synthesis providers.
Score: 55🌐 MovesJun 8, 2026https://www.forbes.com/sites/craigsmith/2026/06/08/tech-rivals-unite-to-stop-ai-designed-bioweapons/ - Stock markets fall as concerns persist over tech firms at heart of AI boom
Drops follow sharp sell-off of US tech stock last week while oil prices seesaw after Iran and Israel exchange strikes Stock markets have fallen amid concern about the prospects for tech stocks, while oil prices have risen after renewed conflict in the Middle East dampened hopes that the strait of Hormuz would soon reopen. Markets in Asia and Europe fell on Monday after a sharp sell-off in US tech stocks late last week, as investors fretted over how firms at the forefront of the artificial intelligence boom would fund their “eye-watering” spending plans. Continue reading...
- BYD Secretly Develops Humanoid Robot Codename 'Yao-Shun-Yu' as Auto Giants Race Into Embodied AI
BYD Secretly Develops Humanoid Robot Codename 'Yao-Shun-Yu' as Auto Giants Race Into Embodied AI BYD, China's largest electric vehicle manufacturer, has confirmed it is secretly developing humanoid robots under a project codenamed "Yao-Shun-Yu." The revelation came from BYD Executive Vice President Li Ke in a recent interview, shedding light on the automaker's ambitions beyond electric vehicles and into the rapidly emerging field of embodied AI. The project was initiated in 2022 and operates under BYD's 15th Business Unit, which focuses on electronic integration and intelligence.
Score: 55🌐 MovesJun 8, 2026https://pandaily.com/byd-secretly-develops-humanoid-robot-codename-yao-shun--jun2026 - Goldman, JPMorgan Explore Trading Compute Futures as AI Financing Hedge
Goldman, JPMorgan Explore Trading Compute Futures as AI Financing Hedge The Information
Score: 55🌐 MovesJun 8, 2026https://www.theinformation.com/articles/goldman-jpmorgan-explore-new-ways-tame-ai-lending-risks - SK chief vows to partner with Nvidia on future AI factory
SK chief vows to partner with Nvidia on future AI factory 매일경제
- Cosine secures industry backing for Britain’s first sovereign frontier model
Cosine, the British AI company selected by the UK Government as part of its Sovereign AI initiative, today announced that it has brought together a coalition of leading UK institutions to co-design Lu...
Score: 55💰 MoneyJun 8, 2026https://tech.eu/2026/06/08/cosine-secures-industry-backing-for-britain-s-first-sovereign-frontier-model/ - Linux Foundation targets AI’s cost-management problem with Tokenomics Foundation
For many CIOs, the challenge of scaling AI is no longer about building applications but about understanding what they cost. With AI models priced through complex token-based structures, enterprises deploying multi-agentic AI are facing a fast-growing and often opaque expense, making it harder to benchmark providers, measure efficiency, and prove returns on AI investments. Seeking to address that problem, the Linux Foundation has announced its intent to launch the Tokenomics Foundation, a vendor-neutral organization that will develop open standards, benchmarks, and best practices for AI cost management. The foundation, instead of working alone, will collaborate with the FinOps Foundation, another Linux Foundation initiative focused on advancing the discipline of cloud financial management and technology value management. A key part of that collaboration will be the expansion of FinOps Open Cost and Usage Specification (FOCUS ), an open standard originally developed to normalize cloud spending and usage data across providers. The two foundations plan to adapt the specification to account for token-based AI consumption, creating a common framework for measuring, comparing, and governing AI costs across models and platforms, the Linux Foundation said in a statement. While the Tokenomics Foundation’s Governing Board will help set strategic priorities and allocate funding for the initiative, its Technical Committee will work with FinOps Foundation contributors to develop the specifications, benchmarks, and reference frameworks needed to incorporate AI token usage into FOCUS, it added. The new foundation, the Linux Foundation further said, has already attracted support from a broad group of enterprise technology vendors, cloud providers, and end-user organizations, including Accenture, Booking.com, Flexera, Google Cloud, IBM, JPMorganChase, KPMG, Microsoft, Oracle, Salesforce, SAP, and ServiceNow. Why will the new foundation matter for CIOs? The move to launch a new foundation to standardize metrics around token usage addresses a growing gap in enterprise AI rollouts in production use cases. “The industry lacks common standards for measuring AI costs and efficiency. That makes it difficult for CIOs to compare models and vendors, in turn making it difficult to track ROI, optimize spending, or decide which models deliver the best value,” said Pareekh Jain , principal analyst at Pareekh Consulting. “A neutral body, in contrast, can help define shared terminology, benchmarks, and accounting frameworks, similar to how FinOps brought discipline to cloud spending. Standardization can make AI costs more transparent and comparable across vendors for CIOs,” Jain added. The challenge is not just about managing AI costs but also about how enterprises architect, deploy, and govern AI applications at scale, even as per-token pricing continues to decline, pointed out Yugal Joshi , partner at Everest Group. “The gross AI bill for enterprises continues to rise due to poor design practices such as throwing unwanted models at specific problems. In addition, the same workflow ends up costing differently based on input queries and user prompts,” Joshi said. “No CIO can make meaningful budgeting decisions or calculate RoI of such workflows. This creates complexity around selecting workloads that can derive the best outcome with the least investment.” The Tokenomics Foundation’s emphasis on standards and best practices could help CIOs make more informed decisions about workload design, Joshi noted. Another area of focus? Cost governance and design best practice, however, are only one part of the equation. Another challenge facing CIOs is determining when open-source models make more economic sense than proprietary alternatives, Jain said. “When enterprises use models such as Llama or Mistral on their own infrastructure, costs shift from paying per token to paying for GPUs, electricity, and infrastructure utilization,” Jain pointed out. “The long-term challenge for the Tokenomics Foundation will be creating a common framework that lets CIOs compare the true cost of self-hosted models with commercial AI APIs. Determining the point at which it becomes cheaper to run a private AI platform rather than consume AI as a service could become one of the most important financial decisions in enterprise AI,” Jain added. While the initiative has been announced, the Linux Foundation has not provided a target date for the Tokenomics Foundation’s formal launch or operational rollout.