Eight developments in AI compute infrastructure, developer tool economics, cybersecurity disclosure, and frontier valuations.

NVIDIA – the semiconductor company whose GPUs power the majority of AI data centers worldwide – used its Computex 2026 keynote in Taipei on 01 June to announce that the Vera Rubin platform, its next-generation AI factory architecture, has entered full production. The platform unifies five purpose-built rack types into a single POD-scale system: Vera Rubin NVL72 GPU racks, Vera CPU racks, Groq 3 LPX networking, Vera BlueField-4 STX storage, and Spectrum-6 SPX Ethernet racks. NVIDIA claims 10x agent throughput at scale compared with the previous-generation Grace Blackwell platform. The Rubin GPU features two reticle dies delivering 50 PFLOPs of NVFP4 inference and 35 PFLOPs of NVFP4 training performance – 5x and 3.5x increases over Blackwell, respectively – along with HBM4 memory (the latest generation of high-bandwidth memory used in AI accelerators) offering up to 22 TB/s bandwidth per chip, a 2.8x jump over Blackwell.

The production ramp involves hundreds of ecosystem partners – 150 in Taiwan alone – across more than 350 factories in 30 countries. Major system builders in full-scale production include Dell Technologies, HPE, Lenovo, Supermicro, ASUS, Foxconn, GIGABYTE, IBM, Inventec, MSI, and Wistron. NVIDIA also introduced Spectrum-X Ethernet Photonics – now in production – combining co-packaged optics with Spectrum-X switching to enable million-GPU AI factories. Vera Rubin NVL72 racks are 100% liquid cooled with cable-free modular tray designs, reducing installation time from two hours to five minutes compared with Blackwell equivalents. Early Vera CPU customers confirmed by CNBC include Anthropic (the company that makes the Claude AI assistant), OpenAI (the company behind ChatGPT), SpaceX’s xAI, Dell, Oracle, and CoreWeave.

At the same Computex keynote, NVIDIA CEO Jensen Huang unveiled RTX Spark, NVIDIA’s first-ever processor built specifically for Windows PCs. The Arm-based superchip packs a 20-core Grace CPU (developed alongside MediaTek, the Taiwanese chip design firm), a Blackwell-class RTX GPU with 6,144 CUDA cores, an NPU (neural processing unit – a dedicated chip for running AI tasks efficiently on-device), and up to 128 GB of unified LPDDR5X memory – all on a single TSMC 3 nm package. NVIDIA claims 1 petaFLOP of AI performance from the chip, which it also refers to internally as the N1X.

RTX Spark systems will debut in the fall of 2026, with over 30 laptops and 10 desktops announced from Microsoft (Surface Laptop Ultra), Dell, HP, ASUS, Lenovo, and MSI. NVIDIA says the chip’s GPU is “roughly equivalent” to the RTX 5070 laptop GPU. Alongside RTX Spark, NVIDIA announced DLSS 4.5 Ray Reconstruction, expected in August 2026, and confirmed it is working with all major game developers to bring popular anti-cheat software natively to Windows on Arm. The move marks NVIDIA’s first entry into the PC processor market in over a decade and represents a direct challenge to Apple’s M-series silicon, Qualcomm’s Snapdragon X platform, and the long-standing Intel/AMD x86 duopoly in Windows PCs.

GitHub Copilot’s new usage-based billing system took effect on 01 June 2026, replacing the flat subscription model with token-based “GitHub AI Credits.” Under the new model, monthly subscription prices remain nominally unchanged – Pro at $10/month, Pro+ at $39/month, Business at $19/user/month, Enterprise at $39/user/month – but each plan now purchases a fixed allotment of AI Credits (1 credit = $0.01) that are consumed based on actual token usage across inputs, outputs, and cached context. Code completions and Next Edit suggestions remain included at no credit cost, but chat, agentic coding (where the AI takes multi-step actions autonomously), and code review sessions now consume credits at model API rates.

The change has generated significant developer backlash. TechCrunch reported that developers have flooded Reddit and X with complaints, with some reporting projected cost increases from $29/month to $750/month and from $50/month to $3,000/month. The previous fallback mechanism – which switched users to a lower-cost model when premium requests were exhausted – has been eliminated. GitHub is offering promotional credits for June, July, and August to ease the transition and has introduced pooled usage for organizations. Annual plans are being retired; annual subscribers will remain on existing terms until expiration. GitHub CPO Mario Rodriguez wrote that the change was necessary because Copilot now powers “far more complex, agentic workflows that consume far more compute.” The timing is notable: it coincides with reports that Claude Code – Anthropic’s command-line coding tool – has overtaken Copilot as the dominant developer AI tool, per The Information.

On 01 June 2026, Microsoft’s Security Response Center (MSRC) issued a public statement on X clarifying that it has “no intention to pursue action against individuals conducting or publishing their security research.” The statement was a direct response to widespread backlash triggered by Microsoft’s 28 May MSRC blog post, which condemned an anonymous researcher known as Nightmare Eclipse for releasing six unpatched Windows zero-day exploits – previously unknown vulnerabilities with no available fix – between April and mid-May 2026. The blog post was widely interpreted as a blanket legal threat against the security research community.

Nightmare Eclipse released working proof-of-concept exploit code for six vulnerabilities targeting Windows Defender (Microsoft’s built-in antivirus software) and BitLocker (Microsoft’s drive encryption tool): BlueHammer (CVE-2026-33825), RedSun (CVE-2026-41091), UnDefend (CVE-2026-45498), YellowKey (CVE-2026-45585), GreenPlasma, and MiniPlasma. At least three of these – BlueHammer, RedSun, and UnDefend – have been confirmed as actively exploited in real-world attacks, a fact corroborated by CISA’s (the Cybersecurity and Infrastructure Security Agency) addition to its Known Exploited Vulnerabilities catalog. Microsoft’s patches are not expected until the 09 June Patch Tuesday cycle; no out-of-band release has been committed. Nightmare Eclipse, who moved to GitLab after GitHub removed their account on 23 May, has threatened additional disclosures for 14 July.

Microsoft is preparing to unveil a suite of in-house AI models at its Build developer conference opening 02 June in San Francisco, including a coding model designed to power GitHub Copilot and compete directly with Anthropic’s Claude Code, per reporting from The Information and Reuters. The homegrown models include MAI-Image-2.5 (text-to-image generation in standard and fast variants with image editing support), MAI-Voice-2 (multilingual text-to-speech across 15+ languages with expanded emotional range), and MAI-Transcribe-1.5 (upgraded speech-to-text claiming the lowest word error rate across 25 languages).

The models will target Azure AI Foundry, Copilot, Teams, and Windows integration. The coding model specifically is aimed at reclaiming GitHub Copilot’s competitive position after Claude Code reportedly overtook it as the dominant developer AI tool. Microsoft CEO Satya Nadella’s opening keynote on 02 June at 12:30 PM ET is expected to frame Windows as a platform for AI agents. The MAI model family was formed ~six months ago under Mustafa Suleyman’s AI division and has already released transcription, audio generation, and image generation models. Build 2026 marks the most significant public debut of the in-house stack.

Anthropic – the company that makes the Claude AI assistant – is expected to close a funding round of at least $30 billion at a pre-money valuation exceeding $900 billion, according to Bloomberg reporting from 22 May citing unnamed sources. If finalized at these terms, the round would make Anthropic the world’s most valuable AI startup, surpassing OpenAI’s $852 billion valuation from its March 2026 round. The company expects to more than double its quarterly revenue to $10.9 billion in Q2 2026 and projects its annualized run rate to top $50 billion by the end of June, per the same Bloomberg report.

The deal has not been finalized and no term sheet has been signed as of the most recent reporting. Anthropic was valued at $380 billion in its February 2026 Series G round, in which it raised $30 billion led by GIC (the Singaporean sovereign wealth fund) and Coatue (a technology-focused investment firm). Google has previously committed $10 billion at a $350 billion valuation, with an additional $30 billion contingent on specific technical milestones. The company is simultaneously exploring a potential IPO as early as October 2026.

Beijing-based startup Vast raised ~$200 million in a funding round that values the company at $1 billion or more, according to Bloomberg. Ince Capital and a venture fund backed by China Life Insurance Co. led the financing. Other participants included Genesis Capital and existing investors Eminence Ventures and Primavera Venture Partners. The company had previously raised a $50 million Series A in March 2026 led by Alibaba Group (the Chinese e-commerce and cloud computing conglomerate) alongside Hengxu Capital, with Baidu Ventures (the venture arm of China’s largest search engine company) also participating.

Vast specializes in AI-powered 3D modeling, offering a product called Tripo AI that converts text and image prompts into 3D objects. The company claims 20 million global users and counts NetEase (a major Chinese gaming company) and Sony among its clients. Founded in 2023 by a 29-year-old gamer, the startup’s rapid ascent to unicorn status reflects the intensifying competition in China’s generative AI sector, particularly in spatial computing and 3D asset creation for gaming, film, and e-commerce applications.

SpaceX – the space launch and satellite company founded by Elon Musk – made its S-1 registration statement public on 20 May 2026, laying bare the financials of the combined SpaceX/xAI entity ahead of what would be the largest IPO in history. The company is targeting a listing on the Nasdaq under ticker SPCX as early as 12 June, with the investor roadshow planned to begin the week of 04 June and pricing expected on 11 June, per Reuters and the Wall Street Journal. The target valuation is $1.75 trillion – up from $1.25 trillion set when SpaceX merged with Musk’s xAI (the company behind the Grok AI models) in February 2026 – with a planned raise of more than $70 billion.

The S-1 reveals that SpaceX’s full-year 2025 revenue reached $18.67 billion, up 33% Y-o-Y. Starlink (SpaceX’s satellite broadband service) generated $4.4 billion in operating profit, but the xAI segment lost $6.4 billion in the same period, and the Space segment lost $619 million. Q1 2026 cash reserves declined to ~$16 billion from $25 billion, with total liabilities at $60.5 billion, primarily driven by the xAI merger. SpaceX plans to earmark a large share for retail investors and will host 1,500 retail investors at a dedicated event, with participants from the US, UK, EU, Australia, Canada, Japan, and South Korea. Morgan Stanley, Bank of America, Citigroup, JP Morgan, and Goldman Sachs are leading the deal.