AI News Roundup: Pentagon Taps 7 AI Giants for Classified Networks, Big Tech’s $700B Bet, Anthropic Passes OpenAI

Pentagon Clears Seven Tech Giants for Classified AI Networks

The Department of Defense announced agreements with Amazon Web Services, Google, Microsoft, NVIDIA, OpenAI, SpaceX, and Reflection to deploy AI on its classified Impact Level 6 and Impact Level 7 networks. Oracle was added shortly after, bringing the total to eight. Over 1.3 million defense personnel already use the department’s official AI platform, GenAI.mil.

The contracts focus on streamlining data synthesis, improving warfighter decision-making, and elevating situational awareness. The Pentagon described the move as building “an architecture that prevents AI vendor lock and ensures long-term flexibility for the Joint Force.”

Notably absent: Anthropic, which was shut out after refusing to provide unrestricted access to Claude for “all lawful use,” citing concerns over potential applications in mass surveillance and autonomous weapons. The exclusion deepens an already-public rift between Anthropic and the Trump administration over AI safety guardrails in defense settings.

Big Tech’s AI Spending Crosses $700 Billion

The four largest cloud hyperscalers — Amazon, Microsoft, Alphabet, and Meta — now plan to spend as much as $725 billion in capital expenditures this year, up sharply from roughly $410 billion in 2025. The spending surge is almost entirely directed at AI data centers, semiconductors, and power infrastructure.

Alphabet and Meta both raised their full-year capex guidance during Q1 earnings, while Microsoft disclosed its first full-year estimate at $190 billion, matching Alphabet. Amazon held steady at $200 billion. Meta’s stock slid despite beating earnings expectations, as investors balked at its $125–$145 billion AI spending plan. Some analysts warn of a potential overbuild, but all four companies signaled sustained investment as AI demand continues to accelerate.

Anthropic Hits $30B Run Rate, Officially Passes OpenAI

Anthropic reached $30 billion in annualized revenue as of early April, overtaking OpenAI’s $25 billion and marking roughly 1,400% year-over-year growth. For context, Anthropic’s run rate was $1 billion in January 2025, $4 billion by June, $9 billion by December, and $30 billion just four months later.

Enterprise adoption is the engine: more than 1,000 business customers now spend over $1 million annually on Claude, a figure that doubled in two months. Claude Code — the company’s coding agent launched publicly in mid-2025 — hit $1 billion in annualized revenue within six months, with enterprise usage now accounting for more than half of Claude Code’s total revenue. Google also committed to invest up to $40 billion more into Anthropic, with an immediate $10 billion tranche and future milestone-based commitments.

NIST Benchmarks DeepSeek V4 Pro: Capable, Cheap, Eight Months Behind

The Center for AI Standards and Innovation (CAISI) at NIST published its evaluation of China’s DeepSeek V4 Pro, the most capable open-weight model from a Chinese lab to date. Using 16 benchmarks across 35 models, CAISI found that DeepSeek V4 performs similarly to GPT-5 — placing it roughly eight months behind the current U.S. frontier.

Where DeepSeek V4 stands out is cost. Compared to GPT-5.4 mini (the most cost-competitive U.S. reference model), DeepSeek V4 was cheaper on five of seven benchmarks, ranging from 53% less expensive to 41% more expensive depending on the task. The evaluation covered cyber, software engineering, natural sciences, abstract reasoning, and mathematics.

EU AI Act Reform Talks Collapse, August Deadline Holds

Trilogue negotiations to reform the EU AI Act under the Digital Omnibus for AI broke down after 12 hours of talks on April 28–29. The Parliament and Council could not agree on how the AI Act’s high-risk system rules interact with existing sectoral safety law (Machinery Regulation, MDR, IVDR). Talks will resume next month.

The stakes are significant. The Omnibus was designed to push the compliance deadline for high-risk AI systems from August 2, 2026 to December 2, 2027 for stand-alone systems, and to August 2028 for systems embedded in regulated products. Without it, companies face the original August deadline — now just three months away.

Google I/O 2026: Gemini 4, 42.5-Exaflop TPUs, and AI Glasses

Google I/O 2026 is set for May 19–20 at Shoreline Amphitheatre, and the headline is Gemini 4, which is expected to support a 2-million-token context window and score 84.6% on ARC-AGI2. Beyond the model, Google is unveiling Ironwood TPUs pushing 42.5 exaflops, AI glasses built in partnership with Warby Parker, and a robotics deal that puts Gemini inside Boston Dynamics’ Atlas.

Perhaps most notable is Personal Intelligence, which connects Gemini to Gmail, Photos, Drive, Calendar, YouTube history, and Search data. Google plans to roll it out to 2 billion users across 200+ countries, making it the largest personal-AI deployment to date.