Edition: August 17, 2026

Nvidia is committing $1.5 billion to SB Energy, a SoftBank‑backed data‑center developer tied to OpenAI, to become the exclusive supplier of compute hardware for the new Ports‑Pike facility near Cincinnati. The deal also includes up to $105 billion in credit to help build a massive 9.2‑GW natural‑gas power plant on former DOE land, highlighting the soaring costs and energy demands of AI infrastructure.

Anthropic’s annualized revenue run rate jumped to $65 billion by July, up from $47 billion in May and $9 billion a year earlier, fueling expectations it could reach $100‑$120 billion by year‑end and pursue a $2 trillion IPO valuation. The rapid growth outpaces rival OpenAI, whose revenue hit $40 billion, and positions Anthropic for the largest market debut ever.

Researchers at Dharma‑AI introduced a constraint‑aware GPU allocator that outperforms traditional FIFO scheduling, boosting GPU utilization by up to 33 percentage points and priority‑weighted output by as much as 105% across seven benchmark scenarios. By intelligently ordering allocation decisions for mixed workloads—training, batch inference, real‑time inference, and quantization—the new scheduler reduces idle reservations and improves overall cluster efficiency without any hardware changes.

Flock, a major police‑tech firm operating 120,000 license‑plate readers, announced new safeguards—like requiring a case number for searches—to curb officer misuse after reports of stalking and harassment. Critics say the measures are weak, allowing bogus case numbers and failing to address broader privacy concerns, fueling a backlash that’s prompting contract cancellations and legislative action.

Amazon is reportedly acquiring rare books, removing their spines and digitizing them to feed large language models, according to a 404 Media investigation that tracked a book to Amazon’s Las Vegas facility. The company says the purchases are to improve customer services, but the practice raises concerns about the use of scarce, out‑of‑print works for AI training and potential copyright violations.

Groq has secured $350 million in new funding to accelerate its shift from building AI chips to operating a "neocloud" AI inference service, valuing the startup at $3.5 billion. The capital will fund expansion of its data‑center capacity to over 200 MW by 2027, positioning Groq alongside Nvidia‑backed rivals in the fast‑growing AI inference market.

Wispr, the AI dictation startup, secured $280 million in a Series B round at a $2 billion valuation to expand beyond transcription into meeting note‑taking and new interfaces. The funding backs its new Canto model, which cuts error rates dramatically, and fuels growth in markets like India and the UK while it battles rising competition.

The new DiG‑bench suite of 70 text‑based games tests AI’s ability to infer hidden rules through exploration, revealing that current frontier models still struggle with these discovery tasks. Developed by researchers from top institutions, the benchmark offers tiered difficulty and private games to prevent training leakage, highlighting a gap in AI’s intuitive and creative reasoning.

Geochemists revisiting data from Ontario's Kidd Creek mine estimate the site releases about 140 metric tons of natural hydrogen annually, enough to power a significant portion of the mine if captured. While the amount is modest, it demonstrates the feasibility of extracting geologic hydrogen—a potentially zero‑carbon fuel—amid growing global interest and investment in underground hydrogen sources.

A ten‑year‑old neurodivergent boy, Xander, relies on Moxie, an AI‑powered social robot, for companionship and therapeutic support, illustrating the growing market of AI toys designed to aid children with autism. The piece highlights both the promise of such devices in providing constant, personalized interaction and the looming reality that these robots, like Moxie, have limited lifespans, raising questions about long‑term reliance.

A MIT Technology Review newsletter highlights two pressing tech issues: the shutdown of the Moxie robot, a therapeutic companion for neurodivergent children, exposing the fragility of AI‑powered toys; and the rise of the “censorship‑industrial complex,” a theory that government, academia, civil society and Big Tech are colluding to curb conservative speech, now influencing US policy. Both stories underscore how AI’s integration into daily life and governance can have profound social consequences.

AI workflow startup Relay is shutting down, with its founder Jacob Bank and other staff moving to Google to lead Chrome product and developer relations. Bank, a former Google executive, aims to embed AI tools into Chrome, building on Google’s recent Gemini integration. The closure highlights the challenges of AI‑focused automation platforms and Google’s push to deepen AI in its browser.

Anthropic detailed its plan to embed invisible watermarks in Claude-generated text, using a version of Google DeepMind’s open‑source SynthID‑Text technique, to meet EU AI Act transparency requirements. The move also adds C2PA support for images, ensuring machine‑readable marks across synthetic media.

Microsoft has delayed the release of a major update to its Exchange Server Subscription Edition due to the extra work created by AI bug-finders, with no release date announced. The company is prioritizing security above all else, following previous flaws in Exchange that led to a major attack. The delay is attributed to the need to ensure the update is stable and inclusive of all security updates.

Google’s Gemini AI and Pixel smartphones have struck long‑term partnership deals with five top football clubs—Arsenal, Barcelona, Bayern Munich, Liverpool and Paris Saint‑Germain—to deliver AI‑driven match insights and exclusive behind‑the‑scenes content, including for women’s teams. Fans will be able to use Gemini for real‑time stats and Pixel’s camera tools for unique media, aiming to deepen engagement and boost visibility of women’s football.

Mozilla has begun rolling out an optional ad‑blocking feature in Firefox for iOS, using an EasyList‑based filter to block most third‑party ads, trackers, pop‑ups and overlays, though it leaves search‑engine ads and sponsored content untouched. The feature is off by default and currently experimental, offering a convenient toggle but less comprehensive coverage than dedicated blockers.

Microsoft announced it will retire the Excel COPILOT() function on September 14, 2026, ending a preview that let users invoke AI directly from a worksheet cell. The company says the Copilot side pane will provide the same capabilities, making the dedicated function redundant. Users will need to replace any formulas that relied on the now‑removed function.

The Verge reviews Whisker’s Litter‑Robot 5 Pro, a high‑end automatic litter box that uses dual cameras and AI to recognize individual cats and monitor health‑related litter habits. While the tech promises detailed insights, the piece treats it as a novelty for cat owners rather than a transformative AI breakthrough.