Good morning. It's Monday, March 9, and we're covering ChatGPT-5.4, the AI data-center land rush, drone strikes hitting AWS cloud infrastructure, and more.

New here? Keep up with the future of tech, sign up here. If you liked this email, share it with a friend.

Iran Targets Gulf Cloud Infrastructure (Paywall): Drone strikes hit Amazon Web Services data centers in the UAE and Bahrain, disrupting services and raising alarms as the first known military attack on major US hyperscale cloud facilities.

AI Helps North Koreans Land Remote Jobs: Microsoft says state-backed operatives use AI voice changers, face swaps, and fake identities to secure Western IT roles and funnel salaries to Pyongyang, expanding a long-running cyber-funding scheme.

OpenAI Robotics Lead Quits Over Pentagon Deal: Robotics chief Caitlin Kalinowski left OpenAI after the company struck a U.S. Department of Defense partnership.

AWS Launches AI Agents for Healthcare: Amazon Web Services unveiled Amazon Connect Health, a HIPAA-eligible platform using AI agents to automate scheduling, documentation, and patient verification.

SoftBank Seeks $40B Loan for OpenAI: SoftBank is arranging a record $40 billion bridge loan, led by lenders including JPMorgan, to finance a major investment in OpenAI—marking the largest dollar-denominated borrowing in the company’s history.

Claude Usage Surges After Pentagon Rift: Anthropic’s Claude app hit 11.3M daily mobile users and 149K daily U.S. downloads after refusing Pentagon surveillance and weapons use—suggesting its stance boosted consumer adoption.

Claude Finds 22 Firefox Vulnerabilities: Anthropic used Claude Opus 4.6 to audit Mozilla’s Firefox codebase for two weeks, uncovering 22 bugs—14 high-severity.

City Detect Raises $13M for AI Monitoring: The startup uses computer vision cameras on city vehicles to detect graffiti, dumping, and building damage at scale. It operates in 17 cities as governments seek faster ways to tackle urban blight.

Anthropic Launches AI App Marketplace: The company unveiled Anthropic Marketplace, letting enterprise customers buy third-party software built on Claude from partners like Snowflake, Harvey, and Replit.

OpenAI’s new unified AI model combines coding, reasoning, and agents with a 2M-token context window, strong benchmarks, and real-world demos.

While major tech companies pour hundreds of billions into AI infrastructure, Apple is taking a markedly different approach. Amazon, Google, Microsoft, and Meta are collectively spending an estimated $650 billion on AI data centers and compute, while Apple’s annual capital spending is roughly $14 billion—essentially flat year over year.

Instead of building massive AI server capacity, Apple is licensing models such as Google’s Gemini and focusing on running AI directly on its devices through custom chips like the M-series. The strategy reflects a bet that controlling the customer ecosystem—and distributing AI across billions of devices—may prove more durable than competing in the capital-intensive race to build the world’s largest AI infrastructure. → Continue reading here.

A fictional scenario circulating online illustrates how legally purchasable data—credit card transactions, cell tower pings, browsing logs, and call metadata—could be combined by AI to build a detailed personal dossier in seconds. The example illustrates tracking at a massive national scale, utilizing warehouses of data containing detailed records of roughly 330 million people to infer relationships, health concerns, travel plans, and potential political activity.

None of the data is hacked or obtained through warrants; the scenario emphasizes that many datasets are already commercially available through data brokers. The exercise highlights growing concerns among privacy advocates that AI could dramatically amplify mass surveillance by automating the analysis of vast consumer data streams. → Read the full article here.

The Recap: As demand for artificial intelligence infrastructure surges, companies like Cloverleaf Infrastructure are acting as intermediaries that secure electricity and land for massive data centers before selling those sites to tech developers. The firm, founded in 2024 by former Microsoft energy executive Brian Janous, has raised $300 million and is assembling “powered land” for projects tied to companies like OpenAI, Oracle, and Meta. The business opportunity is fueled by a looming power shortage as AI companies race to build facilities requiring tens of gigawatts of electricity.

Highlights:

Forward Future Takeaways:
The AI boom is creating a new infrastructure economy, where electricity access—rather than chips alone—has become a critical bottleneck. Companies that can secure power, permits, and land quickly are emerging as pivotal intermediaries in the AI supply chain. → Read the full article here.

A research paper by Judah Goldfeder, Philippe Wyder, Yann LeCun, and Ravid Shwartz-Ziv argues that the widely used concept of Artificial General Intelligence (AGI) is poorly defined and may be the wrong goal for AI development. The authors contend that humans themselves are not truly “general,” but highly specialized at tasks relevant to survival, making AGI an unrealistic benchmark.

Instead, they propose Superhuman Adaptable Intelligence (SAI)—AI systems that exceed human performance on important tasks while rapidly learning new ones. The framework emphasizes adaptability, self-supervised learning, and world models as core mechanisms for building powerful AI systems. → Read the full paper here.

Krea: Generate images and videos with AI, real-time editing, and custom style training.
ModiFace: AR beauty tech enabling virtual makeup, hair, and nail try-ons with AI skin analysis.
Pythagora: Build full-stack web apps with AI using plain language, automated backend, and deployment.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team