Good morning. It's Tuesday, April 28, and on deck: AI energy breakthroughs, a billion-dollar superlearner, the OpenAI courtroom clash, and don't miss the prompt of the week at the end.

Know someone who'd love this? Forward it their way. First time reading? Sign up here to keep up with the future of tech.

Musk Altman Trial Tests OpenAI Mission Shift
As Elon Musk sues Sam Altman, a dispute over OpenAI’s nonprofit roots reaches trial just as it nears a $1tn IPO. The case in Oakland centers on claims that restructuring into a for-profit entity breached founding agreements. Musk seeks over $134bn and leadership changes, while OpenAI—backed by Microsoft—argues he supported the shift in 2017. Witnesses include Satya Nadella.

OpenAI Expands Beyond Microsoft While Extending Partnership
Despite deep ties with Microsoft, OpenAI will now distribute its models across multiple cloud providers. Sam Altman said Microsoft remains its primary cloud partner under an updated agreement. OpenAI will continue supplying models through 2032 and share revenue with Microsoft until 2030. The shift introduces flexibility while preserving one of AI’s most consequential commercial alliances.

OpenAI Eyes Agent-Driven Phone Beyond App Ecosystems
Rather than competing within app stores, OpenAI is exploring a smartphone where AI agents replace traditional apps entirely. Analyst Ming-Chi Kuo says the company is working with MediaTek, Qualcomm, and Luxshare on chips and manufacturing. The device could combine on-device and cloud models, with specs finalized by late 2026 or early 2027 and mass production targeted for 2028. The approach would give OpenAI deeper system access and user data beyond current platform limits.

China Orders Meta To Unwind $2 Billion AI Deal
In a rare late-stage intervention, China is forcing Meta to reverse its completed $2bn acquisition of AI app Manus. Regulators say the deal violated foreign investment rules, despite integration already underway. Authorities demand a full unwind, including returning funds and halting use of Manus technology, with penalties possible. The move signals tightening scrutiny as US-China AI tensions escalate.

DeepSeek dropped V4 last week. Open source, open weights, frontier-level, and a fraction of the price of Opus 4.7 or GPT-5.5. This is a bigger deal than R1 was, and R1 dropped the stock market 20% basically overnight. → Read the full article here.

Researchers at MIT and the MIT-IBM Watson AI Lab unveiled “EnergAIzer,” a tool that estimates the energy consumption of AI workloads in seconds instead of hours or days, according to an April 27, 2026 report. The system analyzes repeatable patterns in AI computations and applies real-world GPU measurements to deliver estimates with about 8% error—comparable to slower traditional methods.

As data centers are projected to consume up to 12% of U.S. electricity by 2028, the tool could help operators allocate computing resources more efficiently and reduce wasted power. It also allows developers to evaluate the energy impact of models before deployment, potentially shaping more sustainable AI design choices. → Read the full article here.

David Silver, the researcher behind AlphaGo, has founded a new AI company, Ineffable Intelligence, backed by $1.1 billion in seed funding at a $5.1 billion valuation. The startup aims to build “superlearners” using reinforcement learning—systems that learn independently through trial and error rather than relying on human-generated data like large language models (LLMs).

Silver argues current AI approaches are limited because they depend on human knowledge, comparing them to “fossil fuels,” while self-learning systems could scale indefinitely. The company plans to train AI agents in simulated environments to develop broader, potentially superhuman capabilities, though questions remain about alignment and real-world applicability. → Read the full article here.

Meta has signed an agreement with startup Overview Energy to deliver up to 1 gigawatt of electricity generated by satellites that beam infrared energy to Earth-based solar farms. The system would allow solar-powered data centers to operate at night without relying on batteries or fossil fuels, as Meta’s energy use already exceeds 18,000 gigawatt-hours annually.

Overview plans to launch its first orbital test in January 2028 and deploy a fleet of around 1,000 satellites by 2030, each designed to operate for over a decade. The approach aims to bypass challenges tied to laser or microwave power transmission by using lower-intensity infrared beams compatible with existing solar infrastructure. The deal underscores how AI-driven energy demand is pushing tech companies toward unconventional, large-scale power solutions. → Read the full article here.

Google Staff Oppose Military AI Use: Over 600 employees, including DeepMind leaders, urge Sundar Pichai to block Pentagon access to classified AI, warning such work risks harmful, unaccountable applications without oversight.

DeepMind Partners With South Korea: Google DeepMind teams with MSIT to launch a Seoul AI Campus and deploy models like AlphaFold, already used by 85,000 researchers, to accelerate science and talent development.

Qualcomm Tied to OpenAI AI Chip Plan: Shares jump 7% after report of partnership with OpenAI and MediaTek on smartphone AI chips, with analyst Ming-Chi Kuo projecting device mass production by 2028.

Anthropic Bug Triggers Unexpected Charges: Developer reports Claude Max billed $200 extra after “HERMES.md” in commits allegedly rerouted usage to API rates; Anthropic acknowledged issue but declined refund.

Use this if you want the AI to help you learn, think critically, and solve your own problems rather than just handing you the answers.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team