Good morning. It's Wednesday, July 15, and we're covering DeepMind's push for AGI safety standards, NVIDIA's expansion beyond GPUs, and Apple's race to run AI on device.
Plus: yesterday’s poll results on the risks of building atop AI giants, and test your AI detection skills in today's Real or AI challenge.
YOUR DAILY ROLLUP
Top Stories of the Day

OpenAI’s first hardware product will reportedly be a screenless, mobile AI-guided smart speaker pitched internally as a "humanlike AI companion that lives in the home," with self-moving mechanical elements, a personality, and access to the user's emails and broader digital life. Designed with heavy input from ex-Apple engineers behind the iPhone and Mac (many of whom joined via last year's $6.4B io Products acquisition), the device lands one week after Apple sued OpenAI for trade-secret theft; OpenAI told Bloomberg the product diverges enough from anything Apple sells today that it is unlikely to infringe.
PrismML says it shrinks a 54 GB AI model to under 4 GB, making a 27-billion-parameter model run directly on an iPhone 15 or newer. Apple is evaluating the startup's technology as it seeks faster, more private on-device AI for Siri. PrismML claims 10–15x lower memory use, 6–8x faster responses, and raised a $16.25 million seed round. Analysts say large-scale testing will determine whether the gains hold up.
New York becomes the first state to halt approvals for large AI data centers, signaling growing resistance to the industry's rapid expansion. Gov. Kathy Hochul's order pauses permits for new facilities of 50 megawatts or larger while the state completes a yearlong environmental review. Officials cite concerns over power costs, water use, and local impacts, even as AI demand drives larger projects.
American workers are becoming more productive even before generative AI delivers its expected economic boost. Economists attribute the gains to tight labor markets, digitization, remote work, and leaner staffing rather than AI alone. Labor productivity has grown at its fastest pace in over two decades, while tech and finance continue cutting jobs despite rising output. The long-term question is whether wage growth will keep pace with productivity.
VIDEO
Codex Power-User Guide
Best GPT-5.6 model choices, multi-thread workflows, browser automation, loops, plugins, remote control, and safety tips.
SAFETY
DeepMind CEO Calls for Frontier AI Standards Ahead of AGI

DeepMind CEO Demis Hassabis argues that Artificial General Intelligence (AGI) could arrive within the next few years and describes it as a transformative technology on par with electricity or fire. He warns that rapid advances in frontier AI are outpacing society's ability to understand and manage risks, including cybersecurity, biological threats, and increasingly AI systems.
To address those challenges, Hassabis proposes a U.S.-led Frontier AI Standards Body that would evaluate advanced models against evolving safety benchmarks before deployment, with mandatory oversight eventually replacing an initial voluntary review process. He also calls for international cooperation, saying technical safeguards must be paired with broader public debate on the economic, ethical, and societal consequences of AGI. → Read the full article here.
SEMICONDUCTORS
NVIDIA Bets on AI CPUs to Expand Beyond GPUs

NVIDIA is positioning its new Vera CPU platform as its next major growth driver, with Wedbush Securities arguing the move could significantly expand the company's addressable market beyond AI graphics processors. CEO Jensen Huang says Vera is designed for AI workloads and agentic AI systems, with NVIDIA claiming it delivers about 1.8x the performance of comparable x86 processors while working alongside its AI chips.
The company continues to post strong financial results, reporting first-quarter fiscal 2027 revenue of $81.6 billion, up 85% year over year, alongside $75.2 billion in data center revenue and guidance for roughly $91 billion in second-quarter revenue despite ongoing China export restrictions. Analysts see the CPU strategy as a way for NVIDIA to compete more directly with Intel and AMD while strengthening its position. → Read the full article here.
LITIGATION
Apple Lawsuit Accuses OpenAI of Coordinated Trade Secret Theft

Apple has filed a 41-page lawsuit accusing OpenAI of orchestrating a coordinated effort to obtain its confidential information through current and former Apple employees. The complaint alleges that OpenAI leadership encouraged employees and recruits to access internal systems, bring Apple prototypes and design materials to interviews, and avoid security measures during their departures.
Apple also claims OpenAI and its hardware subsidiary io improperly used Apple's proprietary industrial design, manufacturing, and supplier knowledge while building new hardware products. OpenAI has denied the allegations, saying it has "no interest in other companies' trade secrets," while Apple says it attempted to resolve the dispute privately before filing suit. → Read the full article here.
NEWS
What Else is Happening

Chai Discovery Raises $400M (Paywall): AI drug startup secures funding at a $3.8B valuation, reflecting strong investor interest in AI-driven drug discovery.
iOS 27 Beta Debuts a Gemini-Powered Siri: Apple's iOS 27 public beta ships a conversational Siri AI built with Google's Gemini, limited to iPhone 15 Pro and newer.
Meta Faces AI Layoff Lawsuit: Employees claim AI-assisted tools unfairly targeted disabled workers and those on medical leave. Meta denies the claims.
DeepSeek Eyes $1.5B Raise: Chinese AI startup reportedly seeks new funding at a $71B valuation while preparing for a future IPO.
U.S. Confirms Limited H200 Shipments: A trade official said very few NVIDIA H200 AI chips have reached China under export licenses.
Google Refreshes Image Search: Google adds personalized image galleries and AI image generation, bringing a major update to Image Search.
SURVEY
AI Beyond Chat for Teams
Turn everyday AI use into repeatable workflows your team can use in real work.
Forward Future and Hiten Shah are exploring a live, hands-on cohort course for company teams that already use ChatGPT, Claude, Copilot, or similar tools and want to go further.
The course would be built around real work. Each company would enroll at least two colleagues who work together on the same recurring workflow or business problem. Tuition would be charged per participant.
One person should complete this form on behalf of the team.
We are gathering interest before finalizing the first cohort.
REAL OR AI
Can You Still Tell if an Image Is AI?

POLL RESULTS
Is Every Prompt a Competitive Risk?
Here's how you voted: Flexibility won, with nearly half opting for a multi-provider strategy, while 26% preferred running open models.
🟨⬜️⬜️⬜️⬜️⬜️ Build on frontier models anyway, speed wins (13%)
🟨🟨🟨⬜️⬜️⬜️ Hedge with open models I can run myself (26%)
🟩🟩🟩🟩🟩🟩 Stay multi-provider, easy to swap (48%)
🟨⬜️⬜️⬜️⬜️⬜️ I don't see it as a real risk (13%)
That's All for Today
Before you go, what did you think of today's issue?
Thanks for reading. See you next time!
— Matthew Berman, Nick Wentz & the Forward Future Team

