AI’s Next Race Is About Efficiency

/Welcome back apprentices! 👋

Hey {{first_name|friend}},

AI had a very normal week: one mystery model got unmasked, OpenAI started making chips, and Elon Musk wants computers in space.

Apparently, building smarter AI wasn’t complicated enough.

So grab your metaphorical hard hat.

The race is getting much bigger than chatbots.

_{Read Time: 4 minutes}

🕵️ Hackers Just Got an AI Upgrade. China-linked hacking groups are reportedly pulling off more than twice as many attacks after adding low-cost, open-source AI models like DeepSeek to their toolkit. Researchers say they’re using AI to write attack code, sift through emails, and map targets at scale — proving that cyber threats don’t need the fanciest AI to get a serious upgrade. The bigger concern: as AI gets better at cyber tasks, cheap models with fewer safeguards could make sophisticated attacks easier for far more groups to launch.

🧴 Lady Gaga’s Next Hit Might Be Skincare. Outer Bio, co-founded by Lady Gaga and Michael Polansky, has unveiled Yuna — a system that keeps human skin alive in the lab for up to four weeks while AI studies the results and suggests promising new skincare compounds. The company says that feedback loop has cut its discovery cycle from roughly 18 months for two leads to a new candidate about every six weeks, though the lab-grown skin still can’t fully mimic blood flow or immune responses.

🧬 AI Spots What Cancer Tests Miss. Researchers used AI to study breast tumor samples cell by cell and uncovered hidden patterns linked to how aggressively the cancer behaves. In a small study of 27 patients, certain cell-division abnormalities were associated with better or worse outcomes, giving researchers new clues about disease progression. The system is now being tested with more data and other tissues to see whether it can help guide more personalized treatment decisions.

Most AI lives on a screen. This kind drives itself through a soybean field at 2 a.m. and cuts weeds to the centimeter, with no herbicide. Greenfield Robotics has 82 machines running across 16 states. This year's fleet sold out, and every robot was delivered. 

The platform keeps adding jobs: weeding now, feeding, spraying, and cover-crop planting on the way. Farmers are done with chemicals linked to Parkinson's and tied to the price of oil. 

The pull is real, and the machines work. Greenfield's Regulation A+ offering lets everyday investors own a stake ahead of the scale-up the company is building toward.

Invest in Greenfield Robotics

^{Greenfield Robotics is Testing The Waters under tier 2 of Regulation A. No money or other consideration is being solicited, and if sent in response will not be accepted. No offer to buy the securities can be accepted and no part of the purchase price can be received until the offering statement filed by the company with the SEC has been qualified by the SEC. Any such offer may be withdrawn or revoked, without obligation or commitment of any kind, at any time before notice of acceptance given after the date of qualification. An indication of interest involves no obligation or commitment of any kind. “Reserving” shares is simply an indication of interest. There is no binding commitment for investors that reserve shares in this manner to ultimately invest and purchase the shares reserved of the company, or to purchase any shares of the company whatsoever.}

The AI race is entering a new phase.

China’s Z AI revealed that the mysterious Ox Alpha model was its own GLM-5.3-Flash, offering strong coding performance at unusually low prices; OpenAI showed off a custom chip designed to make AI responses faster and more power-efficient; and SpaceXAI is working with Nvidia on hardware it eventually wants to put in orbit. 

Together, these developments point toward AI’s next big battleground: squeezing more intelligence out of every dollar, watt, and chip. 

In other words, the AI industry is discovering that being brilliant is great — but somebody still has to pay the electricity bill.

Ox Alpha spent days inspiring speculation about which major company had secretly built it. Z AI eventually revealed that it was a preview of GLM-5.3-Flash.

The model is designed for efficiency: it has 320 billion total parameters, but activates only about 18 billion per token. It supports context windows of up to one million tokens and aims to deliver expensive-model performance without the same computational appetite.

Z AI says it scored 57 on Artificial Analysis’ Intelligence Index at $0.045 per task, while its weights are available for developers to run or customize. The company also says the anonymous Ox Alpha test ran entirely on Chinese-developed accelerators, with software-hardware optimization bringing per-token costs closer to mainstream Nvidia infrastructure.

That does not mean China has eliminated its dependence on advanced foreign chips, especially for training frontier models. But it does suggest that efficient models and better software can offset some hardware disadvantages.

OpenAI is attacking AI costs from the hardware side with Jalapeño, its first custom inference chip, developed with Broadcom.

Inference is the part where an already-trained model answers questions, writes code, or completes tasks. In published benchmarks against Nvidia GB200 and GB300 systems, OpenAI reported roughly 1.5–1.9× higher throughput per kilowatt and 1.7–3.6× lower latency, depending on the workload.

These are early, company-published results, and Jalapeño is not a general Nvidia replacement. It is designed for inference, while OpenAI says it will continue using Nvidia and other accelerators for training and additional workloads.

Still, specialized hardware could help OpenAI reduce its dependence on expensive general-purpose systems. The company says second- and third-generation chips are already in development, with the first expected to enter its infrastructure by the end of 2026.

The goal isn't necessarily to defeat Nvidia.

It's paying Nvidia less often.

SpaceXAI and Nvidia are developing an orbital computing system for Starmind, based on an optimized version of Nvidia’s Vera Rubin NVL72 architecture.

The idea is straightforward: AI data centers need enormous amounts of electricity, land, cooling and grid capacity. Space offers abundant solar energy and, theoretically, room to scale without building another massive terrestrial facility.

The problems are equally straightforward. Chips are difficult to cool in a vacuum, hardware must withstand radiation, launches and replacements are expensive, and communications add another challenge. Orbital AI is therefore not about replacing Earth-based data centers soon.

It is a long-term bet that cheaper launches, better satellites and rising demand for compute could eventually make space-based infrastructure economically viable.

You probably aren't buying 700-watt AI chips or launching server racks into orbit anytime soon.

But the economics underneath these announcements eventually reach your laptop, your workplace and the AI tools you pay for.

Cheaper inference can mean lower subscription prices, larger usage limits and AI features appearing inside software where they previously weren't economical. More efficient models could make powerful local AI practical on smaller hardware, keeping more data away from the cloud. And competition between Nvidia chips, custom silicon and alternative hardware could reduce one of the industry's biggest costs.

For businesses, there is an even bigger implication.

Today you might ask, “Can AI do this task?”

Increasingly, the better question will be:

“Can AI do this task cheaply enough that doing it 100,000 times makes sense?”

That's when efficiency stops being a nerdy benchmark number and starts becoming a business model.

Hey, you didn’t get all this info for nothing — share it! If you know someone who’s diving into AI, help them stay in the loop with this week’s updates.

Sharing is a win-win! Send this to a friend who’s all about tech, and you’ll win a little surprise 👀

Creators lock in holiday content calendars 90 days out, before brands figure out commissions. Waiting too long to launch an affiliate program means less runway to build demand and a missed shot at the best partnerships.

The 90-Day Holiday Sprint covers commissions, recruiting, and scaling a program at Day 30, 60, and 90.

Download the Free Guide

A playground for your imagination (and low-key prompt skills).

Each send, we give you a customizable DALL·E prompt inspired by a real-world use case — something that could help you in your business or job if you wanted to use it that way. But it’s also just a fun creative experiment.

You tweak it, run it, and send us your favorite. We pick one winner to feature in the next issue.

Bonus: you’re secretly getting better at prompt design. 🤫

Last week, we challenged you to test GPT-4o’s visual generation skills with this prompt.

Here’s the WINNER:

Congrats to Arthur for this creation!🥳

Want to be featured next? Keep those generations coming!

Inside a vast circular studio, six fully immersive environments surround a single central object: [your object]. Each section of the studio places the same object in a radically different setting — a futuristic laboratory, a tropical jungle, a snow-covered mountain, a luxurious penthouse, a deep-sea research station, and a spacecraft interior. The environments blend seamlessly at their boundaries while the object remains visually consistent in every scene. Cinematic lighting, vibrant colors, ultra-detailed textures, photorealistic materials, dramatic depth, and high-end concept-art realism.

If AI outputs feel inconsistent, it’s usually not the model, it’s missing structure.
We documented the exact 6- Part System we use to get reliable results across ChatGPT, Claude, and Gemini.

It’s a short guide you can finish in under an hour, with plug-and-play prompts + exercises so you actually build the skill and fix the frustrating AI inconsistencies.

Subscriber Price: $10 (normally $19).

_FEEDBACK