In partnership with
🧠 THAT ONE AI - This Week’s Signal
This Week:
🇯🇵 Japan's Sakana AI takes aim at the frontier
🧬 Google DeepMind loses a Nobel Prize winner to Anthropic
💧 Nvidia's zero-water AI factory design picks up steam
🧰 Tools worth testing
🇯🇵 A Japanese Startup Claims It Can Match Anthropic's Best Models
With Anthropic's Fable 5 and Mythos pulled from global access, a gap opened at the frontier. Sakana AI moved fast to fill it.
The Tokyo-based lab launched Fugu, a multi-model orchestration system that routes each request across a pool of specialized models behind a single API. The core model picks the right helpers, divides the work, checks the outputs, and merges the results. From the outside, it looks like one model. Under the hood, it is a coordinated team.
Two versions ship on the same API:
Fugu for everyday coding and chat tasks
Fugu Ultra for heavier work like patent research and security testing
Sakana claims both versions perform at or above Fable 5 and the original Mythos Preview on coding, reasoning, and science benchmarks. They also pitched Fugu explicitly as a model that delivers frontier capability without the export control risk that took down Anthropic's top models.
The early reception, though, is mixed. Users report real-world performance that does not match the benchmarks, and there is genuine skepticism around the cost structure and lack of visibility into which models Fugu is actually using under the hood.
Interesting approach. Worth watching. Not ready to call it a frontier replacement yet.
The bigger signal: 👉 Model orchestration is becoming a legitimate strategy for labs trying to reach the frontier without building a single massive model. Whether it can actually get there is still an open question.
The ones showing up in LLMs convert 3× better than Google
They optimized for LLMs, not just Google.
FAQs. Comparison pages. Transparent pricing. LinkedIn presence. These aren't vanity plays. They're what gets you cited in ChatGPT, Gemini, and Claude when your buyers are researching, your investors are looking, and your future hires are deciding where to work.
Download the free AEO Playbook for Startups from HubSpot and get the exact checklist. Five minutes to read.
🧬 Google DeepMind Just Lost Its Nobel Prize Winner to Anthropic
John Jumper, co-creator of AlphaFold and 2024 Nobel Prize winner in Chemistry, is leaving Google DeepMind for Anthropic after nearly nine years.
Jumper led the team that built AlphaFold, the protein structure prediction model widely considered one of the most consequential scientific contributions in recent AI history. He shared the Nobel with DeepMind CEO Demis Hassabis.
His departure follows Noam Shazeer, Gemini co-lead and Transformer co-author, who left for OpenAI just days earlier. Two foundational researchers out the door in one week.
A few additional details worth noting:
Jumper had reportedly been contributing to enterprise coding tools at Google, an area where the company has struggled relative to Anthropic and OpenAI
He says he is taking time to recharge before joining Anthropic
His arrival comes ahead of an Anthropic science event on June 30, which now carries obvious significance
Google DeepMind built its reputation on scientific AI. AlphaFold was the crown example. Losing the person who built it, to a direct competitor, in the same week as another high-profile departure, is a meaningful signal about where top researchers see the most interesting work happening right now.
The bigger signal: 👉 Talent concentration at Anthropic and OpenAI is accelerating. For Google, the science edge it held in 2024 and 2025 is no longer something it can take for granted.
💧 Nvidia's Zero-Water AI Factory Is Getting Serious Attention
AI data centers consume enormous amounts of water for cooling. That has become one of the loudest points of opposition to new data center projects globally. Nvidia thinks it has a workable answer.
The company's new factory design uses a closed-loop liquid cooling system that achieves near-zero water consumption in the right climate conditions. The design was first introduced in early June, but a recent post from Nvidia pulled in over 12 million views and pushed the story back into wide circulation.
The technical concept is straightforward: instead of evaporating water to remove heat, the system recirculates coolant in a closed loop. No water leaves the system. In climates where ambient temperatures support it, the approach works without meaningful water draw.
For communities pushing back on AI infrastructure on environmental grounds, this is the kind of engineering response that changes the conversation.
The bigger signal: 👉 The bottleneck for AI infrastructure is no longer just compute. Energy and water constraints are real. Labs and chip companies that solve for both will build faster than those that do not.
That One AI 🧰 TOOLBOX
A few tools quietly worth exploring:
⚡ Nex N2 → A new model family built specifically for action-oriented tasks including coding, search, and tool use
💻 Kimi K2.7 Code → Moonshot AI's latest coding agent designed to handle complex, multi-step developer workflows
🏷️ Monoshoot → Turns raw product photos into studio-quality e-commerce visuals without a photoshoot
🔢 IMG2Excel → Converts any image directly into an editable Excel file
Your best prompts are the ones you'd never bother typing.
The detailed ones. The ones with examples and edge cases. Wispr Flow lets you speak them instead — clean, structured, ready to paste into any AI tool. Free on Mac, Windows, and iPhone.
🔚 EXIT NODE
A Japanese lab challenges the frontier with orchestration instead of scale.
Nvidia is solving for water the same way it solved for compute.
Google's best scientific mind just walked out the door to a competitor.
The AI race is not just about who has the biggest model anymore.
It is about who has the best researchers, the cleanest infrastructure, and the most trust from the people building on top of it.
All three are in motion right now.
See you next issue.