SubQ is the new hero AI

Welcome back to The Shift. Let’s get straight to what matters in AI today…

Today we have:

🤖 A New AI Startup Just Launched a Model That Can Hold 12 Million Tokens at One-Fifth the Cost

🤓How to Make a High-Stakes Business Decision With Serno

🤖 Anthropic Just Signed a Compute Deal With SpaceX and Doubled Claude Usage Limits

🔨Tools and Shifts you cannot miss

🤖 A New AI Startup Just Launched a Model That Can Hold 12 Million Tokens at One-Fifth the Cost

Subquadratic came out of stealth this week with $25 million in seed funding, backed by a former SoftBank Vision Fund partner and Tinder cofounder, launching SubQ, the first commercial LLM built on a sub-quadratic architecture.

The Shift:

The Problem It Solves - Standard AI attention costs quadruple every time you double the input; that is why RAG and chunking exist. SubQ's SSA architecture scales linearly, running 52x faster at 1 million tokens.

The Cost Gap Is Wild - SubQ scored 97% on long-context accuracy for just $8 versus $2,600 on frontier models. On multi-needle retrieval, SubQ scored 83 versus Opus 4.6's 78, GPT-5.4's 39, and Gemini 3.1 Pro's 23.

What It Can Actually Do - SubQ holds a native 12 million token context window more than any frontier model, hitting 92% recall at that length. Two products are live: a 12 million token API and SubQ Code for entire codebases.

Where It Still Falls Short - On real-world coding, SubQ scored 81.8% on SWE-Bench versus Opus 4.7's 87.6%. Cross-session memory tools like CLAUDE.md also remain useful. SubQ solves cost and context length, not persistent memory.

Subquadratic is targeting 100 million tokens by Q4. If the architecture holds at scale, RAG and chunking become less necessary; you skip the engineering overhead and just ask the model directly.

Together with Attio

Attio is the AI CRM for high-growth teams.

Connect your email, calls, product data and more, and Attio instantly builds your CRM with enriched data and complete context. Whether you’re running product-led growth or enterprise sales, Attio adapts to your unique GTM motion.

Then Ask Attio to plan your next move.

Run deep web research on prospects. Update your pipeline as you work. Find customers and draft outreach emails. Powered by Universal Context, Attio's intelligence layer, Attio searches, updates, and creates across your data to accelerate your workflow.

Ask more from your CRM.

Ask Attio

🤓How to Make a High-Stakes Business Decision With Serno

Stop asking a chatbot. Get a team that debates before it answers.

Step 1: Drop your question - Type any decision you’re sitting on, build vs buy, pricing change, vendor selection. Serno assembles a team of AI expert personas around it instantly.

Step 2: Let them disagree - One expert is mandatory adversarial, their only job is to find the holes in every argument. The pushback is built in, not prompted.

Step 3: Audit the canvas - Every claim shows who made it, where it came from, and a confidence signal. No stats you can’t trace. No confident nonsense.

Step 4: Pick your answer - Claude, GPT, and Gemini all run the same question. See which model’s reasoning actually holds up on your specific decision. You can try Serno here.

🤖 Anthropic Just Signed a Compute Deal With SpaceX and Doubled Claude Usage Limits

Anthropic signed a deal to use all of SpaceX's Colossus 1 data center this week, immediately using the new capacity to double Claude Code usage limits across all paid plans.

The Shift:

The SpaceX Deal - Anthropic gets access to all of Colossus 1, over 220,000 Nvidia GPUs, and 300-plus megawatts coming online within the month. Musk said SpaceX will rent to AI companies taking the right steps for humanity.

What Changed for Users Claude - Code five-hour rate limits are doubling for Pro, Max, Team, and Enterprise plans today. Peak hour restrictions are removed for Pro and Max and API rate limits for Claude Opus models are significantly raised.

The Broader Compute Picture - Anthropic now has deals with Amazon for up to 5 GW, Google and Broadcom for 5 GW in 2027, Microsoft and Nvidia for $30 billion of Azure capacity, and a reported $200 billion Google Cloud commitment.

Orbital Compute Is Also Coming - Anthropic expressed interest in developing multiple gigawatts of orbital AI compute with SpaceX. International expansion is also underway, with Amazon adding inference in Asia and Europe for regulated industries.

This is the same Musk who called Anthropic "Misanthropic" now renting them his supercluster. The compute arms race has made for strange bedfellows and Anthropic is clearly not slowing down on infrastructure.

Together with BELAY

Growth Requires Letting Go of the Wrong Work

The work that got you here won’t take you further.

Without support, it’s easy to stay stuck doing everything long after you’ve outgrown it.

Download Operator to Owner: How to Exit the Middle to learn how to refocus your time on the work that actually deserves you.

Download the Free Guide.

🔨AI Tools for the Shift

🎯 Reloop – Creates high-converting AI video ads with realistic avatars, making ad production possible without creative or editing skills.

🎵 BeatMV – Turns your music into full AI-generated music videos with automated storyboarding, visuals, and scene creation.

🎼 CreateYourMusic.ai – Instantly generates professional-quality music from your ideas without needing production experience.

🔍 Bulker – Automates user research and delivers AI-powered insights in seconds instead of weeks of manual interviews.

▶️ Noodle Tomato – Generates full-length faceless YouTube videos automatically so creators can scale content and ad revenue faster.

🚀Quick Shifts

🌐 Google shuts down Project Mariner, its experimental web-task AI agent, after integrating its technology into products like Gemini Agent, AI Mode, and Chrome’s automated browsing features.

🚀 xAI is being folded into SpaceX under the new “SpaceXAI” branding, with Elon Musk saying xAI will no longer exist as a separate company and its AI products will operate within SpaceX.

🖥️ OpenAI partners with AMD, Broadcom, Intel, Microsoft, and NVIDIA on a new networking protocol called MRC, designed to improve performance and reliability in massive AI training supercomputer clusters.

🧠 Barry Diller says he trusts OpenAI CEO Sam Altman, but warns that trust may become irrelevant as AGI advances, arguing the biggest risk is the unpredictable consequences of superintelligent AI.

🔍 Google updates AI Search to include quotes and discussion snippets from Reddit, forums, blogs, and subscribed news sources, adding more context and attribution within AI-generated search responses.

🧩 Prompt of the Day

Checkout Progress Bar That Pushes Users Toward Completion

Most checkout pages feel endless. A clear progress bar reduces uncertainty by showing customers exactly how close they are to finishing.

Turn checkout momentum into fewer abandoned carts.

Paste the prompt: Drop this into ChatGPT, then fill in your checkout flow.

Prompt to paste

Create a checkout progress bar strategy for [Insert brand or store]. Define the key checkout steps, such as cart, shipping, payment, review, and confirmation. Write short progress messages that reassure customers and motivate them to continue. Suggest where the progress bar should appear and how it should reduce uncertainty, improve flow clarity, and increase completed purchases.

🤳AI Nugget of the Day

That’s all for today’s edition. See you tomorrow as we track down and get you all that matters in the daily AI Shift!

If you loved this edition, let us know how much:

Forward it to your pal to give them a daily dose of the shift so they can 👇