In partnership with

Login or Subscribe to participate

WTF is a compute moat?: Training frontier AI needs thousands of chips, enormous power, fast networking and data centres that can run reliably.

A company that controls more of this infrastructure can run larger experiments and repeat them faster.

Compute buys attempts. Research decides whether those attempts become a better model.

xAI announced itself in July 2023.

OpenAI had been building models for eight years. Anthropic had Claude. Google had decades of AI research.

Four months later, xAI released the first Grok. It had trained for about two months, and xAI said that models trained with more data and compute were still ahead.

Grok was fast, funny and connected to X, but it did not look like a future frontier model.

Three years later, Grok 4.6 has entered the same benchmark tier as the leaders.

Axios reports that it scored essentially even with GPT-5.6 Sol Max and just behind Claude Fable 5 Max on the Artificial Analysis Intelligence Index.

How did a late lab erase that much distance so quickly?

I am Alex. Welcome to ShortCu8 by Innov8.

Lets Dive Deep🐰🐰.

Today's Shortcut

When an AI lab launches a strong model, do not look only at its rank.

Watch the interval between competitive releases.

Grok 4.6 suggests xAI has built a system that can return to the frontier.

Elon built the infrastructure first

xAI entered with less research history, so it compressed the physical build. It raised billions, bought chips and assembled Colossus quickly.

xAI says Colossus became operational with 100,000 Nvidia Hopper GPUs in 122 days. Workloads began 19 days after the first servers arrived. Read xAI's account.

Grok 4 was later trained using reinforcement learning on a 200,000-GPU Colossus cluster.

Tesla and SpaceX already operate around factories, power systems, supply chains and large physical projects. xAI applied the same instinct to chips, substations, cooling, networking and buildings.

Compute bought a seat, not the result

The GPUs created training capacity. The model still depended on what xAI did with it.

xAI credits Grok 4's improvement to algorithms that made training six times more compute-efficient, more verifiable training data and reinforcement learning at pretraining scale. The company explains the training here.

For Grok 4.5, xAI describes filtering, deduplication, quality scoring and domain-specific data selection. Read the Grok 4.5 release.

Colossus gave the team more attempts. Researchers, data and better methods converted those attempts into progress.

The result came from capital, compute, power, talent, data and fast iteration.

The price is the second surprise

OpenRouter's current listing says Grok Latest redirects to xAI's latest Grok model and costs $2 per million input tokens and $6 per million output tokens. Plenty of smaller models cost less. Among systems near the top of this benchmark, the price is unusually low. See the live model listing.

Reasoning length, cache use and tool calls still change the final bill.

xAI is competing on capability and the cost of using it repeatedly. A cheaper model near the frontier forces every expensive lab to explain what the extra money buys.

The moat moved behind the model

Grok 4.6 could lose its benchmark position next week. The infrastructure behind it remains.

Musk says Grok 4.7 should arrive in three to four weeks after more training on SpaceX data. That is a promise, not a shipped model. The timeline shows how xAI wants to compete: train, release, learn and return before the market settles.

The moat is becoming the whole machine behind the model:

  • access to compute and power

  • researchers who can use it efficiently

  • proprietary and verifiable data

  • products that generate feedback

  • enough capital to start the next run immediately

Reality check

One composite benchmark does not prove that Grok 4.6 is best at coding, writing, research or everyday use.

Artificial Analysis places Grok in the leading group. Your own work decides whether it belongs in your workflow.

xAI has shown that a research lead can shrink quickly when another company combines serious researchers with enormous infrastructure.

Grok 4.6 may not remain near the top for long.

Its rank is less important than the speed of the climb. In three years, xAI went from a two-month beta to a lab that can plausibly return with another frontier attempt in weeks.

The next leaderboard will change. Colossus, the training pipeline and what the team learned from this run remain. They are already the starting point for the next model.

Now go and test grok bots 🫏

🛠️Cool Tools of the Week:

  • Claude Cowork: Anthropic's work agent is now available in the Chrome side panel, allowing it to take on tasks in your browser and work across tabs.

  • Wan-Animate-2: The latest animation model from Alibaba, featuring high-fidelity character animation and real-time streaming generation. 

  • ChatGPT Work and Codex: Users can now keep work from other agents in sync with those from OpenAI products by importing projects, chats, skills, and plugins.

📩 Innathe Shortcu8 engane undarunnu 👇️?

We read every reply - just reply to this email and let us know how we can improve !

Appo adutha Shortcu8il kanaam bie…👋

If you read till here, you might find this interesting

#AD1

Domain Names + Web and Email Hosting Done Right

Tired of overpaying GoDaddy or Namecheap? There's a smarter option with Porkbun!

Porkbun is the domain registrar trusted by creators, developers, entrepreneurs, and anyone who wants low prices without the nonsense or the forced upsells.

Here's why so many people are making the move to the Bun:
• Most domains sold at cost
• Low, transparent registration and renewal pricing
• Free features like WHOIS privacy and SSL certificates
• Powerful web and email hosting options
• Real human support 24/7, 365 days a year
• Named the #1 domain registrar by Forbes Advisor and USA Today

Whether you're launching a business, building a personal brand, starting a side project, or creating your very first website, Porkbun makes the whole thing easy.

#AD2

What is an EOR—and why are companies using it?

Opening entities in every country can be slow, expensive, and hard to scale.

That's why more companies are using EOR to hire globally faster.

See how Oyster helps teams hire, pay, and support talent in 180+ countries while staying compliant along the way.