😸 Elon thinks Grok 5 = AGI

PLUS: Elon thinks Grok 5 = AGI?!

Welcome, humans.

Remember almost a year ago when Anthropic got roasted for its weird branding campaign? Well, turns out they’re back with a totally new vibe, and TBH, it’s working on us.

Keep thinking with Claude

There’s just one problem with this: this should have accompanied a new model. Why would you release a baller new video and NOT release a baller new product?! Where’s Claude Sonnet 4.1 team?? Just saying.

Next up: Here’s what might happen if you try to bring home your robot partner to meet your parents…

Here’s what happened in AI today:

Microsoft and xAI are building two gigantic data centers for new models.
Nvidia acqui-hired Enfabrica execs to solve GPU connectivity issues.
OpenAI and Jony Ive poached 24+ Apple engineers for new AI devices.
The UAE is getting three new data centers.

Advertise in The Neuron here

Microsoft and xAI Race to Build World's Most Powerful AI Factories

In yet another sign of the times, two GIANT datacenter buildouts caught our eye this week that were worth highlighting this Sunday.

You know the phrase “Follow the money?” Well in AI, “the money” pretty much all goes to datacenters… so whatever datacenters are being built now wlil give you a sense of what kinds of new AI capabilities could be coming next.

First up: Microsoft is building Fairwater in Wisconsin, what it calls the world’s “most advanced AI datacenter”, set to open in 2026:

The 315-acre campus spans 1.2 million sq ft across three buildings, packed with hundreds of thousands of NVIDIA GB200 GPUs (leading edge AI chips), connected by enough fiber to wrap Earth 4.5 times.
Coming online in early 2026, it'll use closed-loop cooling to minimize water usage.
Microsoft's also committed another $4B for a second similar-sized Wisconsin site.

Not to be outdone, Elon Musk’s xAI is also scaling its current Memphis datacenter cluster to build Colossus 2, the first “gigawatt scale” data center in the world. For context, a gigawatt = ~ one nuclear power plant’s worth of power.

As for where that electricity comes from?

Let’s just say Elon’s using, ahem, creative maneuvers across two state borders to get it the power baby Colossus needs.
After pushback in Tennessee, they acquired a former Duke Energy plant across the border in Mississippi and got approval to run gas turbines for 12 months without permits.
They're also building an $80 million wastewater facility to recycle 13M gallons daily for cooling.

Where’s all this going? Better and better models, of course. xAI just dropped Grok 4 Fast, claiming frontier performance with 40% fewer “thinking” tokens. The 2M-token context model runs both reasoning and non-reasoning modes, priced at just $0.20/M input and $0.50/M output (under 128k tokens). And at the moment, it's temporarily free on OpenRouter and Vercel.

Grok 4 Fast builds on the heat from Grok Code Fast-1, a (you guessed it) really fast model for coding agents, which is also free on GitHub Copilot, Cursor, and other platforms. With Colossus 2 underway, Elon now thinks Grok 5 could “reach AGI.”

Is this building up hype? Probably. Could it be true? Possibly. We’ll find out soon after it starts training in a few weeks. Anyone can claim AGI… measuring it is another story entirely.

What about Microsoft and the rest of the gang? We can assume Microsoft’s datacenters will go to training whatever new models OpenAI is cooking up, plus the new versions of Microsoft’s own MAI-1 models.

OpenAI believes infrastructure is destiny, and it’s certainly been true that if you build it, it’ll (eventually) get used. Between all its various projects and cloud spending, OpenAI anticipates spending $17B next year.

Let us not forget Google, reportedly working on Gemini 3, with leaked benchmarks suggesting impressive scores (the folks who have leaked their impressions of it have been very impressed). From those leaks, we also heard that Gemini Flash 3 will be better than Gemini 2.5 Pro… if you can believe it.

While OpenAI has the hype train chugging, and Google might have lost last week’s competitive coding round, Google still has a wickedly powerful inherent advantage in AI. We couldn’t explain it better than Yuchen (of Hyperbolic), so we’ll just let his tweet do the talking:

No matter who ends up having the best of these upcoming models, we're watching infrastructure investments that dwarf entire countries' GDPs… all fueled by the equivalent of an infinite money glitch… and it could be that whoever designs the best infrastructure today may control the destiny of the AI industry…

FROM OUR PARTNERS

Don’t Let SOC 2 Kill Your Startup

A startup was on the rise—steady growth, happy customers, investors circling. Then a big prospect asked for SOC 2.

They signed up with one of the big platforms. Suddenly: endless tasks, confusing checklists, engineers pulled off product. Deals stalled. Competitors moved in.

Those competitors? They used Delve.

15 hours later, they were audit-ready and closed the deal.

The first startup never recovered. One by one, their deals slipped away. Keys handed over. Game over.

Don’t be that startup.

Delve automates compliance—SOC 2, ISO 27001, HIPAA, GDPR, PCI-DSS and more—so you’re ready in days, not months.

Today, compliance is done in Delve.

Book a demo, use TLDR1KOFF for 1K off.

Prompt Tip of the Day

Want to stop struggling with vague image editing prompts? Reddit user promptingpixels found a game-changer: annotate directly on your images.

The tool: Google’s Nano Bananas image model.

The method: Use any image editor to draw red boxes around areas you want changed, add red text with specific instructions, then prompt: “Read the red text in the image and make the modifications. Remove the red text and boxes.”

Gemini supposedly gets this right 9/10 times, even with complex group photos or multiple edits.

Why do we like this? You're giving AI visual instructions instead of playing prompt guessing games. Much more precise than describing “the thing on the left near the blue object.” Visual tools need visual prompting!

Treats to Try

*Asterisk = from our partners (only the first one!). Advertise in The Neuron here.

*Gamma instantly creates presentations and websites from your rough ideas.
Dyad lets you build full-stack AI applications locally on your device with complete privacy and no cloud lock-in—free forever.
HeyGen Video Agent turns your text prompt into a complete video with script, visuals, and voiceover in minutes—free to try.
Nanobot provides an open-source MCP agent framework for building AI applications with scientific communication expertise.
MagicLight AI transforms your scripts into animated videos up to 50 minutes long with character consistency and AI voiceovers.
ElevenLabs Studio 3.0 combines audio and video editing in one browser platform with AI voices and custom sound effects.

Around the Horn

NVIDIA “reverse-acquihired” AI networking startup Enfabrica’s top management for $900M+ to solve GPU connectivity bottlenecks in massive data centers (relevant, as you can imagine based on today’s main story!).
LinkedIn updated its privacy policy to let Microsoft use more of your profile data for AI training starting November 3rd (but you can opt out).
OpenAI and Jony Ive poached 24+ Apple engineers with $1M+ stock grants to build AI devices like smart glasses, voice recorders, and wearable pins for late 2026 release.
Khazna Data Centers is building three AI-optimized facilities in the UAE while planning a massive 1GW expansion across Europe and the Middle East.

FROM OUR PARTNERS

Explore AI Innovation at AWS Symposium Ottawa on October 1

AWS Public Sector Symposium is returning to Ottawa and featuring the latest in emerging technology, data and AI. Join tech leaders and those driving transformation across the public sector in Canada, and leave inspired by what’s possible.