🍭 Microsoft’s $14B GPU Buffet 😋

Good morning. Some people start their day with meditation.

We start ours by refreshing Google’s AI blog and pretending it’s self-care.

Let’s dig in.

🍭 What’s Cookin’:

• Microsoft taps Nscale

• Google’s Veo 3 → Veo 3.1

• 5 AI Tools to Feel Superhuman

• Everything Else: X, Meta, Spotify, and more!

_Google
🎬 Google’s Drops Veo 3.1 and Moves it Right Into Flow

The Bite: Google rolled out Veo 3.1 and tucked it straight into Flow, its built-in AI video editor.

The update adds richer audio, tighter prompt adherence, better image-to-video, and new edit powers like insert/remove objects and extending shots.

Flow’s now a one-stop shop:
model + timeline + tools under one roof, which is exactly how you win attention back from flashy demos;

By being easy to use daily.

Since May, creators have made 275M videos in Flow, so the on-ramp is real.

If Google keeps lowering the friction, the conversation can swing back from Sora 2 viral video formats to “what can I ship today?”

Snacks:

• Flow First: Model and editor live together. Prompt, tweak, export.
  

• Audio Everywhere: Generated audio now works across Flow’s core features.
  

• Granular Edits: Insert new objects, soon remove them, and keep it seamless.
  

• Image-to-Video+: Better realism and adherence when turning pics into clips.
  

• Extend Shots: Grow scenes into longer, connected sequences.
  

• Real Usage: 275M videos made in Flow since May launch.

Why it bites: Attention follows access.

By bundling Veo 3.1’s control inside Flow, Google is aiming at the people who make videos for a living:
Marketing teams, product launches, social ops.

Not just viral video hunters.

“Boring” features like precise inserts, removals, and longer shots are exactly what turn AI video from stunt to standard tool.

And as we learned yesterday from Cisco;
“Boring” is where budget comes from.

_ToolBox™
⚒️ 5 AI Tools to Feel Superhuman

🎬 Sora 2
→ OpenAI’s new text-to-video engine for cinematic, ultra-realistic footage. Think “type a sentence, get a trailer.”

🧩 Apps in ChatGPT
→ Chat with real apps inside ChatGPT—or build your own. No tabs, no APIs, just direct integration.

🎥 Tight Studio 2
→ Record or generate gorgeous product demos with AI-driven pans, zooms, and captions. Screen Studio vibes, zero effort.

✉️ Dreamlit
→ Write and automate outreach emails that actually sound human—on-brand, on-tone, on autopilot.

🎬 Veo 3.1
→ Google’s next-gen text-to-video model built for real production work.
Longer clips, richer audio, and style control with reference images.

Steal This Prompt
🎮 Fun Fortnite Character Builder

Design outrageous Fortnite-style characters from scratch or modeled after a known persona, perfect for mockups, memes, or quick concept art.

Bonus chaos-minimizers: a CFO skin with a spreadsheet pickaxe; a “Startup Founder” with a latte launcher.

1. Click this link (Prompt)

2. Paste into GPT

3. Replace the #’s with your subject details

4. Watch it cook (your squad just unlocked “CEO With a Nerf Budget”)

   
   🍭 Try the prompt on Snack Prompt 🍭

_{Everything Else}
🧠 You Need to Know

🧩 X Tests Profile Transparency Features

→ X will pilot new profile info (account age, country/region, username changes, app-store origin).

Intended to help users assess authenticity, with opt-outs flagged and a staged employee rollout.

🔗 Meta Partners With Arm on AI Efficiency

→ Multi-year deal moves Meta’s ranking/recommendation systems onto Arm’s Neoverse platform.

Objective: to scale AI with better performance-per-watt amid major data-center expansion plans.

🪶 Anthropic Updates Claude Haiku (4.5)

→ New Haiku aims for Sonnet-level capability at lower cost and latency;

Available on free plans with emphasis on fast, lightweight production use.

🎧 Spotify’s AI DJ Can Now Text

→ Premium users can type requests (English and Spanish).

Get prompt suggestions, and mix voice or text control across 60+ markets.

🇦🇷 OpenAI Signs LOI for ‘Stargate Argentina’

→ OpenAI and Sur Energy will explore a clean-power data-center project after meetings with President Milei;

OpenAI may serve as offtaker and expand “OpenAI for Countries” efforts.

Hardware
📶 Nscale’s $14B GPU Haul With Microsoft

The Bite: Microsoft just tapped UK upstart Nscale for one of the biggest AI infra orders on record: ~200,000 Nvidia GB300 GPUs headed to Azure sites in the U.S. and EU, with Dell in the mix. The bulk ships to a new Texas build; Portugal gets a smaller slice; Norway keeps humming from a prior deal. Nscale’s using the momentum to whisper “IPO by 2026,” while everyone else rushes to lock racks, power, and patience. Casually bullish? Yes. But the real bottleneck isn’t chips—it’s the grid.

Snacks:

• $14B Deal: One of the biggest AI infra contracts ever;
  About 200,000 Nvidia GB300s.
  

• Texas Takes the Lead: Over half headed to a new Texas site;
  Portugal and Norway get the rest.
  

• Dell in the Loop: Dell joins as hardware and deployment partner.
  

• IPO Watch: Nscale eyes a late-2026 public debut, keep an eye on them.
  

• Part of a Pattern: Fits Microsoft’s global AI buildout from the UK to the U.S.
  

• Power Crunch: Texas cluster could hit gigawatt scale.
  Energy is the real choke point.
  Watch

Why it Bites: We’re in the “steel-and-substations” (hardware infra that backs AI) phase of AI.

Deals look like chip counts, but the limiting reagent is energy.

Microsoft gets supply certainty;
Startups on Azure may see quota relief.

But every new cluster now competes for transformers, cooling, and clean electrons.

This energy competition makes deciding on where to put data centers (Texas vs. EU) just as strategic as model training itself.

This is the same throughline we’ve been tracking across OpenAI/Nvidia/AMD deals:

Compute is abundant on press releases, scarce on the grid.