io.net Re-poster

$250B down to $120B.

Not because the demand shrank. Because the risk got real.

NVIDIA just scaled back its guarantee for OpenAI's Ohio data center because of shareholder pushback.

Even the biggest chip maker is having to recalculate how much centralized bet-making it can absorb.

Building more compute doesn't have to mean building bigger, more centralized infrastructure.

There's already GPU capacity distributed around the world. The real opportunity is connecting it.
https://blockonomi.com/nvidia-nvda-slashes-openai-data-center-backing-by-more-than-half-amid-investor-concerns/

AI refusals aren't neutral.

They're someone else's risk tolerance built into a model you're renting.

Closed models: partial control and zero visibility.

Open-weight: your rules and your infra.

And, running it yourself is cheaper than the API bill.

Open models and open compute are the foundation for better AI.

AI's future shouldn't hinge on two companies.

But right now it does.

70% of AI revenue and 25-35% of cloud revenue for the four biggest tech companies ride on the same bet on @AnthropicAI and @OpenAI .

But as models multiply and open-source competition accelerates, that bet is looking riskier by the day.

What the market needs now is compute that's flexible, open, and globally distributed, not concentrated.

That's exactly what @ionet is building.
https://qz.com/steve-eisman-ai-boom-openai-anthropic-warning-081326

AI's future shouldn't hinge on two companies.

But right now it does.

70% of AI revenue and 25-35% of cloud revenue for the four biggest tech companies ride on the same bet on Anthropic and OpenAI.

But as models multiply and open-source competition accelerates, that bet is looking riskier by the day.

What the market needs now is compute that's flexible, open, and globally distributed, not concentrated.

That's exactly what @ionet is building.
https://qz.com/steve-eisman-ai-boom-openai-anthropic-warning-081326

300%.

That's what AI chip demand is running at, compared to supply right now.

H100 rentals are pricier today than at launch. In 2023.

More chips alone aren't the fix.

Better orchestration is.

http://io.net is here to make it happen.

Nvidia guarantees OpenAI's debt.

OpenAI leases from SoftBank.

SoftBank is funded by Nvidia's investment.

$250B backstop. $350B chip financing. Same three names.

This isn't a way to scale AI or create more access.

It's circular economics built to keep corporate profits flowing, not builders building.

@ionet was created for builders.

Three companies shouldn't get to decide who builds AI.

But they have been.

Not because they had the best GPUs. Because they controlled the only door in.

A researcher in Lagos with a brilliant model and no AWS relationship simply didn't get compute, regardless of the ability to pay.

That's not a shortage. That's a gate.

DePIN removes the gatekeeper.

Thousands of independent operators, not three hyperscalers, compete to fulfill your workload. The price is set by the market. The access is permissionless. A wallet and a training script is all it takes.

The compute oligopoly was never inevitable. It was just uncontested.

https://t.co/ZuybGWvjv9 is changing that.

64 H100s, 10 days.

@awscloud on-demand: $62,918.

@ionet: $22,886.

Same fine-tune. Two price tags.

Hyperscalers gate capacity behind enterprise agreements and quota approvals. @ionet pulls from idle GPUs across 130+ countries and spins up in minutes.

That's the difference between a platform built for builders, and one built for corporate profits.

AI infrastructure is broken.

A handful of companies own it. Everyone else waits in line.

The IDE changes the equation.

It's more than a new tokenomic model. It's the foundation for a fairer, open, and accessible era in AI compute.

Everyone, everywhere should have access to the AI economy.

@ionet makes it possible.

$15 billion per year.

That's what @AnthropicAI is paying SpaceX for compute.

That is enough money to pay for the compute needs of over 25,000 startups.

Instead, high infrastructure costs mean that only 5-10% of AI projects will make it past their first year.

But it doesn't have to be this way.

@ionet makes it possible for projects of all sizes, anywhere in the world, to get immediate access to compute at up to 70% less than hyperscalers.

So if your project's budget is less than $15 billion a year, we can help.

Affordability. Check.

Accessibility. Check.

Yes, @ionet lets you spin up GPUs in seconds at up to 70% less than hyperscalers.

But, that's only part of the story.

When you move away from centralized compute you also increase resilience, flexibility, and data security.

In a time of global instability, these matter more than ever.

@ionet CEO @Gaurav_ionet offers his thoughts on how distributed compute can help ensure critical systems stay online, even when centralized data centers go down.

Most AI teams aren’t limited by their ideas.

They’re limited by access to compute.

No matter how great the idea, the team, or the tech, if you can't get access to the right GPUs at affordable prices your project can't grow.

https://t.co/IjHEvTwGWy needed hundreds of GPUs to support real-time image generation at scale.

Using hyperscaler would have meant getting crushed by traditional cloud pricing and procurement delays.

But with https://t.co/ZuybGWvjv9 they were able to:
- Cut GPU costs by 50%+
- Provision faster
- Test new hardware sooner
- Keep scaling without slowing product velocity

The result? They grow from 14K → 19M users in a year.

Making AI affordable and accessible isn't a nice to have, it's a must have.

Affordable and accessible compute gives teams a fair chance to compete, bring their products to market, and create sustainable businesses.

Without it, we get something dystopian.

Companies now using surveillance software to tracks every click employees make at their computers to train AI to replace them.

This is what happens when you put profits over people, competition, and innovation.

@ionet we know first hand how important affordable and accessible AI is, and have built a platform to make it possible. 70% cheaper than AWS. No waitlists.

Check out our Head of Brand Strategy talk about the choices behind AI development in today's @Independent

Hyperscaler customers have caught a case of Supply Chain Paranoia.

And it's not surprising.

AI companies are hoarding GPUs or reserving capacity blocks as a business tactic against competitors.

But startups and enterprises don't need billions in the bank to tap into this idle compute.

With https://t.co/ZuybGWvRkH you bypass the supply chain paranoia.

Our decentralized GPU network gives you immediate access to the 95% of idle compute, at prices up to 70% less than hyperscalers.

If you want to build and ship without the fear, we've got you covered.

Deploy GPUs today, or whenever you need them:

Major tech companies are laying off up to 10% of their workforce, and blaming AI.

They are spending hundreds of billions of dollars on new data centers while up to 85% of existing GPUs are being underutilized due to inefficient infrastructure.

This isn't a human resources issue, it's what AI looks like when everything happens behinds closed doors and is controlled by a handful of companies.

It is AI for the few, not the many.

We believe in AI for the many, not the few.

@ionet makes underutilized GPUs from around the world instantly accessible at prices that are 70% less than major hyperscalers, so that anyone anywhere can build great products, and sustainable businesses.

Choosing the right GPUs for your project isn't about picking the "best" one.

It's about choosing the right one. For the right job. At the right time.

Each GPU has a different price/performance ratio. Understanding how to cluster them for your unique workloads can make the difference between burning through your runway, and having the resources to scale your project

Our new GPU cluster cheat sheet helps you get it right:

• H100 vs A100 vs L40S (when to use each)
• Cluster configs that actually work
• Networking + NCCL sanity checks
• Cost optimization rules that save real money

Check out the full guide:

According to a recent study, GPU utilization across enterprise servers sits is at 5%.

Yes, just 5%.

That means 95% of provisioned GPU capacity is not being used.

Hyperscalers are putting people on waitlists, costs continue to rise, billions are being spent on new data centers and utilization sits at 5%.

Something is very wrong here. We should be increasing access to AI, not hoarding it.

That's why @ionet gives you the flexibility to access affordable compute when and how you need it by orchestrating underutilized GPUs from around the world.

The US has 5,427 data centers.

That's 10X more than any other country.

This kind of centralization and concentration of power only benefits the few, not the many.

While hyperscalers continue buying up the market, the majority of the worlds developers and AI startups don't have access to the tools and resources they need to even compete.

This isn't okay. It limits opportunity. It limits innovation. And, it makes AI worse.

That's why we built a platform that makes AI accessible to everyone, everywhere.

Find out more at https://t.co/ZuybGWvjv9

Source:

Hyperscalers will soon control 2/3 of global data center capacity.

This will allow them to further control access, set terms and prices that exclude all but the largest companies, and ultimately decide who gets to participate in the AI revolution.

Except when they can't.

Open networks like https://t.co/ZuybGWvjv9 are pushing back by offering affordable and accessible compute to everyone, everywhere.

No backroom deals. No hidden costs. No gatekeeping. Transparency. Access. And prices that are 70% less.

The future of AI isn't for the few, it's for the many.

https://t.co/ZuybGWvjv9 is turning on the lights.

Haven't gotten a chance to read full the @ionet tokenomics litepaper yet?

No problem, here's the lowdown.

The IDE:

→ Replaces fixed, inflationary token emissions with a dynamic, demand driven model

→ Provides GPU supplier income in USD-targeted payouts to remove token price volatility

→ Burns at least 50% of remaining emissions supply over time, creating real scarcity tied to real usage

→ Introduces a dual-vault system that protects suppliers in downturns, absorbing surplus in upturns

Find out more: