The Scene
HUMAIN CEO Tareq Amin recognizes Saudi Arabia’s latest effort to transform its economy for what it is: an audacious bet that it can join the US and China as a global AI superpower.
asserted
it → recognize → superpower
Ahead of US President Donald Trump’s visit to the Kingdom last year, the Jordanian-American tech executive helped launch the state-backed AI company under Saudi’s Public Investment Fund.
asserted
executive → help → Fund
The plan was to use the Kingdom’s cheap energy, vast land, and sovereign billions to become one of the world’s largest compute hubs.
asserted
plan → use → hubs
Amin, who previously led Saudi Aramco’s tech arm, clocked an early win with a well-timed $3 billion investment into Elon Musk’s xAI before it was folded into SpaceX ahead of its historic IPO.
asserted
it → lead → IPO
Now, HUMAIN has plans to launch a large venture capital fund and expand significantly with offices in the US, France, and the UK, in order to “accelerate the pace of investment,” Amin said in an interview.
asserted
Amin → have → interview
It also wants to reduce its reliance on the kingdom’s sovereign wealth fund, its largest shareholder, open up to more global investors, and position itself as a sort of “Switzerland” for open models that sidesteps the dueling ambitions of the US and China.
asserted
that → want → US
Here’s how Amin is thinking about what’s next for HUMAIN:
asserted
what → ’ → HUMAIN
This interview has been edited for clarity and brevity.
asserted
interview → edit → clarity
How much external financing have you raised so far?
asserted
you → raise → financing
We’re not there yet, but I’m getting close to it,.
asserted
I → ’re → it
I would love nothing else but to open HUMAIN to the rest of the world and say this is a great place to invest, a great company to invest in.
asserted
this → love → world
The story that we talked about a year ago was a PowerPoint, it was a vision.
asserted
it → talk → that
Today it’s no longer a vision, nor a PowerPoint.
asserted
it → ’ → ?
People can see it with their own eyes and understand the value proposition of what we offer.
asserted
we → see → what
When you offer predictable energy tariff rates, an abundance of land, low-cost energy, and a [total cost of ownership] that is one of the lowest in the world, it becomes a no-brainer to start looking seriously at investing here.
asserted
it → offer → investing
I can’t tell you the exact number, but you could assume today an average build of $11 to $12 million per megawatt is what the industry builds on for [data centers].
uncertain
industry → tell → centers
What we’ve declared is a build towards 3 gigawatts, and you could really stipulate how much capital we need.
uncertain
we → declare → capital
On one hand, we do build-to-suit for hyperscalers and they purchase the GPUs.
asserted
they → do → GPUs
On the other hand, I’m really accelerating our AI inferencing, which is endpoint APIs to sell frontier models in our marketplace.
asserted
which → accelerate → marketplace
But for third-party financing, I am extremely pleasantly surprised that we have large companies such as Blackstone that have committed to put $3 billion, and many others that will be coming in to accelerate our debt and equity raise for the company.
asserted
that → have → company
The US has temporarily cut people off from access to some of the models, which has made everybody think very seriously about the idea of sovereign AI.
asserted
everybody → cut → AI
What does that mean for HUMAIN and how are you thinking about it?
asserted
you → mean → it
It’s not trivial to build frontier models.
asserted
It → ’ → models
You may take an approach and say, ‘Look, we’re going to try to create from scratch an Arabic large language model.’
uncertain
we → take → model
I think it did relatively well; it taught my team how the transformer model architecture works.
asserted
architecture → think → team
But when we created Allam [an Arabic-first large language model], we used about 4,000 GPUs.
asserted
we → create → GPUs
Compare that to OpenAI that has hundreds of thousands of GPUs.
asserted
that → compare → GPUs
Practically speaking, there are two big constraints to building a frontier model: You need scientists and researchers that have the depth and know-how, and you need large compute.
asserted
you → speak → compute
We did a strategic partnership with a leading company in the US called Reflection AI.
asserted
We → do → US
That relationship does two things: it continues to educate my own team on how we tune, optimize, and improve Allam to be a frontier model, and they provide us the sovereignty that we need and require to do this.
asserted
we → do → this
In addition, we made a big announcement with Mistral.
asserted
we → make → Mistral
We basically want to be a marketplace — think of it as a Switzerland for open models.
asserted
We → want → models
Open models are becoming fascinating; their performance is absolutely reaching frontier levels.
asserted
performance → become → levels
With this open-weight architecture, I feel confident that we have an answer regardless of whatever geopolitical scenario comes our way.
asserted
scenario → feel → answer
I don’t think anybody in [the Middle East] yet could say they have built a state-of-the-art frontier model.
uncertain
they → think → model
Europe is trying, and I think eventually we will catch up, but
asserted
we → try → ?
we need a bit of time.
asserted
we → need → time
So short term, my open-weight models are the strategy to guarantee sovereignty.
asserted
models → guarantee → sovereignty
Going back to what you were saying about how fast you’re moving and the demand for compute capacity, how is that making you think about your data center targets?
asserted
you → go → targets
My entire capacity is sold before it’s deployed.
asserted
it → sell → ?
…and 12 more, not listed.
The Aporia
