We've joined Jump Trading
I am excited to announce that the Itô team has joined Jump Trading to lead its compute trading business.
When my co-founder Alejandro (@absurdistphil) and I first started Itô, our thesis was simple: find the highest-leverage thing we could do given our skillsets and experience to solve a burgeoning problem in a rapidly changing financial and technological ecosystem.
Originally, we settled on utilizing new primitives and contract markets to hedge corporate risk that the current system simply could not support. How is one supposed to protect themselves from a particular AI supply chain risk without taking on exposure to everything else within AI Capex?
We spoke to over a thousand funds, corporations, VCs, and people in every industry, from reinsurers and AI CFOs to startup founders and hedge funds, looking for the right angle on the concentrated risk problem: where the bottleneck lies, and where it becomes exponential.
Looking back through all the pivots and product ideas, we gave it our all in everything we pursued. Countless all-nighters, flying across the country last minute just to get into a room to have a conversation, building and researching every other moment we could get. Sharing what we were learning and creating group chats along the way.
We noticed some things. At a macro level, companies are taking more and more targeted but highly leveraged risks, skewing and concentrating risk exposure to just a select few industries and other companies. You might have an idea of what I'm about to say next here.
With an ever-changing venture capital landscape and high volatility, where capital accrues and doubles down heavily on the tail ends, a growing set of companies in the early to mid stages struggle to raise rounds. This is the stage where the numbers begin to matter, but being highly dependent on a set of resources and companies makes things difficult. An era where variable costs often made up 60+% of company expenditures for the layer that sits above the models.
Too small to take on significant debt, too big for meaningful subsidies, and facing more demand than they could afford to serve.
That brings us to the big question...
Compute and Inference Costs (Volatility + Variability):
Capacity contracts commonly start at 3 to 5 year commitments with around 30% paid upfront. Meanwhile, these companies' revenue is new, usage-based, and often month-to-month, and they need the compute before the first dollar comes in. Add spiking, volatile secondary-market prices for the most in-demand chips, and the squeeze is real.
The conversation was no longer just about the Hyperscalers and Labs contracting out billions of dollars of capacity for the next 5 years of buildout. It was: how does the middle market for compute and inference form?
There's a fast-growing cohort of late-seed to Series C companies using venture capital and equity to fund compute, and they need new ways to qualify for these commitments. These clusters can carry TCVs in the low to mid hundreds of millions for the mid-market at times.
How we got here:
So we began building a platform for this market. We had also been the end buyers ourselves, training and serving models across providers. We knew what it felt like to build around these costs and constraints.
Alejandro was coming from creating credit baskets at GS. I was coming from building the #16 most-starred repository of all time, used by millions globally (the #1 agentic engineering toolkit).
Prior to that, I'd been in the academic, research, and open source space for some time, always early. Social agents and harnesses in web3, agent frameworks before then, high-dimensional data processing and transformation on HPC clusters at the San Diego Supercomputer Center for the Economics department.
Many don't know, but my history with compute goes back further. I built my first PC at 11 and later ran a PC-building and e-commerce business for years while also trading casually in neomarkets, creating tools for pricing, sourcing, and tracking restocks. We sold thousands of consumer GPUs, motherboards, CPUs, and RAM.
I was always obsessed with getting the most compute per dollar. Looking back, compute has been a through-line for a long time.
A common question we got asked was: why not operate this as a desk in a larger org?
The fundamentals and economics favor forward-thinking orgs. Good credit, risk controls, deep trading experience, hardware expertise, talent, resources, and freedom from the constraints of equity rounds or private-credit debt let us think bigger.
These things must come together, and at Jump they do. When we were presented with the opportunity to lead a new internal compute trading desk, we saw a strong path to achieve our vision of being a leading player in the compute space.
Now:
I want to thank everyone who supported us along the way, our customers, our investors, our partners and friends - and invite you to follow and work with us on this new chapter.
Looking forward to working with the companies building this market, understanding what they need, and finding ways to trade and work together. This market is forming. We’re active and trading in it today, and we want to help shape what it becomes.
For inbound: compute@jumptrading.com
Full speed ahead.
