Register interest

Investor brief

Leveraging the capability of unified-memory to lower the barrier to entry for sovereign enterprise AI.

SnapStak.ai is raising its seed round to take the CON10X AI Ecosystem to market. The investment roadmap, from seed to Series C, is shared with registered investors.

Register your interest Explore the ecosystem

Scroll to explore

Why now

The demand is ready. The data centres are not.

$600B

Sovereign AI market by 2030. McKinsey expects sovereignty requirements to shape 30 to 40 percent of all AI spending.

7 years

Wait for grid power in Europe, against about two years to build the data centre. In parts of Europe the wait reaches a decade.

0.6%

Africa's share of global data centre capacity, with 18 percent of the world's population. Power is the main constraint on new builds.

Demand is on the road map.

Nearly three quarters of enterprises have sovereign AI on their 2026 road map, and 71 percent of the executives, investors and officials McKinsey surveyed call it existential or a strategic imperative. Few yet have the plan or the budget, and sovereign cloud and AI migrations take three to four years.

Power is the bottleneck.

The IEA estimates that around one in five planned data centre projects is at risk of delay from grid constraints. Microsoft's chief executive has said the company holds GPUs it cannot plug in for lack of powered buildings, and Dublin has paused new data centre connections until 2028.

The books are out of step.

Data centre GPUs last two to three years in service but are written off over five to six, and Michael Burry puts the understated depreciation at $176 billion for 2026 to 2028. Resale prices hold up today only because memory is scarce. Bernstein expects DRAM prices to fall by about half in 2028, and once new hardware costs what used hardware does, old power-hungry GPUs lose most of their resale value. More than $200 billion has been lent against AI infrastructure, so the gap lands hardest on operators who borrowed to buy.

Sources: McKinsey, The sovereign AI agenda (opens in a new tab), December 2025, and Sovereign AI ecosystems (opens in a new tab), March 2026. IEA, Energy and AI (opens in a new tab), April 2025. AWS EMEA to Reuters, reported by The Register (opens in a new tab), February 2026. Satya Nadella on the BG2 podcast, reported by DCD (opens in a new tab), November 2025. Baker McKenzie (opens in a new tab), June 2026. Quinn Emanuel (opens in a new tab), March 2026. Michael Burry, reported by Tom's Hardware (opens in a new tab), August 2026. Bernstein, reported by Notebookcheck (opens in a new tab), July 2026.

When memory recovers

The same price drop cuts both ways.

Middle-tier providers

  • Neoclouds and debt-financed sovereign builds bought at peak memory prices, and borrowed to do it. The loans are serviced whatever the market pays.
  • Their hosted capacity competes with customers who can buy newer, cheaper hardware outright.
  • The hardware they planned to resell is outdated and power hungry, and few buyers want it.
  • Hyperscalers can absorb the write-down from other revenue. The middle tier cannot.

SnapStak.ai

  • Licenses software and owns no hardware, so there is nothing to depreciate, refinance or resell.
  • Cheaper memory makes every customer's next box cheaper, which lowers the cost of entry and widens the market for licences.
  • Customers add boxes as their users grow, at the prices of the day, from AMD, Apple or NVIDIA.
  • Capital raised goes into product and market reach, not data centres.

A hardware downturn that hurts the landlords works in favour of the licensor.

1,621

Concurrent users on one 512 GB unified memory workstation, at sixty tokens per second each on a 2K window. A calculated upper bound for the reference configuration. Adding a node adds its users without changing the others.

Where CON10X fits

Sovereign AI without the data centre.

Enterprises that want sovereign AI now cannot wait years for a data centre to be powered. CON10X serves a full precision domain model from unified memory workstations on standard office power, so an enterprise can start now and grow one box at a time.

The corpus and the weights belong to the enterprise. A model trained on its own knowledge is an asset it owns, and an asset it will want to keep.

CON10X does not replace what an enterprise already runs. Existing infrastructure and models stay in place behind the CAT parent, governed by the same four engines, alongside new unified memory hardware.

What sets it apart

What sets CON10X apart

Built in, not bolted on

The first AI platform built from the ground up with agentic security and governance designed in.

Governance acts before the agent does. Every mission is signed, every agent sits on a closed roster, and every action is requested from and granted by the orchestrator, so agents keep their judgement but hold no authority to act alone. Guardrails wrapped around other platforms inspect what a model has already done.

Unified memory, fully used

The only enterprise sovereign AI platform built to use the full capability of unified memory for maximum concurrency on weights the enterprise owns.

Indexed clusters keep only the working set in memory and give the rest to the people being served: up to 1,621 concurrent users on one 512 GB workstation, on the enterprise's own weights, at full precision.

Where we stand

Built, not planned.

The CON10X AI Ecosystem is developed and in final refinement before going to market. SnapStak.ai licenses it, so the company carries no infrastructure.

Ecosystem
Six products on four governance engines: CON10X Pythagoras, Academy and Coder, with Aegis, Prometheus and Multiverse built into Coder.
Model
Software licences only, per user in blocks of 100 at one flat rate. Licences renew and accumulate, so revenue compounds as the installed base grows.
Recognition
Selected by SAAIA as one of South Africa's top 10 startups in 2026.
Research
Five working papers on SSRN set out the theory behind the engines.
In use
CON10X 4N6, free on the Microsoft Store, runs the engines on a user's own documents.
Founder
Russel Hawkins, architect of the CON10X AI Ecosystem, brings 27 years of software engineering. In 2004 he invented OTP two factor authentication, launched on SABC2 Business Focus.

The journey

Seed to Series C in about three years.

The seed round starts it, and each round after it funds the next stage of reach. Amounts, valuations and terms are shared with registered investors.

  1. 01Year one

    Seed

    The start

    Africa first, through African sovereign AI providers.

  2. 02Year two

    Series A

    Global launch

    A presence in each market SnapStak.ai serves.

  3. 03Expansion

    Series B

    Major markets

    Expansion into the major sovereign AI markets.

  4. 04Scale

    Series C

    Global footprint

    SnapStak.ai established across the sovereign AI market worldwide.

A two-year head start.

No other sovereign AI provider has the same architecture. SnapStak.ai has two years to lock in customers before formidable competitors arrive.

SnapStak.ai

Built in Africa. Built to deploy globally.

SnapStak.ai is a South African intellectual property licensing company, registered in Gauteng. The CON10X AI Ecosystem starts with African sovereign AI providers, and nothing in it ties it to one market.
SnapStak.ai

Register interest

The full roadmap goes to registered investors.

Register below and SnapStak.ai will arrange a meeting and share the investment roadmap, from seed to Series C.

  1. ConfirmationAn email confirms your registration.
  2. MeetingSnapStak.ai contacts you to arrange a meeting.
  3. RoadmapYou receive the full investment roadmap.

Register your interest

For angel investors, funds, family offices and strategic investors.

We use these details only to reply to you. See our Privacy Policy.

This page is for information only. It is not an offer to sell, or a solicitation of an offer to buy, shares or any other securities in SnapStak.ai. Any investment is made only on the terms of signed agreements.