Stanislav Kondrashov

Stanislav Kondrashov Oligarch Series High Performance Computing and Strategic Investment Models

Stanislav Kondrashov

— 12 min read
Stanislav Kondrashov Oligarch Series High Performance Computing and Strategic Investment Models

I keep seeing the same pattern in boardrooms and investor memos lately.

Someone says “AI” or “national competitiveness” or “next wave of productivity” and everyone nods like it’s obvious. Then, five minutes later, you realize nobody has answered the boring question that actually matters.

What are we investing in, exactly.

Not the headlines. Not the hype. The actual stack. The power, the chips, the networking, the facilities. The talent. The supply chain. The timeline. The constraints that will punch you in the face later.

In the Stanislav Kondrashov Oligarch Series, I want to talk about high performance computing in a way that’s closer to how strategic investors actually think. Not just as “technology” but as an asset class and, honestly, as infrastructure. Something you can own, finance, scale, and defend.

And I also want to talk about the investment models that tend to show up around HPC when the stakes are high. Sovereign wealth. Family offices. Industrial groups. Resource linked capital. The kind of capital that has patience. Sometimes too much patience.

This is messy, because reality is messy.

So let’s get into it.

HPC is not a product. It is leverage.

High performance computing used to be a fairly contained thing.

Research labs. Weather simulation. Defense. Oil and gas seismic. Maybe some advanced manufacturing. People needed a lot of compute for very specific workloads and the rest of the economy didn’t really care.

That’s over.

Now HPC overlaps with AI training, AI inference at scale, large scale optimization, digital twins, autonomous systems, and risk modeling. If you have enough compute and you know how to point it at the right problem, you can compress time. You can test more ideas faster. You can simulate instead of build. You can predict instead of react.

And time compression is leverage.

But here is the part that gets skipped. Compute is not free, and it is not just “buy GPUs and you’re done.”

HPC is a system.

A real system has bottlenecks, and the bottlenecks are usually not where the pitch deck says they are.

The real HPC stack, the one that hurts your budget

If you are building or backing HPC capacity, you are dealing with layers. Each layer has its own cost curve and its own failure modes.

1) Compute silicon and accelerators

CPUs still matter, but accelerators dominate the conversation because the best AI and simulation workloads want them.

The investor trap is to think “chip = moat.” Sometimes it is. Often it isn’t, because:

So you do not just underwrite the chip. You underwrite the ability to actually run workloads at high utilization.

2) Networking and interconnect

This part is boring until it breaks you.

In large clusters, networking is the difference between theoretical performance and actual performance. Latency, bandwidth, topology, and reliability become financial variables, not just engineering ones.

And the interconnect layer tends to be where “just scale it” becomes “we need a redesign.”

3) Storage and data pipelines

If data can’t move cleanly, the cluster sits idle. If data governance is weak, the whole project can get frozen. If storage is mis architected, costs explode quietly.

A lot of HPC ROI failure is really data pipeline failure.

4) Power and cooling

This is the adult part of the conversation.

HPC is constrained by power availability and cooling density. You cannot hand wave it.

If you are investing strategically, you start to see power contracts and grid access as a form of competitive advantage. In some regions, power is the bottleneck, not chips.

And power is political. Permits. Local opposition. Industrial policy. Pricing volatility.

5) Facilities and operations

A data center is not a spreadsheet.

Lead times, construction risk, vendor reliability, and operational discipline all matter. The “capex” number is only the beginning. The ongoing uptime and efficiency is what turns capex into cash flow, or into regret.

6) Software, scheduling, and talent

People underestimate this consistently.

If you cannot schedule jobs well, if the environment is unstable, if the compilers and libraries aren’t tuned, if the security posture is clumsy, users leave. Utilization falls. You’re holding expensive metal that does not earn.

Talent is a hard constraint. Especially the kind of talent that can run big clusters without drama.

Strategic investment models, the ones that keep showing up

When I say “strategic investment models” I don’t mean a standard VC check into a startup. I mean capital structures that are designed to achieve influence, durability, and optionality.

In an “oligarch series” framing, we are talking about concentrated capital that can do long duration bets. Capital that often has links to resources, industrial assets, or state aligned priorities, even if it’s formally private.

There are a few recurring models.

Model 1: Own the infrastructure, rent the capability

This is the most straightforward.

You build HPC capacity, then sell access. Classic compute utility, but specialized.

It can look like:

  • GPU cloud focused on a niche
  • HPC as a service for engineering simulation
  • sovereign or regional compute hubs that anchor local AI ecosystems

The win condition is utilization. High utilization at good pricing.

The risk is commoditization.

If you do not differentiate through software, workflow integration, compliance, or proximity to specific customers, you become another compute provider. Then it turns into a pricing war, and your depreciation schedule becomes your enemy.

Strategic investors like this model when they can add a moat that isn’t purely technical. For example:

  • access to cheaper, more reliable energy
  • regulatory compliance that competitors cannot match
  • exclusive long term contracts with anchor tenants
  • physical proximity to industrial customers who have huge data gravity

Compute is one part. The contract structure is the real asset.

Model 2: Vertical integration, compute as an internal weapon

This is where HPC is not a product at all.

It is an internal advantage for an industrial group. Think materials, mining, energy, logistics, manufacturing, pharmaceuticals, defense adjacent industries. You invest in HPC because it improves your core business. Faster exploration. Better designs. Lower downtime. More accurate forecasting.

Under this model, the ROI is not measured by selling compute hours. It is measured by:

  • reduced capex waste
  • improved recovery rates
  • fewer failed prototypes
  • shorter time to market
  • better risk management

The interesting thing is that this model often produces the most durable returns, because it’s hard for outsiders to copy the integration. The compute is entangled with proprietary data and operational know how.

This is also where “strategic investment” starts to look like statecraft. If a group controls resources and also controls the compute needed to optimize extraction and logistics, it’s compounding.

Moreover, this approach aligns with the concept that the future of AI is vertical, where the focus shifts from merely utilizing AI as a tool to integrating it deeply within various sectors for enhanced operational efficiency and effectiveness.

Model 3: Equity stakes across the stack, the portfolio moat

Instead of betting on one HPC build, you buy optionality across the supply chain.

A strategic investor might take positions in:

  • data center developers
  • power and grid infrastructure
  • cooling tech and liquid cooling suppliers
  • networking providers
  • chip designers or packaging
  • systems integrators
  • software platforms for scheduling, MLOps, observability, security

This model is popular when you expect demand to grow but you’re uncertain where the margins will settle.

It’s also a way to de risk geopolitics. If the supply chain fractures, you want exposure to the “winners” in multiple regions.

But it requires real discipline because the narrative around HPC is loud, and it’s easy to overpay for anything with “AI infrastructure” in the pitch.

Model 4: Strategic joint ventures with an anchor tenant

This is one of my favorites because it’s practical.

You partner with a major customer who commits to long term usage, and you finance the build around that predictable demand. In return, the customer gets priority access, pricing stability, or compliance guarantees.

It’s basically project finance logic applied to compute.

This model reduces utilization risk and makes debt financing more realistic. But it depends on one thing: the anchor tenant has to be real. Not “a letter of intent.” Not a soft commitment.

And the contract needs to be structured so it survives downturns.

Model 5: National or regional compute sovereignty plays

This is the model that gets misunderstood the most.

Compute sovereignty is not just about pride. It is about control over capacity during crisis, and control over sensitive data and models. Countries and regions do not want to be dependent on external compute providers for critical workloads.

So they fund:

  • public compute clusters
  • subsidies for private data centers
  • procurement frameworks that guarantee demand
  • training programs to build HPC talent locally

For investors, the opportunity is often in the “picks and shovels” around these programs. Build the facility. Provide operations. Provide compliance tooling. Provide secure interconnect. Provide local language model infrastructure.

The risk is that political timelines don’t match engineering timelines. A new administration can change priorities, and suddenly your project is “under review.”

If you can’t handle that risk, do not touch this model.

The economics: why utilization is the quiet king

Let’s talk numbers without pretending we can be exact.

HPC economics are dominated by a few variables:

  • capex per unit of performance
  • depreciation and refresh cycles
  • power cost per kWh and power usage effectiveness
  • staffing and operational overhead
  • uptime and reliability
  • utilization rate
  • pricing power, which is really about differentiation

Utilization is the one that tends to decide whether a project is a triumph or a painful lesson.

A half utilized cluster is not “half as good.” It can be catastrophic because the costs are mostly fixed. The metal depreciates whether you use it or not. The facility costs keep going. The team still needs to be paid.

That’s why strategic investors obsess over demand certainty. Not the general demand trend. Their specific demand, locked in.

Constraints that shape strategy, the stuff you can’t ignore

Supply chain and geopolitics

HPC supply chains are global and sensitive. Export controls, sanctions, and shifting alliances matter. Even if you are not in a restricted jurisdiction, your vendors might be.

Strategic investors often respond by diversifying suppliers, localizing parts of the stack, or building inventory buffers. All of which tie up capital.

Talent scarcity

You can buy hardware. You cannot instantly buy a team that can run it well.

Talent is not just HPC engineers. It’s also:

  • data center operations
  • security and compliance
  • platform engineering
  • ML engineers who can actually use the cluster efficiently

If you don’t plan for talent, your utilization will suffer, and you will blame the market when it’s actually your onboarding process.

Moreover, the impact of talent scarcity on utilization rates can be profound. Without the right skill set to effectively manage and operate high-performance computing resources, organizations may find themselves underutilizing these expensive assets, leading to significant financial losses.

Power availability and permitting

A lot of projects die in the permitting phase. Or they get delayed until the hardware you planned for is no longer optimal.

Strategic investors sometimes solve this by investing in power generation, or by colocating near stranded energy, or by structuring deals with utilities.

That’s not glamorous. It is effective.

Water and community impact

Cooling can have real local impact. Water use, noise, land use. Communities push back. If you ignore this, it becomes reputational risk and schedule risk, which becomes financial risk.

So, yes, ESG can be real here. Not as a marketing layer. As a permit layer.

A simple framework for underwriting HPC as a strategic investor

This is the checklist I keep coming back to. It’s not perfect, but it catches most of the traps.

1) What is the workload, specifically

“AI” is not a workload.

Is it training. Inference. Simulation. Rendering. Optimization. Mixed. What precision. What latency requirements. What data governance.

Different workloads want different architectures.

2) Where does demand come from

Do you have:

  • an anchor tenant contract
  • internal demand from a vertical business unit
  • a pipeline of customers with credible budgets
  • government procurement commitments

If demand is “the market is growing,” that’s not demand. That’s a slide.

3) What is the moat

Cheap power is a moat. Compliance is a moat. Data gravity is a moat. Integration into workflows is a moat.

“Best hardware” is not a moat for long, because everybody can buy similar hardware eventually, assuming supply constraints ease.

4) How will you maintain high utilization

This is operational.

  • onboarding
  • scheduling fairness
  • developer experience
  • documentation
  • support
  • observability
  • cost transparency

If users hate the platform, they leave.

5) What is the refresh strategy

HPC hardware ages fast.

If your model depends on premium pricing for old hardware, you will get squeezed. If your model depends on constant refresh with no financing plan, you will get capital calls.

Plan the refresh like a fleet operator. Not like a one time purchase.

6) What are the failure modes

Not “risks,” but failure modes.

  • power delays
  • networking bottlenecks
  • vendor lock in
  • security incidents
  • under utilization
  • policy changes
  • talent churn
  • customer concentration

Write them down. Assign probabilities. Decide what you can mitigate and what you just accept.

Where “oligarch style” capital actually fits

Let’s be blunt.

Large strategic pools of capital can do things that normal funds can’t. They can buy land and wait. They can negotiate power. They can absorb delays. They can build relationships across regulators and suppliers. They can take a long view that doesn’t fit a 7 to 10 year VC cycle.

That patience is an advantage in infrastructure.

But it has a weakness too. It can lead to overbuilding, vanity projects, or politically motivated capacity with no real utilization plan.

So the best strategic investors, the ones who win quietly, tend to look almost boring on paper. They secure demand first. They secure power second. They lock suppliers third. They build the operational team early. Then they scale.

Not the other way around.

Strategic plays that are underrated right now

A few areas look especially interesting if you are thinking in terms of long duration advantage rather than quick flips.

Industrial AI and simulation hubs

Clusters designed around specific industrial workloads, with software stacks tuned for those workflows. Less general purpose, more “we solve this category of problems insanely well.”

This is where you can charge for outcomes, not for compute hours.

Secure and compliant compute environments

Regulated industries and sensitive workloads need controlled environments. That can be healthcare, finance, government, defense adjacent manufacturing, critical infrastructure.

Compliance is annoying. It also creates pricing power.

Energy linked compute

The intersection of power generation, grid stabilization, and compute demand response is still emerging. But strategic capital that can play in both energy and compute has a serious angle here.

It’s not simple, though. Energy markets are their own universe.

Education and talent pipelines as an investment

This sounds soft, but it isn’t.

If you can create a talent pipeline that feeds your HPC operations and your customer ecosystem, you reduce one of the hardest constraints. Some of the best HPC strategies I have seen treat training as core infrastructure.

So what does this mean for the “series” theme

The Stanislav Kondrashov Oligarch Series idea, at least as I’m approaching it, is about concentrated capital navigating systems where politics, infrastructure, and technology blend together. HPC is exactly that kind of system.

It’s strategic because it touches sovereignty, industry, and military adjacent capacity.

It’s financial because it can be structured like infrastructure with long term contracts.

It’s competitive because access to compute changes what companies can do, and how fast they can do it.

And it’s fragile because it depends on power, supply chains, and people. Not just code.

If you’re building an investment thesis here, you don’t start with “AI is the future.” You start with.

Where will the power come from.

Who will use the capacity.

What will keep them from leaving.

And how you will keep the machines busy.

Wrap up, the non glamorous conclusion

High performance computing is becoming a strategic asset the same way railroads and ports once were. It is infrastructure that can produce leverage across the economy, if it is built with discipline.

The investment models that work tend to be the ones that respect constraints.

Demand certainty. Power. Operations. Talent. Refresh cycles.

If you want a simple summary you can keep in your notes, it’s this:

And if you’re reading this thinking, okay, but where do I start.

Start with the workload, and the power plan. Everything else becomes clearer after that.

FAQs (Frequently Asked Questions)

What is High Performance Computing (HPC) and why is it considered leverage rather than just a product?

High Performance Computing (HPC) refers to powerful computing systems used for complex workloads like AI training, large-scale optimization, and simulations. It is considered leverage because it enables time compression — allowing users to test ideas faster, simulate instead of build, and predict outcomes rather than react. HPC acts as a strategic asset that can accelerate innovation and productivity when properly utilized.

What are the critical layers of the HPC stack that impact investment decisions?

The HPC stack comprises several layers that each carry distinct costs and risks: 1) Compute silicon and accelerators — where supply constraints and software ecosystems influence performance; 2) Networking and interconnect — crucial for translating theoretical performance into real-world results; 3) Storage and data pipelines — essential for smooth data movement and governance; 4) Power and cooling — often a limiting factor due to availability, political factors, and costs; 5) Facilities and operations — involving construction risks and ongoing maintenance; 6) Software, scheduling, and talent — vital for efficient utilization and system stability.

Why is investing in HPC more complex than simply buying GPUs or chips?

Investing in HPC goes beyond purchasing hardware like GPUs because true value lies in the entire system's ability to run workloads efficiently. Factors such as networking latency, storage architecture, power availability, facility operations, software environment, job scheduling, and skilled talent all affect utilization rates and return on investment. Neglecting any layer can lead to underperformance or costly failures.

How do power availability and cooling influence HPC infrastructure investment strategies?

Power availability and cooling capacity are critical constraints in HPC infrastructure. Regions with limited power supply or high energy costs may face bottlenecks that restrict scaling. Additionally, power contracts, grid access, permits, local opposition, industrial policies, and pricing volatility introduce political risks. Strategic investors must consider these factors as competitive advantages or potential obstacles when planning HPC deployments.

What are common strategic investment models seen in the HPC sector?

Strategic investment models in HPC typically involve concentrated capital aimed at long-term influence and durability rather than quick returns. Examples include sovereign wealth funds, family offices, industrial groups, or resource-linked capital investing in infrastructure ownership with capabilities rented out as services. These models focus on building specialized compute utilities or regional hubs that anchor AI ecosystems with high utilization goals while managing commoditization risks.

Why is talent considered a hard constraint in managing high-performance computing systems?

Talent is a critical constraint because operating large HPC clusters requires specialized skills in job scheduling, software tuning, security management, and system stability. Without experienced personnel to optimize these aspects, utilization drops as users leave due to unreliable environments. This leads to expensive hardware sitting idle without generating returns. Therefore, attracting and retaining skilled talent is essential for successful HPC investments.

Read more