Space Compute Is Moving From Science Fiction To Infrastructure

The first meaningful market will not be an AWS region in orbit. It will be the movement of processing, storage and decision-making closer to where space-based data is created.

For most of the cloud era, computing has been treated as an Earth-bound problem.

The industry built larger data centers, connected them through faster networks and improved nearly every layer of the stack—from chips and software orchestration to power and cooling. Cloud regions were placed close enough to customers to reduce latency, meet regulatory requirements and improve reliability.

That model is not going away. Most computing will remain on Earth.

But the model is becoming incomplete.

An increasing share of strategically important data is now created in space, transmitted through space or dependent on space-based infrastructure. Earth observation satellites, defense systems, communications constellations, weather platforms and remote-sensing networks are generating enormous volumes of information far from terrestrial cloud regions.

This raises a simple question: Why send every piece of raw data back to Earth before deciding what to do with it?

Increasingly, we will not.

Compute, storage and intelligence are beginning to move closer to where space-based data is created. The market remains early, and the technical and economic risks are substantial. Yet space compute is beginning to look less like science fiction and more like a legitimate infrastructure category.

The question is no longer whether space will matter to computing. It already does.

The more interesting question is how much intelligence should move closer to the data itself.

The First Market Is Not A Data Center In Orbit

The phrase “data center in space” invites the wrong mental image.

It suggests a giant floating warehouse—something resembling an AWS facility launched into orbit, filled with racks of servers and powered by vast solar arrays.

That may be part of the long-term vision. It is unlikely to be the first meaningful market.

Launching heavy infrastructure remains expensive. Maintaining it is difficult. Radiation affects electronics, heat must be managed differently, power is constrained and communications capacity is limited. Orbital debris, replacement cycles and the difficulty of repairing hardware make space a far harsher environment than Northern Virginia.

The near-term opportunity is more practical.

Satellites already collect more information than they can always transmit economically. Earth observation systems need to process imagery. Communications payloads must manage increasingly complex signals. Defense and intelligence missions require greater autonomy and resilience. Future lunar and deep-space missions will not be able to rely on constant Earth-based decision-making.

In each case, the same logic applies: process more information locally, determine what matters and transmit only what is useful.

Customers do not always want the raw data. They often want the answer.

The first wave of space compute, therefore, is not about moving the entire internet into orbit. It is about processing space-native data in space.

That is a far more credible starting point.

Why The Market Is Emerging Now

The market is not forming because of a single technological breakthrough. It is forming because several curves are moving in the same direction.

There are more satellites, and those satellites are becoming more capable. As constellations expand, space assets are shifting from rare, isolated machines into distributed networks. A larger and more connected space economy naturally requires more computing, storage and communications infrastructure.

Payloads are also becoming more software-defined.

Historically, satellite hardware was designed for a narrow function and expected to perform that function for years. Once launched, its capabilities were largely fixed. That model is becoming less attractive.

Operators increasingly want systems that can be updated and reconfigured after launch. They want to deploy new algorithms, change workloads and adapt a satellite as mission requirements evolve. In effect, they want space assets to behave less like fixed hardware and more like programmable infrastructure.

At the same time, the volume of data being generated is rising rapidly. Earth observation, synthetic aperture radar, hyperspectral imaging, signals intelligence and broadband communications can all produce more information than is practical to transmit continuously.

AI changes the equation again.

Not every AI workload belongs in orbit. Training large models in space is unlikely to be the first major use case. But inference, filtering and pattern recognition can be highly valuable at the edge. A satellite that can identify what matters before transmitting information becomes more useful—and potentially more economical.

Terrestrial data centers are also encountering constraints of their own. Land, power, cooling, grid access, permitting and geopolitical sensitivity are becoming more significant considerations in infrastructure planning.

Space will not solve those problems overnight. Nor will it replace Earth-based cloud infrastructure. But it adds another strategic layer to the global compute map.

The result is a broader infrastructure question:

Where should intelligence live?

For most workloads, the answer will remain on Earth. For a growing number of space-native workloads, it will increasingly be in orbit.

The Market Will Develop In Layers

The space compute market is often discussed as a single category. In practice, it is likely to develop through several overlapping layers.

The first—and most immediate—is onboard edge computing.

Satellites need to process data locally, compress it, filter it, encrypt it and determine what should be transmitted. These systems must be compact, power-efficient, software-defined and capable of operating in a radiation-heavy environment.

That last requirement is critical.

A powerful chip is not enough. A sophisticated software platform is not enough. The product must survive the environment in which it operates.

In space, reliability is not simply a feature. It is part of the product itself.

Storage represents another important layer.

A satellite may generate more information than it can immediately send to Earth. That data must be stored safely until a communications window becomes available or until the system determines what is worth transmitting.

Space-grade storage must withstand radiation, operate within strict power limitations and maintain data integrity over long mission durations.

Losing data in space is not like losing a file on a personal computer. The information may represent a defense signal, a mission-critical observation or an expensive collection window that cannot easily be repeated.

Communications processing is the third layer.

As satellite communications systems become more flexible, onboard digital processing becomes increasingly important. Channelization, beamforming, regenerative processing, routing and software-defined payload management can make satellites more adaptable and efficient.

This is where traditional satellite hardware begins to converge with digital infrastructure. Before large orbital data centers can become viable, satellites must first become better at processing, routing and managing information.

Orbital data centers sit at the most ambitious end of the market.

Companies are exploring off-planet storage, cloud-like computing, AI processing, data sovereignty and potentially power-intensive workloads in orbit. The opportunity could be significant if the economics work.

So could the execution risk.

Large-scale orbital infrastructure will require lower launch costs, reliable power, sophisticated thermal management, high-speed connectivity, maintainability and enough recurring utilization to justify the investment.

This segment may eventually become strategically important. It will almost certainly develop in phases.

The Competitive Landscape Is Taking Shape

Different groups are approaching the market from different starting points.

Large technology companies and commercial space leaders bring scale, capital and long-term ambition. SpaceX and Blue Origin bring launch capabilities, space infrastructure expertise and the potential to integrate vertically. Google brings experience in cloud infrastructure, chips and software orchestration.

Google’s Project Suncatcher represents one version of the long-term vision: solar-powered satellites equipped with machine-learning compute in space.

The advantage of these companies is clear. They have the technical depth and financial capacity to influence the underlying infrastructure.

Their challenge is timing.

The most ambitious versions of orbital compute depend on several technical and economic curves improving simultaneously. Launch costs, power generation, thermal management and communications infrastructure must all become sufficiently capable—and sufficiently economical.

Orbital infrastructure companies are approaching the opportunity from another direction.

Axiom Space, for example, has discussed orbital data center concepts within a broader commercial space station and in-space infrastructure strategy. Companies in this category could eventually control or operate the physical environments in which compute and storage are deployed.

Their challenge will be utilization.

Without recurring customer demand, orbital platforms risk becoming expensive capacity in search of workloads.

Storage-focused companies such as Lonestar are pursuing off-planet data storage and resilience. The proposition may appeal to governments, financial institutions and other customers concerned with continuity, sovereignty or strategic independence.

Storage can be an attractive entry point. Over time, however, customers are likely to expect storage, computing, networking and software management to work together as a unified infrastructure layer.

Traditional aerospace companies will also play an important role.

Lockheed Martin, Northrop Grumman, Airbus and Thales bring mission experience, manufacturing capabilities and deep relationships with governments. Their advantage is trust and heritage, both of which matter enormously in space.

They should not be underestimated.

At the same time, large aerospace companies have historically operated through complex, customized programs. The emerging space infrastructure market may increasingly favor modular products, faster development cycles and software-defined systems.

That creates an opening for specialized infrastructure companies focused on space computing, storage and communications processing.

Ramon Space is one example. Founded in Israel, the company positions itself around space-resilient computing systems, including onboard processing, storage and communications-related capabilities.

What makes this category strategically interesting is not simply the technology. It is the sequencing.

Specialized infrastructure companies can begin with problems that already exist. Satellite operators already need to process more information, store it reliably, communicate more intelligently and make their systems more programmable after launch.

These companies do not need customers to immediately believe in a fully developed orbital cloud. They can solve current mission requirements while building toward a broader infrastructure platform.

That distinction matters.

New infrastructure markets are rarely won solely by the company with the boldest long-term vision. They are often won by the company that solves the first unavoidable bottleneck.

A company that becomes embedded in early missions can accumulate operating experience, establish technical credibility and gradually expand its position across the architecture.

The hyperscalers may define the long-term ambition. Orbital platforms may define future deployment environments. Storage-first companies may validate specific resilience use cases. Aerospace primes will remain powerful.

But the companies building the space-grade compute, storage and communications layer are closest to a bottleneck that already exists.

What The Winning Architecture May Look Like

The competitive question is therefore not simply who has the grandest vision for computing in space.

It is which architecture best fits the first real market.

The strongest systems will likely share four characteristics: they will be software-defined, modular, radiation-resilient and vendor-neutral.

Software-defined systems can be updated after launch. That allows operators to deploy new algorithms, improve performance and adapt to changing mission requirements without replacing the physical satellite.

Space assets cannot remain frozen machines.

Modular systems can serve different use cases without requiring an entirely new architecture for every mission. Earth observation, defense, communications, lunar infrastructure and deep-space exploration will have different requirements, but they may rely on common building blocks across compute, storage, networking and software control.

Radiation resilience is essential because terrestrial assumptions break down quickly in orbit. Reliability, data integrity and power efficiency are not secondary product features.

They are the product.

Vendor neutrality may also become increasingly important.

Not every satellite operator will want to depend on a vertically integrated ecosystem controlled by a hyperscaler, launch provider or sovereign platform. The space economy will need independent merchant infrastructure—systems capable of serving multiple customers across different mission profiles.

That may ultimately become one of the market’s most important strategic positions.

Hyperscalers may define the vision. Launch companies may provide the deployment layer. Orbital platforms may create new environments for computing. Aerospace primes will continue to dominate many complex government programs.

But independent infrastructure providers could become the connective tissue between them.

Space compute will not replace the terrestrial cloud. It will extend it.

The near-term market is not about launching enormous server farms into orbit. It is about moving processing, storage and decision-making closer to space-native data.

The cloud era was built by centralizing computing in increasingly powerful data centers. Its next chapter may be defined by distributing intelligence to places where data cannot efficiently wait for the cloud.

The companies that become the default infrastructure layer for that transition may ultimately help determine how computing evolves beyond Earth.

Your Product Cycle Is Dead. Revenue Just Hasn’t Gotten The Memo

In military intelligence, one of the easiest mistakes is confusing activity with momentum. A unit can be moving, communicating and burning through ammunition — and still be losing ground.

Software markets work the same way.

Revenue can keep climbing long after a product has stopped creating the market it sits in. By the time the dashboard turns red, the strategic battle is usually already lost.

I started thinking about this seriously in late 2021, not long after launching my first fund. Public software multiples were falling apart while private prices kept behaving as though nobody had told them the war had started.

For close to two years, every slowdown got blamed on the capital cycle — rates were higher, investors had rediscovered that cash flow matters, customers were suddenly examining budgets like devout auditors. All of that was true. It also missed something more fundamental: some of these categories weren’t just being repriced. They were aging out.

A recent Invest Like the Best conversation with David George, who runs growth investing at Andreessen Horowitz, brought this back into focus for me. The capital cycle sets what investors are willing to pay. The product cycle determines whether the technology is still creating real new value for customers. Those aren’t the same question, and conflating them is how good investors miss the turn.

Investors tend to be decent at spotting the start of a product cycle — a breakthrough makes something possible that wasn’t practical before, early customers put up with rough edges because the improvement is dramatic, incumbents explain patiently why the category doesn’t matter, and TAM slides start multiplying like rabbits.

The end is harder to see. It never arrives with a press release titled “Our Product Cycle Is Now Over.” Financial results can stay attractive for years after the underlying dynamic has already shifted.

Here’s my working definition: a product cycle is ending when incremental innovation stops creating materially more value for the customer, and competition shifts to bundling, procurement and price.

The category can keep growing. The leader can keep taking share. But the thesis has changed — and investors who don’t notice are fighting the last war with the last war’s playbook.

Revenue Is Usually The Last Signal To Move

In my book, Decoding the Software Landscape, I argue that financial metrics are symptoms of deeper strategic forces, not the forces themselves. Growth, retention, margins — they tell you what’s happening, rarely why.

That gap matters most as a cycle matures. A company can post strong net revenue retention because existing customers keep adding seats or modules, even as new-logo growth quietly slows because almost every plausible buyer already owns the category. The blended number still looks fine. Underneath it, the greenfield is disappearing.

So here’s the question I’d put in every growth-stage memo: what share of new ARR comes from first-time category adoption, versus vendor replacement, versus expansion of an existing account?

Early in a cycle, customers usually aren’t replacing anything — they’re leaving spreadsheets, email or some manual process behind for a new category of software altogether, and every new logo expands the market. Think about the first wave of cloud data warehouses: a customer adopting Snowflake generally wasn’t choosing between two similar cloud products, they were escaping the operational weight of an on-prem system. Or Figma — the breakthrough wasn’t a nicer toolbar, it was browser-native, multiplayer design at a moment when emailing files back and forth was still considered normal.

Later in the cycle, one company’s growth increasingly comes out of another company’s pocket. A vendor can still post 30% growth as an excellent consolidator — but that growth isn’t coming from category creation anymore, it’s coming from winning share in a knife fight.

Video conferencing is a good illustration. Zoom pulled millions of organizations into a behavior they’d never adopted at that scale. Today a new video deal is far more likely to be a renewal, a consolidation or a displacement among Zoom, Teams, Meet and Webex. The market is still huge. Where the growth comes from has changed completely.

Cybersecurity works the same way. A company selling the first real defense against a new attack surface is creating budget. A mature endpoint vendor signing its thousandth customer is more likely replacing an incumbent or attaching a module to an existing platform. Both show up as ARR. Only one is opening new ground — and that distinction should show up in the multiple, not just the footnotes.

Listen To Why The Customer Buys

There’s a diligence question I keep coming back to because it’s uncomfortable in a useful way: what can this product do that the next-best alternative fundamentally cannot?

Early in a cycle, customers usually have a crisp answer — this is the only product with the performance, accuracy or scale to do the job, and they’ll forgive a dozen missing features because the core capability doesn’t exist anywhere else. Salesforce let you run CRM in a browser instead of maintaining another on-prem box. Twilio gave developers an API for something that used to require a telecom negotiation. Shopify let a merchant stand up a real online store without hiring anyone to build a commerce stack. Nobody bought any of these because the buttons were nicer.

As the cycle matures, the answers get softer. Customers like the interface, or the support team, or the fact that procurement already has a vendor number for this company. None of that is nothing, but it’s a sign the product frontier is closing.

I think of this as the half-life of differentiation. If a company ships something important and it takes competitors two years to catch up, the frontier is probably still open. If every meaningful feature shows up across the market within a quarter or two, whatever rents existed are already gone.

Generative AI is a live version of this experiment right now. A model or product that does something genuinely nobody else can — much better reasoning, dramatically lower latency, reliable completion of a hard workflow — can drive real adoption. But when that same capability turns up in five competing products before the enterprise even finishes its security review, the feature was real and the moat mostly wasn’t.

You also hear it in the vocabulary. Early buyers talk about adopting, building, deploying. Mature-category buyers talk about consolidating, standardizing, cutting vendors. A CIO who says “we need a vector database because our architecture can’t support this” is describing category creation. A CIO who says “we have four observability vendors and want one bill” might still hand someone a very large check — but that check is funding consolidation, not discovery, and it prices very differently.

“Good Enough” Is A Powerful Weapon

Early in a cycle, the best technology usually wins because the gap between products is enormous. Over time the gap closes — standards emerge, integrations get built, features get copied — until the difference between an excellent product and a merely good one no longer justifies another contract and another security review.

That’s when bundling gets dangerous. A startup with a product customers rate a 9 out of 10 runs into an incumbent offering a 7 or 8 inside a contract the customer already signed. The startup wants to argue about the missing points. The customer would rather skip another six-month procurement cycle. They haven’t stopped caring about the problem — they’ve just stopped being willing to pay a premium for the last bit of product quality.

Slack versus Teams is the textbook case. Slack genuinely redefined how teams communicate. But once chat, meetings, calendar and identity all sit inside the same Microsoft tenant, Teams doesn’t need to win every feature comparison — it just needs to be good enough and already there. Dropbox went through something similar: the folder that magically synced everywhere was a real breakthrough, until sync became a standard feature of Microsoft 365 and Google Workspace and the question shifted from “does it work” to “why would we approve a second vendor for this.” Dropbox’s push into search and organization through products like Dash isn’t a random pivot — it’s what a company does when it needs a new layer of value sitting on top of a core that’s become table stakes.

None of this means incumbents win automatically. An incumbent entering a category is often a sign the category matters, not that it’s dying. The real question is whether the incumbent’s version has become good enough to make the standalone product economically unnecessary. In software, “good enough” has quietly buried a lot of technically superior products.

When “Platform” Starts Appearing In Every Slide

When a category leader starts talking constantly about platforms, adjacencies and new personas, there are two explanations, and in practice it’s usually some mix of both: either the company earned the right to expand into a genuinely larger market, or the original product can’t absorb any more capital at attractive returns and management is looking for somewhere else to point it.

You see this pattern everywhere. Zoom has stretched from meetings into phone, contact center and AI. CrowdStrike has gone from endpoint to a full security platform. Shopify has moved from storefronts into payments, fulfillment and capital. Each expansion might be the right strategic call. But it forces a separate question for the investor: how strong is the original beachhead on its own, and how much of today’s valuation is actually a bet on management’s ability to build the next act.

Docusign is a good case study. E-signature created a genuine product cycle by moving contract execution off paper. As the category matured, the opportunity moved from signing itself toward the broader agreement lifecycle. In fiscal 2026, Docusign’s total revenue grew 8%, while its Intelligent Agreement Management product grew from 2.3% of ARR to 10.8% — and reached 12.6% by April 2026. None of that proves e-signature is exhausted, or that IAM becomes the next growth engine. It does tell you where the center of gravity in the growth story is moving.

Salesforce did the same thing from sales automation into service, marketing and data. Adobe did it from boxed software to subscriptions and cloud workflows. Intuit did it from tax prep into payroll, payments and lending. A strong core can fund a genuinely great second act — or it can be used to paper over a first act that’s stalling. The word “platform” on a slide doesn’t tell you which one you’re looking at; the math does.

Which is why I’d ask management directly: if you stopped launching new categories today, what would the core product’s growth rate be three years out? If the answer is 8% against a company-wide target of 25%, you’re no longer underwriting product-market fit — you’re underwriting management’s ability to invent, launch and sell an entirely new business inside the old one. That’s a much harder thing to get right than a slide with the word “platform” on it seventeen times would suggest.

New Technology Can Reopen An Old Battlefield

One idea I keep coming back to from Decoding the Software Landscape is the difference between the terrain and the architecture sitting on top of it. Businesses will always need to store information, protect assets, run workflows and make decisions faster — that need is the terrain, and it doesn’t go away. The architecture on top of it does: on-prem moved to cloud, manual integration moved to APIs, and human analysis is now moving toward machine-assisted, increasingly autonomous decision-making.

That’s the mechanism by which a mature category reopens: the object being managed changes.

Identity is a clean example. Workforce identity has always been built around employees — someone joins, gets permissions, uses some apps, eventually leaves. AI agents break that model. An agent can act autonomously on someone’s behalf, hold application permissions, exist for ten minutes, and take more actions in a day than a human does in a month. Microsoft’s Entra Agent ID now issues distinct identities and governance for agents specifically, because the old framework genuinely doesn’t fit. Identity didn’t disappear as a category — its underlying object changed, and that reopened the whole product cycle.

You can see the same shift starting in software development, where the first generation of AI coding tools suggested lines of code to a developer sitting at a keyboard. Newer coding agents take an issue, read the repo, write and test the code, and open the pull request themselves. The unit being sold is starting to shift from “assist one developer” to “complete a task” — and that can reopen testing, code review, security and infrastructure, all markets that looked completely settled.

Accounting may be next. The ledger isn’t going anywhere, but if an agent can continuously classify transactions, investigate anomalies and prepare reconciliations instead of a human doing it once a month, the product cycle around that workflow effectively restarts. Payments are moving the same direction — Stripe now documents agentic-commerce flows where an AI agent discovers a product, completes checkout and transacts on someone’s behalf. Payments aren’t new. Autonomous buyers with no human in the loop raise entirely new questions about permissioning, fraud and limits that the old rails weren’t built for.

Customer service is going through the same rewrite. A chatbot that pulls up an article is a feature. An agent that reads the account, changes the reservation, issues the refund and logs what it did has changed what’s actually being purchased — from software that helps an agent to work that gets completed without one.

A mature category can become an early market again whenever a new technological primitive changes the unit customers are actually consuming.

A Three-Front Test For Every Investment Memo

Instead of asking whether a market is “early” or “late,” I look at three separate frontiers:

1. The capability frontier — how many economically important problems remain technically unsolved?

2. The adoption frontier — how many customers or workflows haven’t adopted the category yet?

3. The consumption frontier — can each existing customer consume meaningfully more over time without the company just adding headcount to sell to more logos?

   
   

The best setup is when all three are still open.

Cloud infrastructure shows why adoption alone can mislead you: nearly every large enterprise already runs on the cloud, so the logo frontier looks mature — but AI training, inference and data pipelines keep expanding how much compute and storage each of those same customers consumes. The workload frontier is nowhere near closed.

Cybersecurity is a version of the same story. Endpoint protection is broadly deployed, but the surface being protected keeps growing — laptops, cloud workloads, APIs, identities, now AI agents. A company can sit inside an old, familiar budget line while actually attacking a brand-new consumption object.

The opposite case is a large market where all three frontiers are closing at once: most customers already have a solution, the competing products are converging, and usage is capped by a relatively fixed number of seats. Seat-based software for a stable headcount is the plain example — once every employee who needs it has it, growth has to come from price, cross-sell or taking share. Compare that to usage-based infrastructure, where one customer can create ten times the workload without hiring ten times the people. Two companies can show identical growth this year and have completely different growth physics underneath it. A large TAM slide doesn’t fix that; it just makes the slide look bigger.

Late-Cycle Does Not Mean Uninvestable

None of this means mature categories are bad investments — it means the early-cycle playbook, and the early-cycle multiple, stop applying.

Adobe is a good reminder of what a mature-category winner looks like. Creative software was never a hidden category, but Adobe built extraordinary economics through subscription pricing, file-format lock-in and a professional ecosystem nobody wants to leave. Microsoft has repeatedly turned mature categories into durable cash flow through distribution and bundling. Intuit benefits from trust and workflow depth in categories that businesses don’t casually swap out on a whim. None of these are category-creation stories anymore — they’re mature-market power stories: switching costs, distribution efficiency, pricing discipline. The returns can still be excellent. The source of the return is just different, and it should be underwritten differently.

Early in a cycle you’re underwriting product superiority and technology risk. During scaling, you’re underwriting execution and distribution. Late in a cycle, you’re underwriting consolidation, pricing power, capital allocation and whether management can credibly build an Act II. All four can work. They require different assumptions about how long growth lasts and what’s actually defending it.

In the 3X Framework from Decoding the Software Landscape, this sits inside “Nail The Target” — before deciding how a company should attack a market, you first need to know what kind of target the market has become: still being created, still expanding, or already being consolidated.

The most dangerous phrase in growth investing is “it’s still growing.” The better question is always what’s causing the growth. When customers pick a company because it does something real that nothing else can, the cycle is probably still alive. When they pick it because of the bundle, the existing relationship or the lower price, the cycle is probably ending — the income statement just hasn’t gotten the memo yet.

The Missing Map Of The Modern Software Industry

The Eight City-States and How Capital Flows Between Them

The software industry is often described as if it is constantly reinventing itself. Cloud, SaaS and now AI are each framed as a reset. But beneath the surface, the structure of the industry is far more stable. What changes is not the terrain, but the tools built on top of it.

To understand how software markets actually behave - especially in the current wave of AI-driven investment - it is useful to think of the industry not as a collection of categories, but as a continent. A system composed of distinct “city-states,” each governed by its own economic logic. Companies do not just build products; they operate within one of these cities, whether they realize it or not.

At the base of the continent sit the foundational systems - the layer where software becomes part of the environment itself. Companies like Amazon Web Services or Snowflake are not chosen repeatedly; they are embedded. Their success is not driven by features, but by indispensability. Once integrated, they are extremely difficult to replace, and over time, they quietly accumulate power.

Closely adjacent are the systems of record, the software that owns the “truth” inside organizations. Salesforce in CRM or Workday in HR are not necessarily loved products, but they are deeply entrenched. Control the record, and you control the workflow. This creates a form of durability that is less visible than growth, but far more resilient.

Another part of the continent is driven by a completely different force: fear. In cybersecurity and compliance, spending is not tied to ambition, but to risk. Companies like CrowdStrike or Palo Alto Networks operate in an environment where the buyer is not asking how to improve outcomes, but how to avoid failure. This is why, even in downturns, these budgets tend to hold.

Further along the continent, the logic shifts from protection to optimization. Some systems exist to reduce costs. Companies such as UiPath or ServiceNow often enter organizations with a simple promise: eliminate inefficiencies. The value here is measurable, which makes adoption easier - but it also makes competition harsher. If a better or cheaper solution appears, switching is rational.

In contrast, another set of companies focuses on expanding revenue. Platforms like HubSpot or Shopify succeed when they can directly tie their product to growth. This is a more aspirational category, but also a more fragile one. The central challenge is attribution. If the product clearly drives revenue, it becomes essential. If not, it is one of the first tools to be reconsidered.

Then there are systems built to inform decisions. Historically, this meant business intelligence tools like Tableau. Increasingly, in the age of AI, it means something more ambitious: turning data into action. Companies like Databricks or Palantir are attempting to bridge this gap. But as models and infrastructure become more commoditized, the real question becomes who owns the decision layer, not just the data.

A newer and rapidly growing part of the continent is defined by speed. Products like Notion or Canva succeed not because they are the most powerful, but because they deliver value almost instantly. In modern organizations, where patience for long implementations is low, time-to-value has become a competitive advantage in itself.

Finally, there are the specialized systems built for specific industries. Companies like Veeva in life sciences or Procore in construction operate differently from horizontal software. They win not by being broadly better, but by being deeply aligned with the workflows of a single sector. Over time, these businesses often become highly defensible, precisely because they are so tailored.

Taken together, these eight city-states - foundations, systems of record, risk mitigation, cost reduction, revenue expansion, decision systems, speed-driven tools and vertical solutions - form a coherent structure. Each has its own rules, and each rewards a different kind of company.

The mistake many operators make is assuming that these rules are interchangeable.

A product designed to drive revenue cannot be sold like a system of record. A cost-cutting tool cannot rely on narrative; it must continuously prove its value. A vertical solution without real domain depth will struggle to gain trust. These are not execution failures - they are structural misalignments.

This framework also helps explain how capital is moving today.

Despite the excitement around AI, capital is not flowing evenly. It is concentrated in specific parts of the continent. Infrastructure and foundational layers - such as compute, data platforms and model providers - have attracted enormous investment, because they are becoming embedded early. At the same time, there is a surge of capital into applications that promise direct outcomes, particularly those that automate workflows rather than simply assist them.

In contrast, more crowded horizontal categories are seeing increased pressure. As markets saturate, capital begins to shift toward depth - into vertical solutions where differentiation is harder to replicate and retention is stronger.

Seen through this lens, the current moment is not chaotic. It is a reallocation.

For operators, the implication is straightforward. The most important strategic decision is not just what to build, but where you are building it.

Each city rewards a different behavior. Each requires a different type of proof. And each offers a different path to durability.

The companies that endure are not necessarily the most innovative. They are the ones that understand the rules of the city they are in - and align themselves to them completely.

At Titan, this has become central to how we analyze opportunities. Not by asking whether a company is “good”, but by understanding where it sits on the continent, and how that position evolves over time.

The terrain, after all, does not change. But those who understand it can navigate it with far greater precision.