TL;DR 6 key\u00a0insights<\/p>\n
01 Infrastructure wins by becoming unavoidable; apps win by being loved. Only one of those creates durable\u00a0power.<\/p>\n
02 AI is shifting infrastructure from background layer to primary constraint. Demand for compute, power, and cooling is rising faster than\u00a0supply.<\/p>\n
03 Power is now the binding constraint: a 1 GW AI data center costs ~$30B to build, and a 49 GW shortfall looms by\u00a02028.<\/p>\n
04 Infrastructure is not just physical. Software that becomes embedded in workflows can be just as defensible as hardware.<\/p>\n
05 The easiest thing to build is often the hardest to defend. Feature parity arrives fast at the app\u00a0layer.<\/p>\n
06 The next unicorns will be less visible but more deeply embedded\u200a\u2014\u200aowning constraints, not interfaces.<\/p>\n
Image is generated by\u00a0AI<\/p>\n
In 1849, hundreds of thousands rushed to California chasing gold. Most returned empty-handed. The real winners weren\u2019t the miners they were the ones selling picks, shovels, jeans, and supplies.<\/p>\n
That pattern hasn\u2019t\u00a0changed.<\/p>\n
We\u2019re in a new gold rush now AI. Every week, a new startup promises to reinvent entire industries. Most will fade just as\u00a0quickly.<\/p>\n
But the companies building the \u201cpicks and shovels\u201d compute, data pipelines, networks, and power won\u2019t just survive this cycle. They\u2019ll define the next\u00a0decade.<\/p>\n
The surprising part isn\u2019t that this is happening.
It\u2019s how few people are paying attention to\u00a0it.<\/p>\n
The reality of the app economy is simple: features don\u2019t last. Anything you build can be replicated quickly, often faster than you can scale distribution. Attention is expensive, retention is fragile, and platform dependence keeps\u00a0rising.<\/p>\n
Underneath this noise, something more structural is\u00a0forming.<\/p>\n
AI is consuming resources at a pace physical systems struggle to match. Training costs are compounding. Power grids are tightening. Cooling is becoming an engineering bottleneck rather than an operational detail. Demand for specialized chips, bandwidth, and compliant data systems is outpacing supply.<\/p>\n
This is what a real constraint looks\u00a0like.<\/p>\n
And in markets, constraints are where pricing power\u00a0lives.<\/p>\n
\u201cApp features are copied in weeks. Infrastructure is shaped by physics, regulation, and integration depth.\u201d<\/p>\n
The infrastructure compounding flywheelPower Supply vs Demand\u00a0Gap<\/p>\n
The model is deceptively simple.<\/p>\n
A company identifies a bottleneck inside a large, complex system something slow, expensive, or risky to manage manually. It builds a solution that removes that friction and embeds itself directly into the customer\u2019s operations.<\/p>\n
Once integrated, switching is no longer a product decision\u200a\u2014\u200ait becomes an operational risk.<\/p>\n
Revenue scales not with attention, but with the customer\u2019s own\u00a0growth.<\/p>\n
This is the key\u00a0shift.<\/p>\n
Apps compete for engagement.
Infrastructure compounds with dependence.<\/p>\n
One survives by being loved. The other by becoming impossible to\u00a0remove.<\/p>\n
App vs Infrastructure: Defensibility Comparison<\/p>\n
Real-World Infrastructure Companies: Where They Sit in the\u00a0Stack<\/p>\n
1. AI compute and inference<\/strong> Inference costs have fallen sharply (9x to 900x depending on workload, per Epoch AI), but demand is expanding even faster. Lower unit costs are not reducing spend\u200a\u2014\u200athey\u2019re increasing usage.<\/p>\n 2. Power and energy infrastructure<\/strong> A single 1 GW AI data center costs roughly $30B. U.S. data center demand is projected to rise from ~80 GW in 2025 to ~150 GW by 2028. Morgan Stanley estimates a 49 GW supply\u00a0gap.<\/p>\n That gap is not theoretical. It\u2019s transmission lines, permitting delays, turbines, and grid capacity.<\/p>\n The winners here won\u2019t just be tech companies\u200a\u2014\u200athey\u2019ll be energy builders.<\/p>\n 3. Developer tooling and deployment infrastructure<\/strong> Once teams build around CI\/CD pipelines, preview environments, and edge functions, the workflow becomes the product. Switching platforms means rewriting habits, not just\u00a0code.<\/p>\n That is real\u00a0lock-in.<\/p>\n 4. Data governance and intelligence<\/strong> Tools like Collibra sit at the center of this shift\u200a\u2014\u200acontrolling access, lineage, and compliance across complex data environments.<\/p>\n Removing them isn\u2019t migration. It\u2019s operational surgery.<\/p>\n 5. B2B procurement and embedded commerce<\/strong> Companies like Infra.Market and OfBusiness show how deeply embedded procurement, logistics, and financing can become infrastructure. They effectively become the operating system for entire industries\u2019 purchasing behavior.<\/p>\n That is infrastructure in its purest form: invisible, essential, and deeply integrated.<\/p>\n Training cost\u00a0growth<\/p>\n 3.5\u00d7 2\u00d7 74 GW 49 GW $30B 20% Operational delays<\/strong> Capital intensity mismatch<\/strong> Security and cyber risk<\/strong> Vendor lock-in backlash<\/strong> Bull case<\/strong> Bear case<\/strong> Data Center Load Capacity Growth + Scenario\u00a0Analysis<\/p>\n Infrastructure isn\u2019t boring\u200a\u2014\u200ait\u2019s just less visible.<\/strong> \u201cCheap software\u201d becomes expensive at scale.<\/strong> The easiest thing to build is the hardest to defend.<\/strong> AI scaling trajectory Power permitting speed Hyperscaler vertical integration Sovereignty policy Open-source pressure For founders, the choice is becoming clearer: build for attention, or build for dependency. The first rewards speed. The second rewards\u00a0depth.<\/p>\n The question isn\u2019t \u201cwill people use this?\u201d For investors, this shifts focus toward businesses with deep integration, high switching costs, and revenue tied to customer expansion rather than acquisition.<\/p>\n These companies rarely look exciting early. But they compound in ways consumer apps rarely\u00a0do.<\/p>\n For enterprise buyers, the implication is simpler: choose vendors with care. The same lock-in that creates value also creates long-term dependency.<\/p>\n The internet era created giants at the application layer\u200a\u2014\u200asearch, social, e-commerce. That cycle is largely\u00a0mature.<\/p>\n The next cycle is being built underneath it.<\/p>\n Power, compute, data governance, deployment systems, and embedded commerce are being reassembled in real time under real constraints.<\/p>\n The companies building these systems won\u2019t be the most visible. But they will be the most essential.<\/p>\n And in every cycle, the same pattern\u00a0holds:<\/p>\n The visible layer gets crowded. The more this landscape evolves, the clearer one thing becomes: the most important companies being built right now are often the least\u00a0visible.<\/p>\n They don\u2019t chase users. They don\u2019t go viral. They sit inside supply chains, data flows, and power systems most people never think\u00a0about.<\/p>\n That\u2019s where the real leverage is\u00a0forming.<\/p>\n The gold rush analogy isn\u2019t just history repeating itself\u200a\u2014\u200ait\u2019s a\u00a0filter.<\/p>\n Most people chase the\u00a0gold.<\/p>\n A few quietly build the\u00a0rails.<\/p>\n And over time, the rails always\u00a0win.<\/p>\n Why the Next Unicorns Won\u2019t Be Apps<\/a> was originally published in Coinmonks<\/a> on Medium, where people are continuing the conversation by highlighting and responding to this story.<\/p>","protected":false},"excerpt":{"rendered":" The biggest companies of the next decade won\u2019t be competing for your attention. They\u2019ll be the ones quietly controlling the systems everything else depends\u00a0on. TL;DR 6 key\u00a0insights 01 Infrastructure wins by becoming unavoidable; apps win by being loved. Only one of those creates durable\u00a0power. 02 AI is shifting infrastructure from background layer to primary constraint. […]<\/p>\n","protected":false},"author":0,"featured_media":170131,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[],"class_list":["post-170130","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-interesting"],"_links":{"self":[{"href":"https:\/\/mycryptomania.com\/index.php?rest_route=\/wp\/v2\/posts\/170130"}],"collection":[{"href":"https:\/\/mycryptomania.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/mycryptomania.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/mycryptomania.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=170130"}],"version-history":[{"count":0,"href":"https:\/\/mycryptomania.com\/index.php?rest_route=\/wp\/v2\/posts\/170130\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/mycryptomania.com\/index.php?rest_route=\/wp\/v2\/media\/170131"}],"wp:attachment":[{"href":"https:\/\/mycryptomania.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=170130"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/mycryptomania.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=170130"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/mycryptomania.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=170130"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
The AI stack is creating an industrial layer around GPUs, TPUs, and specialized accelerators. Companies like Together AI are not competing on models they\u2019re building the infrastructure that makes models usable at\u00a0scale.<\/p>\n
This is the real bottleneck of\u00a0AI.<\/p>\n
Vercel didn\u2019t just simplify deployment\u200a\u2014\u200ait redefined how modern software\u00a0ships.<\/p>\n
As enterprises push data into AI systems, governance becomes critical infrastructure.<\/p>\n
Infrastructure isn\u2019t always digital or compute-based.<\/p>\nKey Metrics<\/h3>\n
The Numbers Behind the\u00a0Thesis<\/h3>\n
Annual growth in AI training costs since 2020 (Epoch\u00a0AI)<\/p>\n
Annual doubling of AI power requirements<\/p>\n
Projected U.S. data center demand by 2028 (Morgan\u00a0Stanley)<\/p>\n
Projected power supply shortfall by\u00a02028<\/p>\n
Estimated cost to build one 1 GW AI data\u00a0center<\/p>\n
Data center campuses projected to exceed 1 GW scale by\u00a02030<\/p>\nRisks<\/h3>\n
What Can Go\u00a0Wrong<\/h3>\n
Infrastructure depends on physical execution. Permits, equipment shortages, and grid access can delay revenue even when demand is\u00a0strong.<\/p>\n
Heavy upfront investment with delayed returns creates timing risk. If utilization lags, returns compress\u00a0quickly.<\/p>\n
Infrastructure sits close to critical systems. The closer you are to the core, the higher the impact of\u00a0failure.<\/p>\n
High switching costs can trigger customer resistance and push toward multi-vendor or open-source alternatives.<\/p>\nBull vs Bear\u00a0Case<\/h3>\n
Two Ways to See\u00a0This<\/h3>\n
AI creates a decade-long infrastructure supercycle
Power constraints create durable pricing power
Deep integration drives high net revenue retention
Governments fund sovereign infrastructure
Software-layer infra scales without proportional capex<\/p>\n
Hyperscalers vertically integrate the stack
AI demand growth slows faster than expected
Permitting delays stall supply expansion
Open-source compresses software margins
Overbuilding leads to utilization collapse<\/p>\nScenario Analysis<\/h3>\n
Three Futures for Infrastructure<\/h3>\n
What Most People\u00a0Miss<\/h3>\n
Three Counterintuitive Truths<\/h3>\n
The best infrastructure companies sit under large, non-discretionary budgets. They solve painful problems and earn pricing power through dependency, not\u00a0hype.<\/p>\n
Once software becomes mission-critical, the cost of failure far exceeds subscription cost. That\u2019s where pricing power quietly compounds.<\/p>\n
Apps are fast to ship and easy to copy. Infrastructure is slow to build but difficult to displace. The moat comes from integration depth, not\u00a0speed.<\/p>\nKey Variables<\/h3>\n
What to\u00a0Watch<\/h3>\n
Whether compute demand continues compounding or stabilizes under efficiency gains<\/p>\n
Grid expansion speed is the real limiter\u200a\u2014\u200anot\u00a0capital<\/p>\n
How aggressively AWS, Google, and Microsoft absorb the\u00a0stack<\/p>\n
Local compute requirements could reshape global infrastructure maps<\/p>\n
Margin compression risk in software-only infrastructure layers<\/p>\nStrategic Impact<\/h3>\n
What This Means in\u00a0Practice<\/h3>\n
It\u2019s \u201ccan they realistically operate without\u00a0it?\u201d<\/p>\nConclusion<\/h3>\n
Own the Rails, Not the Interface<\/h3>\n
The essential layer gets expensive.
And value flows to whoever controls the constraints.<\/p>\nPersonal note<\/h3>\n