Artificial Intelligence Ecosystems

Microsoft CEO predicts the end of traditional software. The money is flowing somewhere else... A $273B transformation few people are talking about. A pivot is coming. But not the one everyone expects. Here's what's happening. The Numbers Tell The Story: → $273B SaaS market "at risk" → $220B spent on AI infrastructure in 2024 → $10B in new AI revenue for Microsoft alone → $500B investment for OpenAI's Project Stargate Three shifts reshaping the industry. 1 - Value Creation is Inverting The Old World: → Infrastructure was a commodity, servers and storage. → Applications captured value with 80%+ margins. → Features created moats, being hard to build and copy. → Distribution controlled customers with enterprise ties. The New Reality: → Infrastructure captures value with chips & clusters. → Applications commoditize as AI accelerates dev. → Domain expertise is the moat with expertise. → AI tools get mass distribution by solving real problems. 2 - Enterprise Advantage Remains What's Actually Sticky: → Security with data, governance, and audits. → Complex workflows with decision trees and oversight. → Domain knowledge with industry specific process. → Enterprise reliability with uptime, scale, and SLAs. This Explains Why: → Salesforce’s grew Agentforce with key insights. → Microsoft adds $10B in AI revenue via trust and infra. → OpenAI sticks with seat-based pricing for enterprises. → Enterprise SaaS rebounds first due to customer loyalty. 3 - The Real Transformation What's Dying: → Generic point solutions are replaced by AI platforms. → Seat-based pricing fades as AI lowers costs. → Feature competition disappears as AI copies fast. → Integration moats weaken as AI links systems. What's Emerging: → Domain-specific agents with deep expertise and AI. → Outcome-based pricing focused on results. → Capability-based competition solving real problems. → Expertise-driven moats compounding knowledge. The Uncomfortable Truth: The winners won't be pure infrastructure players or traditional SaaS companies. They'll be hybrid organizations that: → Own key infrastructure like compute and AI models. → Build domain expertise across industry and tech. → Deliver real outcomes, not just features. → Own customer relationships through trust and results. Look at your software company today: → Do you own infrastructure like compute and models? → Do you have deep industry and tech expertise? → Do you deliver real outcomes, not just features? The $273B isn't just redistributing. It's reorganizing entirely. The question isn't whether agents will replace SaaS. Are you positioned to capture value in the new stack?

The real AI war is being fought in the deployment layer. While everyone obsesses over GPT and Claude, these platforms, which process billions of inferences daily, are quietly determining who actually wins in AI. In just 12 months, the AI deployment landscape transformed more dramatically than cloud infrastructure did in 5 years. Big one-year changes in Mosaic scores (company health and trajectory metric) across the model deployment & serving market signal a fundamental reshaping of the AI infrastructure landscape. Market leaders redefining AI deployment: → Databricks dominates with the highest Mosaic score and $100B valuation, reaching $2.6B revenue with 60%+ growth → Baseten’s rapid rise just attracted a fresh $150M in funding, driven by their serverless GPU infrastructure → Together AI capitalized on generative AI demand, raising $533.5M at a $3.3B valuation with in-house LLMs using reinforcement learning → VESSL AI and Modal are winning with pay-per-use GPU compute models Current market leaders are split into distinct camps that will likely converge or consolidate sooner than we all expect. → Infrastructure specialists like Together AI and Fireworks AI focus on serverless inference for production environments. → Platform plays like Databricks leverage their existing enterprise relationships and massive resources to both build and buy innovation. → Developer-centric players like Modal attract startups with zero fixed costs. The winners share proven technical foundations driving their success: ↳Scale: Hugging Face hosts 500,000+ models for 5 million developers ↳Architecture: Serverless infra eliminates DevOps complexity (Baseten, Modal, Fireworks) ↳Business Model: Pay-per-use pricing removes barriers for growing startups ↳AI-Native: 80% of Databricks’ new databases are now AI-created vs. 30% last year ↳Generative AI Focus: Together AI and Fireworks built specifically for LLM inference demands Critically, these platforms combine efficient compute, intelligent orchestration, and developer-friendly abstractions – creating defensible moats against hyperscaler competition. In turn, this makes these companies prime acquisition targets for the established cloud leaders. With 96% of enterprises deploying AI models (up from 25% in 2023), infrastructure choice has become strategic. Massive YoY revenue growth numbers across both the hyperscalers and emerging players demonstrate the market's trajectory. While the world debates which LLM is smartest, the companies controlling how those models actually reach users are building the real moats. Incredible recent funding rounds and major acquisitions (Nvidia acquiring OctoAI) will define which platforms become the become dominant AI infrastructure players. P.S. Want more insights on the companies powering AI deployments? Drop "deploying" in the comments for *free* access to CB Insights' data and insights on the model deployment & serving market.

In a seismic shift for the AI industry, OpenAI co-founder Sam Altman is betting that radical transparency—not proprietary guardrails—will cement his company’s dominance. But will giving away the crown jewels backfire? The Wall Street Journal — This analysis examines OpenAI’s counterintuitive strategy to combat rising competition from Chinese AI firm DeepSeek AI, leveraging unprecedented openness in a field once defined by secrecy. 🔮 Open-Sourcing the Unthinkable OpenAI has begun releasing foundational AI architectures previously considered too dangerous for public access, including advanced reasoning frameworks and multimodal training blueprints. This strategic disarmament aims to undercut DeepSeek’s market position by flooding the sector with state-of-the-art tools—a calculated risk that redefines what “competitive advantage” means in AI. ⚖️ The Ethics Earthquake By open-sourcing models capable of synthesizing complex chemical compounds and analyzing geopolitical scenarios, OpenAI has ignited fierce debate about responsible innovation. Internal documents reveal heated boardroom debates over whether this democratization empowers benevolent researchers or arms bad actors. 🌐 The New AI Cold War The move directly counters DeepSeek’s rapid advances in generative video AI, with leaked emails showing Altman telling staff: “If we don’t break our own monopoly, others will”. Industry analysts note this mirrors geopolitical tech strategies, where controlled proliferation maintains influence over chaotic development. 🧠 Developer Ecosystem Gambit OpenAI’s surprise release of “Model Forge”—a toolkit for building AI assistants with emotional resonance—has already been adopted by 14,000+ developers in its first week. The play: become the indispensable infrastructure layer for AI innovation worldwide, making competitors’ products reliant on OpenAI’s open-source bedrock. 🕳️ The Profitability Paradox While releasing core IP, OpenAI quietly unveiled new premium services for enterprise-scale AI alignment validation—a classic “give away the razor, sell the blades” approach. Early adopters like Pfizer and Airbus are already paying seven figures annually for these certification services, suggesting a blueprint for monetizing openness. This tectonic shift in AI strategy continues to unfold, with regulators scrambling to adapt to an ecosystem where yesterday’s dangerous capabilities are tomorrow’s open-source building blocks. #AIStrategy #OpenSource #TechInnovation #AIEthics #DeepTech #FutureTech #AICompetition #TechDisruption #OpenAI #DeepSeek

80% of companies admit to major gaps in data preparation while investing heavily in AI. Most are about to waste money on tools that can't access the data they need. DHL learned this the hard way. Their AI voicebot kept missing "Ja" - the German word for "yes." One tiny data processing failure taught them everything about infrastructure gaps. Today, that same voicebot handles a million calls monthly, resolving half without human intervention. The difference? They fixed the plumbing first. Coca-Cola took a different approach from day one. No 50,000-person AI bootcamp. No consultant army. They created a sandbox with exactly six people: legal for data governance, communications for internal messaging, and tech for sandbox infrastructure. This specific combination works because legal sets data boundaries, comms manages change resistance, and tech builds safe testing environments. That sandbox birthed Create Real Magic and showed the company what was possible before anyone bet serious money. Here's what works: 1. Fix your data infrastructure before buying AI tools. DHL's "Ja" problem shows why data quality matters more than model sophistication. 2. Build your own six-person team. Legal handles data access and compliance. Communications manages internal expectations and resistance. Tech creates isolated testing environments with real but low-stakes data. 3. Use two-way doors for decisions. Reversible experiments get approved fast. Irreversible bets get reviewed thoroughly. Don't let irreversible review processes slow every reversible idea. Both companies treated AI as infrastructure, not press releases. They built working systems instead of making announcements. Every organization sits somewhere on a simple grid: infrastructure readiness multiplied by urgency. I've mapped this into four zones (see chart below): - Stagnation (low urgency/low readiness): drifting toward irrelevance - Complacency (low urgency/high readiness): resources without motivation - Frustration (high urgency/low readiness): action without infrastructure - Innovation (high urgency/high readiness): where transformation happens As Cisco's Jeetu Patel warns: "Eventually there will be only two kinds of companies: those that are AI companies, and those that are irrelevant." My article is just one of many thought-provoking contributions in this I by IMD issue. The whole magazine is packed with ideas worth exploring (link in comments). P.S. Comment below saying 'free trial please' and I'll DM you the QR code for 3-month free access to our school's I by IMD Magazine.

The next massive software category isn't built for humans; it is built for AI agents. For decades, we optimized software for human eyes and hands. Today, human processing speed is the primary enterprise bottleneck. Autonomous agents can now research, negotiate, and execute complex workflows in milliseconds. They do not need graphic dashboards. They require machine-to-machine infrastructure to communicate, collaborate, and transact natively. We are rapidly moving from a human-to-human (H2H) software architecture to an agent-to-agent (A2A) ecosystem. Consider the emerging agent-native toolstack: - AgentMail: Dedicated email infrastructure that allows AI agents to parse, send, and orchestrate asynchronous workflows entirely via API. - Moltbook: A specialized social forum where millions of agents interact, share data, and validate operational capabilities without human intervention. - OpenClaw: An open-source framework enabling these agents to autonomously execute secure tasks across varied enterprise environments. To build a durable AI strategy, leaders must prepare for this infrastructure shift. Here is how you can adapt: 1. Audit API Readiness: Legacy software lacking robust APIs will stall your automation efforts. Inventory your core systems to ensure they can communicate securely with external agents. 2. Update Procurement Rules: Stop evaluating enterprise software solely on user experience. You must prioritize machine interoperability and "agent-friendliness" in your next vendor assessment. 3. Launch an A2A Pilot: Isolate one high-friction, data-heavy workflow. Deploy an internal agent sandbox to handle the initial data processing and routing before a human steps in. Are you building infrastructure for your future digital workforce, or just buying faster dashboards for humans? #ArtificialIntelligence #AIAgents #EnterpriseAI #Innovation #FutureOfWork

AI's biggest bottleneck isn't capability. It's governance. While tech headlines celebrate each marginal improvement in model performance, the real world often tells a different story. JPMorgan built their own inferior LLM suite entirely in-house, refusing to use OpenAI despite its superior performance. Why? Because when you're handling millions of financial records, a black box —no matter how sophisticated — is a huge risk. This pattern repeats across industries. The obstacle to AI adoption isn't whether the technology works but whether it can be trusted with sensitive data and critical decisions. Lorenza Binkele at SecureAIs spotted this gap early. Instead of building another AI model, she positioned her company in the space between data and AI—providing sanitization, governance, and compliance infrastructure. Her approach challenges conventional startup wisdom in three ways: 1️⃣ She rejected the all-or-nothing platform strategy. When enterprise sales cycles dragged, SecureAI broke their offering into modular components that could be adopted individually. Sales accelerated immediately. 2️⃣ She identified the wedge that opens multiple markets simultaneously. The same technology that helps banks with compliance also helps legal teams process documents and healthcare institutions share research data. 3️⃣ She recognized that the 23andMe bankruptcy wasn't an isolated incident but a preview of AI's future risk landscape. Every company using AI tools is potentially one vendor bankruptcy away from a catastrophic data breach. The real opportunity isn't building the next ChatGPT, but becoming the essential layer all AI depends on. This positions founders to: → Benefit regardless of which AI models ultimately win → Avoid direct competition with tech giants → Build defensible infrastructure that increases in value as AI advances → Target the projected $52 billion AI governance market In the AI gold rush, the biggest opportunities are in building the infrastructure that all AIs need to function safely. The conventional wisdom says focus on agents, models, and applications. The contrarian bet is on the layer that makes those things trustworthy enough for enterprise adoption. Go-to-market success in AI is fundamentally about becoming essential infrastructure rather than just another application. #startups #founders #growth #ai

Everyone’s chasing GPUs. Amazon just secured 1.9GW of clean power... While the #AI world debates models, Amazon is playing a different game: Control the energy, control the infrastructure. In June 2025, Amazon committed $20 billion to transform #Pennsylvania into a core node of its AI cloud network. But this was not just a real estate play. It was a masterclass in AI-era infrastructure strategy. Here's what Amazon did: 1. Selected a site next to a nuclear plant 2. Restructured a 1.9GW power deal after regulators blocked a direct connection 3. Locked in carbon-free baseload energy through 2042 4. Partnered with the state on fast-track permitting, workforce development, and public-private planning 5. No tax breaks. Just execution. This is the new blueprint. In the past, data centers followed fiber and land. Today, they follow power and permitting velocity. Amazon’s Pennsylvania pivot tells us three things: 1. #AIinfrastructure is now energy infrastructure. 2. The bottleneck is no longer land. It's megawatts. 3. States (or countries) that streamline execution will win the next wave of hyperscale investment. If you're building data centers without a long-term power strategy, you're already behind. If your state doesn't offer fast permitting and firm energy access, you're not on the map. The future of AI will not just be trained in the cloud. It will be built where the power is. #datacenters

The AI infrastructure race is no longer theoretical. The companies that win will not be defined only by models, chips, or cloud platforms. They will be defined by who can secure power, land, fiber, construction capacity, and speed-to-market at scale. Hyperscale growth is now a physical infrastructure execution challenge. Power availability, diverse network routes, campus scalability, permitting timelines, supply chain discipline, and construction certainty are becoming competitive advantages. Without that foundation, AI demand cannot be monetized at the pace the market expects. The next phase of AI will be won by the operators, developers, utilities, fiber providers, and capital partners that can turn ambition into deployed capacity. AI may be digital. But the backbone is very real. #DataCenters #AIInfrastructure #DigitalInfrastructure #Hyperscale #CloudComputing #ArtificialIntelligence #Fiber #Power #DataCenterConstruction #Technology #befound

The AI architecture crisis nobody's talking about! Every enterprise is building AI solutions right now. The problem? We're creating a mess that'll take years to untangle. I'm watching organizations speed-run the same mistakes we made during cloud migrations, except faster and messier. Teams are shipping AI features in isolation. Marketing has their chatbot. Engineering built their document search and coding assistant. Sales is piloting something with a different LLM provider. Finance just approved three separate AI vendors. Nobody's talking to each other. The result? AI sprawl. Each team solving identical problems, authentication, prompt management, cost monitoring, data security, from scratch. We're building technical debt at unprecedented speed. But here's the thing - it doesn't have to be this way. Organizations getting this right aren't moving slower. They're building smarter foundations that let teams move faster. So how do we avoid this? 1. Start with an abstraction layer Build an LLM gateway that routes requests based on task requirements. Need complex reasoning? Route to the expensive model. Simple classification? Use the fast, cheap one. Teams don't rewrite code when you switch providers. 2. Implement Model Context Protocol (MCP) This is the game-changer! MCP standardizes how LLMs connect to your data and tools. One integration to your CRM, your docs, your databases, and every AI application can use it. No more rebuilding connectors for each use case. 3. Create a shared RAG infrastructure Stop letting each team build their own vector database setup. Centralize the foundation: Teams customize on top, but they're not rebuilding the foundation every time. 4. Treat prompts like production code Version control. Testing. Peer review. If a prompt drives business logic, it needs the same seriousness as any other code. Most orgs aren't doing this. Build lightweight governance that enables speed! - Define clear security and data handling standards - Set cost thresholds that trigger reviews - Create an AI inventory (you can't manage what you can't see) - Let teams innovate within those guardrails 5. Implement FinOps from day one Token costs aren't like normal compute. They scale unpredictably. Tag everything. Monitor everything. Create visibility before bills become problems. Form an AI Center of Excellence (but keep it lean) Not a committee. Not a bottleneck. A small team that: - Maintains shared libraries and patterns - Prevents duplicate problem-solving - Enables teams rather than gatekeeping them The technical foundations (LLM gateway, MCP, unified RAG) give you the biggest leverage, they let teams move independently while maintaining architectural coherence. Most organizations are six months into building AI solutions with no architectural strategy. The mess is already there. So, will you architect properly now or will you wait for the disaster? #EnterpriseArchitecture #SolutionArchitecture #AI #LLMOps #TechLeadership

Everyone's chasing AI. The smartest founders I know are building what AI needs to exist. Last week, a founder pitched me their "AI for sales" startup. They were the 7th AI pitch I'd seen this month so far. I asked one question: "What's your monthly compute bill?" "$287,000. And growing." That's when I showed them where the real opportunity was hiding. The infrastructure paradox: Every AI startup needs: GPUs they can't get Data centers already at capacity Specialized compliance tools Cost optimization they can't build The gold rush is obvious. The shovel shortage? That's where fortunes get made. 1849: Levi Strauss didn't mine gold. He sold jeans to miners. 1990s: Cisco didn't build websites. They sold the routers. 2000s: AWS didn't create apps. They rented the servers. Today's version? Look at what's actually getting funded: GPU scheduling optimization AI model monitoring platforms Specialized cooling systems Compliance and governance tools Not getting funded: "ChatGPT for [insert industry]" What I tell every founder: You don't need to predict which AI company wins. You need to sell to all of them. One founder pivoted from "AI-powered analytics" to "analytics for AI companies." Before: Competing with thousands After: Serving thousands The difference? Every AI company needs infrastructure. Only a few need another competitor. The opportunities hiding in plain sight: Model versioning systems. Compliance frameworks. Data labeling tools. Cost optimization platforms. Inference infrastructure. Edge computing solutions. Boring? Yes. Necessary? Absolutely. Fundable? Ask any VC focused on infrastructure. The smart money isn't just chasing AI. It's building what AI needs to exist. Here's what founders miss: The best businesses in a gold rush aren't the ones finding gold. They're the ones everyone pays on the way to the mountain. Your competition as an AI startup: OpenAI, Anthropic, and thousands of others. Your competition as infrastructure: Usually just spreadsheets and duct tape. The founders getting funded in Q4 won't all be building the future of AI. Many will be building the foundation it runs on. Stop asking "How do I compete with ChatGPT?" Start asking "What does every AI company buy?" That's where the real opportunity is. #VentureCapital #Infrastructure #StartupFunding #AIInvestment #FidelmanCo