How Big Tech Influences AI Infrastructure

We’re used to thinking of Big Tech as a cage match: Apple vs Google, Meta vs OpenAI, Amazon vs Microsoft. But go deeper down the stack and these rivalries dissolve into something less cinematic - inter-dependent supply chains. Consider just this past week: ➰ Meta signed a $10B+ cloud deal with Google, its fiercest rival in digital advertising. ➰ OpenAI is feeding ChatGPT with Google Search results (via SerpAPI) and renting its GPUs, while trying to make Google Search obsolete. There’s an entangled web of interdependencies in AI: your most threatening competitor is often your most critical vendor. Everyone sells the shovel, even to the guy digging their grave. So, what gives? 1. Moats are now Rentable. And often leased to the very people trying to cross them. What used to be a moat - distribution (iOS/Android), data (Search), or compute (GPUs at hyperscale) - is increasingly sold as a SKU. If your “defensive asset” can be metered, it will be monetized… even to your rivals. That sounds contradictory until you realize the real moat isn’t the resource - it’s the flywheel that replenishes it. Google can lease GPUs and still deepen its Gemini feedback loops. 2. Infrastructure is too Expensive to own Alone. The modern AI stack is fractured and expensive: - Compute (GPUs, interconnects, custom silicon) - Indexing (web crawlers, real-time feeds, proprietary corpora) - Modeling (foundation models, adapters, RAG) - Orchestration (retrievers, agents, tool use) - Distribution (hardware, OS defaults, app interfaces) No single company can win all five. So they do what every industry does when vertical integration becomes unscalable: they trade. AI isn’t owned. It’s assembled - by companies renting from rivals they’d love to replace. 3. Market Power comes from Volume. Take Meta. It’s signed deals with every major cloud provider: AWS, Azure, Oracle, CoreWeave, and now Google Cloud. This isn’t loyalty; it’s pricing arbitrage and regional hedging. At that scale, cloud is a commodity and power comes from being the customer that can move someone else’s earnings call. 4. Time-to-Quality > Ideological Purity. If the fastest path to product quality is to buy accuracy while you build your own index, you do both. You can always replace a vendor. You can’t buy back time. Months matter. In AI, months are market share. Meanwhile, Google selling compute to OpenAI is not charity; it’s toll collection on a rival’s growth curve. 5. Optics matter Turning your enemies into customers is not just good business - it’s good politics. Each hyperscaler that lands a rival as a marquee customer bolsters its narrative: To Wall Street: “We grow no matter who wins.” To regulators: “We’re not a monopoly, we power our competitors.” The stack is too entangled, too capital-intensive, and too unevenly distributed for anyone to play lone wolf. In this economy, independence is expensive and rivalry is mostly theater. Monetize your enemy’s ambition. The best revenge is recurring revenue.

The major tech companies - Amazon Web Services (AWS), Google, Meta Facebook and Microsoft - invested over $65 billion in CAPEX this quarter (Q3) on cloud and AI infrastructure. Year-to-date spending exceeds $171 billion, setting records for quarterly investment: Amazon: $22.79 billion (+79%), marking a new high. Spending primarily targets AWS and fulfillment. Amazon expects around $75 billion in CAPEX for 2024, with further increases projected for 2025. Google: $13.06 billion (+62%), matching nearly all of 2017’s annual spend in one quarter. Investments focus 60% on servers and 40% on data centers. Meta: $9.2 billion (+36%), slightly below guidance due to timing, with increased spending expected in Q4 and 2025 for infrastructure growth. Microsoft: $20 billion (+79%), equivalent to its full-year 2020 spend, aimed at AI-driven cloud capacity. Microsoft’s enterprise offering, Fabric, now has over 16,000 customers, including 70% of the Fortune 500. Detailed Company Quotes: Amazon:  - “We expect to spend approximately $75 billion in CAPEX in 2024. The majority supports AWS’s growing AI demand, alongside infrastructure in North America and internationally. Investments in fulfillment and transportation networks aim to enhance delivery speeds and reduce service costs.”  - “Many of these assets, such as data centers, have useful lives of 20 to 30 years.”  - "Our AI capacity demand currently exceeds available infrastructure."  - "CAPEX growth is particularly driven by generative AI, with anticipated further spending in 2025." Google:  - "We expect Q4 CAPEX to match Q3 levels and project further increases in 2025, though not as substantial as from 2023 to 2024."  - "In Q3, approximately 60% of CAPEX went to servers, with 40% allocated to data centers and networking equipment." Meta:  - “Our full-year 2024 CAPEX range is now $38-40 billion, slightly up from prior guidance, with significant infrastructure growth anticipated in 2025.”  - "The expected increase in Q4 CAPEX will be partly due to server spend and data center investments, with delayed cash outflows from server deliveries appearing in Q4."  - “We’re training Llama 4 on a cluster of over 100,000 H100 GPUs—one of the largest known setups.” Microsoft:  - “Half of our cloud and AI spending is on long-lived assets supporting monetization over the next 15 years, with the remainder for CPUs and GPUs to meet current demand.”  - "Demand, especially for AI inference, continues to exceed capacity."  - "We don’t sell raw GPUs externally due to our own high demand and adverse selection in the current market."  - "Our Fabric platform now has over 16,000 customers, including 70% of the Fortune 500, with Copilot Stack sitting atop Fabric to provide advanced enterprise infrastructure." #ai #digitalinfrastruture

AI Compute Oligarchy: Power, Not Code, Is the New Moat This image is not about GPUs. It is about control. At the top, the chart shows barely 2,259 MW of operational AI data-center power today. At the bottom sits a staggering 35,736 MW of planned capacity. That gap is the story of this decade. AI is no longer limited by ideas or algorithms. It is limited by electricity, land, chips, and who gets access to all three. A small group already dominates what is live. A few logos account for most of the power that is actually running. Training and inference at scale now demand industrial-grade energy footprints. AI has crossed from software into heavy infrastructure. Once that happens, concentration is inevitable. The lower half of the chart is even more revealing. Planned capacity runs into tens of thousands of megawatts, led by hyperscalers, frontier model labs, and energy-aligned data-center players. This is not speculative spending. This is pre-emptive land grab. Whoever builds first locks in power contracts, grid priority, silicon supply, and regulatory leverage for years. This is why the phrase “AI democratization” feels increasingly hollow. You can open-source models. You can publish papers. But you cannot open-source power stations. You cannot crowdsource grid access. You cannot casually finance a multi-gigawatt build-out. Compute has become the choke point. And choke points create oligarchies. The implications are uncomfortable. Innovation risk shifts from talent to access. Startups may invent breakthroughs, only to rent their future from those who own the machines. Nations without energy surplus risk becoming AI consumers, not producers. Even governance debates tilt toward those who can afford to run the largest experiments. There is also a quieter signal here. Nearly 94% of the required infrastructure is not yet built. That means the real race is just beginning. Policy, power pricing, sustainability trade-offs, and grid resilience will shape AI outcomes as much as model design. The next AI wars will be fought in planning commissions, energy ministries, and supply chains, not just labs. This chart is not a forecast. It is a warning. AI’s future will not be decided only by who writes the smartest code, but by who controls the physical backbone beneath it. Compute is destiny now. And destiny, as history shows, rarely distributes itself evenly. DC*

Just two months ago, AI infrastructure stocks were tumbling. Investor confidence was shaken, and whispers rippled through the market that Big Tech might be pulling back.  Even Microsoft, a core pillar of the AI boom, was rumored to be slowing its data center expansion. The narrative was shifting—from boundless optimism to skeptical restraint. But here’s the twist:  AI doesn’t run on hype. It runs on concrete, copper, and gigawatts. In just a few short weeks, we’ve seen a cascade of moves reshaping the landscape. Amazon? A week ago, Amazon revealed a $10 billion investment to expand its AI infrastructure in North Carolina, one of the largest in state history. This move will create over 500 high-skilled jobs and support thousands more in the AWS data center ecosystem. It’s not just about servers and silicon, Amazon is also launching training programs, funding K-12 STEM education, and backing local community projects. North Carolina is quickly becoming a hub for AI-driven innovation, and this investment signals just how fast the future is arriving. And then this week Amazon announced a $20 billion investment to build two AI and cloud computing data center complexes in Pennsylvania, marking the largest private sector investment in the state's history. The Salem Township facility is planned adjacent to the Susquehanna nuclear power plant, aiming for a direct power supply. This "behind-the-meter" arrangement is currently under review by the Federal Energy Regulatory Commission due to concerns about grid fairness and energy distribution. Meta? Meta signed a 20-year PPA with Constellation Energy to secure the full output from the Clinton Clean Energy Center, extending its life through June 2027 and adding 30 MW capacity, powering AI operations while sustaining 1,100 jobs and outputting as much energy as 800,000 homes. This week the news broke that Meta is investing $14.8 billion for a 49% stake in Scale AI, marking one of its largest acquisitions since WhatsApp, and positioning CEO Alexandr Wang to lead a new Meta team focused on developing super intelligence. The UK government just pledged £1 billion to expand AI compute infrastructure, 20× boost in national capacity, announced during London Tech Week. GlobalFoundries just committed an additional $3 billion to expand AI chip manufacturing in Saratoga County, NY, and Essex Junction, VT, on top of a previous $13 billion CHIPS Act-backed build-out. Applied Digital signed two long-term leases with CoreWeave to deliver 250 MW of capacity at its Ellendale, North Dakota data center, expected to generate $7 billion over 15 years. Purpose-built for AI and HPC, the site can scale to 1 GW, with an option for CoreWeave to lease an additional 150 MW, reinforcing Ellendale’s role as a scalable AI infrastructure hub. Now some of those same stocks? Vertiv?+95%. Constellation Energy? +75%. The AI gold rush isn’t just about the algorithms. It’s also about who supplies the picks and shovels.

Your AI startup has a 90% chance of being locked out. Not by competition. By infrastructure. The OECD just mapped the entire AI infrastructure stack. If you're building on AI, you need to see these concentration numbers. The pattern repeats at every layer: - Chip design: 3 companies control 90%+  - Advanced fabs: TSMC has 92% of cutting-edge production - Cloud compute: 3 hyperscalers dominate 65% - Data centers: Top 10 own most capacity This isn't just market share.  It's control over who gets to build. I've been shipping AI systems for over a year. Every bottleneck traces back to this same reality. The report treats AI like electricity or steam engines. General purpose technology that changes everything. Except this time, the infrastructure is already captured. What this means for builders... Access discrimination is coming. Not if. When. The big players are vertically integrating fast. Microsoft x OpenAI. Google DeepMind. Amazon x Anthropic. Cross-shareholdings everywhere. Partnership webs that lock out competition. My production reality: - GPU access: 6-month waitlists or 10x markup - Model access: Sudden API limits when you scale - Compute costs: Unpredictable spikes during launches - Migration lock-in: Switching costs designed to trap you The OECD - OCDE flags three critical risks: 1️⃣ Foreclosure They control the chips. They control access. Your innovative startup competes with their product? Good luck getting compute. 2️⃣ Discrimination Not outright denial. Subtle degradation. Higher latency. Lower priority. "Technical issues." Death by a thousand API timeouts. 3️⃣ Collusion potential When 3 players control everything, coordination is easy. Prices rise together. Innovation slows together. The market can't self-correct. Competition authorities are finally waking up. Probing chip designers. Investigating partnerships. But enforcement takes years. Markets move in months. The report's solution: Public compute infrastructure. Government-funded alternatives to break the stranglehold. Open-source requirements. Interoperability mandates. Until then, every AI builder faces the same reality: You're not just competing on product. You're competing for permission to exist. The infrastructure layer determines who wins. Not because they build better. Because they control who gets to build at all. The OECD's message is clear... This concentration isn't sustainable. Intervention is coming. But if you're building today? Plan for a world where compute access is power. And power is already concentrated. Follow Alex for the infrastructure reality of shipping AI. Save this if you're navigating the AI stack bottlenecks.

Alphabet Inc. just pledged $75B to scale its AI infrastructure in 2025. And it could quietly shift the entire balance of power... This isn’t a flashy model release or a moonshot bet. It’s a systematic land grab for the physical layers of AI, where compute, energy, and geography become Alphabet’s long game. The $75B number? The largest annual infrastructure spend in Big Tech history. Their goal? Build the pipes, platforms, and power flows that AI will depend on for decades. 1. Google Cloud expands globally. (AI-first regions optimized for Gemini and enterprise GenAI) 2. TPUs go custom and dense. (Purpose-built silicon paired with liquid cooling) 3. Energy moves in-house. (On-site generation, PPAs, and grid-level coordination) This isn’t about smarter queries. It’s about controlling the inputs that shape outcomes. And it’s already underway in #Macon, #BeaverDam, #FortWorth, places with space, speed, and silence. The shift is on. Alphabet is locking in power, land, and interconnect before others realize they’re falling behind. This isn’t hub expansion. It’s hyperscale fortification. And in the background? Tariff hikes, power constraints, global chip chokepoints. Still, Alphabet moves. Because the real cost isn’t overbuilding. It’s waiting too long to start. The company isn’t just deploying capital. It’s encoding itself into the AI economy’s fabric. And in emerging markets? From #Chennai to #Querétaro to #Lagos, the next wave of demand is materializing fast, and Alphabet wants to meet it with concrete, not latency. Because the game isn’t about building better models. It’s about owning the world they run on. Alphabet isn’t reacting to AI’s future. It’s preparing to anchor it. Before everyone else realizes there’s no space left to build. #datacenters

AMD stock just jumped 25% on the OpenAI deal. But the real story is what OpenAI is building, and how fast. OpenAI announced a massive multi-year partnership with AMD today: 6 gigawatts of Instinct GPUs starting in 2026, with a warrant structure that could hand OpenAI roughly 10% ownership of AMD. For AMD, this is validation they've been chasing for years. After trailing Nvidia in AI accelerators, they now have OpenAI, the flagship AI customer everyone wants. Lisa Su called it transformative, and she's right. But zoom out for a second. This AMD deal comes just two weeks after OpenAI's $100 billion Nvidia agreement. Together, that's 16 gigawatts of their 23-gigawatt infrastructure roadmap. At roughly $50 billion per gigawatt in construction costs, OpenAI has committed close to $1 trillion in buildout spending in 14 days. Let that sink in. Here's what's really happening: we're watching a tightly wound circular economy form in real-time. NVIDIA supplies capital to buy its chips. Oracle builds the data centers. AMD and Broadcom provide hardware. OpenAI anchors the demand. Everyone is investing in everyone, trading equity for compute, capital for capacity. The interconnected nature of this buildout creates risk, sure. But it also creates alignment. Everyone has skin in the game. That's not fragility—that's commitment. This is the kind of infrastructure bet that defines decades. I think they're getting it right. What's your take? #AI #TechIndustry #AMD #OpenAI #Infrastructure

"Open always wins." That's what Groq CEO Jonathan Ross told me when discussing the enterprise shift to open source LLMs. After extensive interviews with enterprise leaders, I found he's largely right – but with important nuance. The sophisticated players are moving quickly to open source, seeking greater control and customization. Just as Linux won the OS wars and Chromium dominated browsers, open source LLMs are gaining serious momentum in the enterprise: • Meta reports 400M+ downloads for Llama, 10x higher than last year (this isn't only enterprise obviously, but a reflection on Llama's momentum overall) • Major app platforms (Salesforce, Oracle, SAP, ServiceNow etc) are rapidly integrating open LLMs, so that their 10s or 100s of thousands of companies can use open models like Llama, Mistral and Cohere easily in workflows that avoid having to do all the "set up" work themselves. • Even AWS, after its $4B investment in Anthropic, acknowledges the surge But here's what's fascinating: The real story isn't about "open vs. closed" – it's about control. The most advanced enterprises, like Intuit, are building infrastructure to leverage both open and closed LLMs strategically. Some are even bringing compute in-house and/or creating sophisticated orchestration layers to maintain full control over their AI stack. This mirrors a larger shift: As AI becomes mission-critical, enterprises need to assert ownership of their intelligence infrastructure – not just their models. My latest deep dive explores this transformation and what it means for enterprise AI strategy: https://lnkd.in/gySeWXTN Curious to hear your thoughts on this shift. Are you seeing similar patterns in your organization? VentureBeat's team and I will diving deeper into this enterprise infrastructure story in the coming months through our AI Impact Tour events and at Vb Transform 2025. The conversations with technical leaders about how they're actually engineering these strategies, and accommodating the emerging and powerful area of agentic AI, will be intriguing indeed. (Thank you to Intuit's Ashok Srivastava, Meta's Ragavan Srinivasan and Manohar Paluri, AWS' Baskar Sridharan, SAP's Walter Sun, Ph. D., Groq's Jonathan Ross, Salesforce's Jayesh Govindarajan, Inflection AI's Ted Shelton, IBM's Matthew Candy, and Oracle's Greg Pavlik, among others, for their helpful insights.)

Microsoft, Google, and Meta are making unprecedented bets on AI infrastructure. Microsoft alone plans to spend $80B+ in 2025. By 2027 their collective AI infrastructure investment could exceed $1T. The assumption driving these investments: bigger models equal better AI. But here’s the data: → OpenAI's Orion model plateaus after matching GPT-4 at 25% training → Google's Gemini falls short of internal targets → Training GPT-3 uses about 1,300 megawatt hrs of electricity, equivalent to the annual needs of a small town → Next gen models would require significant energy resources The physics of computation itself becomes a limiting factor. No amount of investment overcomes these fundamental barriers in data, compute, and architecture. Researchers are pursuing new architectures to address the limitations of transformers: → State Space Models excel at handling long-term dependencies and continuous data → RWKV achieves linear scaling with input length versus transformers' quadratic costs → World models, championed by LeCun and Li, target causality and physical interaction rather than pattern-matching DeepSeek’s efficiency breakthrough reinforces this trend: AI’s future won’t be won by brute force alone. Smarter architectures, optimized systems, and new approaches to reasoning will define machine intelligence. These constraints create opportunities. While tech giants pour resources into scaling existing architectures I’m watching for founders building something different.

Tech companies will spend $400 billion on AI infrastructure in 2025—exceeding the Apollo program's inflation-adjusted budget, repeated every ten months. Both bubble skeptics and AI bulls present compelling evidence, leaving business leaders caught between FOMO and prudent risk management. Unlike dot-com startups burning venture capital, today's AI leaders (Microsoft, Google, Amazon) are massively profitable and will survive even if AI bets fail. The technology demonstrably works for specific tasks. Infrastructure has alternative uses if foundation model companies collapse. Unit economics worsen with scale rather than improve. Financial engineering obscures true profitability (Microsoft-OpenAI circular revenue bookings mirror WorldCom-era accounting). MIT studies show 95% of AI pilots fail to yield meaningful results. The gap between infrastructure spending ($400 billion) and consumer revenue ($12 billion annually) echoes telecom overcapacity that left 85-95% of fiber "dark" in 2002. Don't bet on timing the bubble—build for multiple scenarios. Prioritize AI applications with 12-month ROI that work whether vendors consolidate or not. Rent compute from hyperscalers rather than building proprietary infrastructure. Develop internal expertise that survives vendor failures. Prepare to acquire distressed assets (GPUs, talent, data centers) at 2027 fire-sale prices if correction arrives. Remember Amara's Law: We overestimate short-term impact, underestimate long-term transformation. The internet crashed in 2000 but enabled Amazon, Google, and Facebook by 2005. Position to benefit from both timelines. The bubble thesis is probably correct for 2026-2027. That doesn't make AI investments wrong—it makes vendor selection, contract structure, and capability building more critical than ever. Companies that survive bubbles distinguish hype from utility, build competency during uncertainty, and stay capitalized to buy when others must sell. Investors who know what’s coming can avoid misfortune.