Key Factors Driving AI Compute Growth

AI holds great potential for the semiconductor industry and will kick-start the next round of innovation for faster, cheaper and more energy-efficient computation – that was my message today at SPIE Advanced Lithography + Patterning. I discussed the potential and the challenges that AI holds for our industry.   The potential is clearly huge. AI is rapidly integrated into applications, and high-performance compute is expected to underpin growth towards $1 trillion of semiconductor sales by 2030. The challenges are around the computing needs of AI models and related energy consumption. The compute workload of training a leading AI model has increased 16x every 2 years in recent years – much faster than the increase in computing power delivered by Moore’s law, which is about 2x every 2 years. The energy needed to train a leading model has not grown so steeply but still rose 10x every 2 years. This computing need has been met by building supercomputers and massive data centers. If you extrapolate these trends, training a leading AI model would need the entire world-wide electricity supply in about 10 years. That’s clearly not realistic, so the trend has to break, by training algorithms becoming more efficient and by chips becoming more efficient. In other words, the needs of AI will stimulate immense innovation in chip design and manufacturing – and the potential value of AI to our society will put urgency and funding behind that drive. As a consequence, chip makers are pulling all levers to accelerate semiconductor scaling. This includes lithographic “2D” scaling: shrinking the dimensions of transistors to pack more into a square millimeter. It will also include “3D” integration, with innovations like backside power delivery, transistor designs like gate-all-around, as well as stacking chips in the package, where holistic lithography will play a critical role to deliver performance requirements. ASML will support these trends through a comprehensive, holistic lithography portfolio. Our 0.33 NA/0.55 NA EUV lithography systems allow chip makers to shrink dimensions at the lowest possible cost on their critical layers, while tightly matched and highly productive DUV systems will continue to reduce cost. More than ever, metrology and inspections tools – whose data is fed into lithography control solutions that keep the patterning process operating within tight specs to deliver the highest possible production yields – will be essential to deliver 2D scaling and 3D integration processes. 3D integration requires wafer-to-wafer bonding, and we have demonstrated the capability to map the stresses and distortions that bonding creates and to compensate for them, reducing overlay errors for post-bonding patterning by 10x or more.   It was a pleasure catching up with the industry’s lithography and patterning experts in San Jose. I’m excited to see our collective innovation power having a go at these challenges. Together, we will push technology forward.

The AI race will not be won by the biggest model. It will be won by the most adaptable infrastructure. Three shifts stand out right now. 1. Image models are becoming reasoning engines.   Multimodal models are now generating accurate charts, legible text, and consistent layouts from natural language. Images are becoming a valid front end for analytics, training, and operations, not just brand and marketing. 2. Coding agents are moving from autocomplete to coworkers.   New models are built to work for hours on a problem, refactor large code bases, and manage complex workflows. This is the start of continuous software delivery by AI, not just a side tool for developers. 3. The AI infrastructure cycle is accelerating and becoming heterogeneous.   Inference is always on and needs to sit closer to data, users, and regulators. That is driving a build out of specialized compute across CPUs, GPUs, TPUs, and other accelerators. Frontier models like Gemini are already trained and served on custom TPUs, while GPUs remain the workhorses for parallel math and CPUs still anchor control and business logic. The question is no longer which chip to choose, but how to compose the right mix and move workloads as cost, regulation, and model options evolve. In my role at Rackspace I see this weekly with leaders in healthcare, financial services, and the public sector. They are not asking whether to use AI. They are asking how to secure the right mix of compute and locations without recreating technical debt. For forward deployed leaders, the ones closest to customers and operations, the agenda for the next 12 to 24 months is clear: • Treat image models as a new experience layer. Take one important customer or employee journey and redesign it so dynamic visuals and copilots are the primary interface, not static reports or dashboards.   • Select one critical workflow and rebuild it with AI at the center. Break it into steps, decide where agents own the work and where humans stay in the loop, and redesign the data and process around that.   • Plan capacity and partnerships around persistent inference demand and a mix of CPU, GPU, and TPU, rather than a single vendor or architecture. The gap will not be who has access to AI. It will be which organizations can rewire their operating model and infrastructure fast enough, while staying flexible enough to pivot as the landscape continues to shift.

The AI GPU market isn’t a battle anymore. It’s a platform shift — backed by data. For a decade, NVIDIA built an empire on CUDA — and today it captures ~80–90% of the AI accelerator market. That’s dominance. But dominance creates pressure… AMD is scaling a different model: OPEN AI ECOSYSTEM Why this shift is accelerating: 🔓 1. Lock-in is becoming too expensive • Rewriting CUDA workloads can take 3–9 months for large models • Engineering cost per migration: $1M–$10M+ depending on scale Open standards (ROCm, HIP): → Reduce porting complexity by 30–50% → Enable multi-vendor deployment 2. AI spend is exploding • Global AI infrastructure spend projected to exceed $400B+ by 2027 • Hyperscalers already investing $50B–$100B annually each in AI At that scale: → A 10% cost delta = $40B impact Open ecosystem enables: → Real pricing competition → Lower total cost of ownership (TCO) 3. Supply is the real constraint • Advanced packaging (CoWoS) capacity grew ~2–3x from 2023 to 2025, yet still constrained • Lead times for high-end AI GPUs reached 6–12 months at peak demand Both NVIDIA and AMD rely on TSMC → No single vendor can satisfy global AI demand Open ecosystem = supply diversification at scale 4. Sovereignty is now measurable risk • Governments committing $100B+ globally to local semiconductor + AI capacity • AI classified as critical infrastructure in US, EU, and Asia Single-vendor dependency is no longer acceptable at national level. 5. Compute is fragmenting fast By 2028: • 50% of AI workloads expected to run on heterogeneous architectures • Custom silicon (ASICs, NPUs) growing >20–30% CAGR Closed ecosystems: → Optimized for one architecture Open ecosystems: → Designed for multi-architecture orchestration Reality check: NVIDIA still leads: ✔ ~80–90% market share ✔ Most mature AI stack ✔ Fastest deployment cycle But… History is consistent: → Linux powers 90%+ of cloud workloads → Android runs on 70%+ of global devices → Open ecosystems dominate when scale matters 🚀 What’s happening now: Hyperscalers are actively moving toward: → Dual- and multi-vendor AI strategies → Reducing single-stack dependency → Optimizing cost per training/inference cycle The real question is no longer: “Who has the best GPU?” It’s: Who enables AI to scale from billions → trillions efficiently? #AI #Semiconductors #AMD #NVIDIA #Cloud #DataCenters #OpenSource #Innovation #FutureOfTech #MachineLearning:

The real moats in AI hardware aren’t just chips Everyone is chasing GPUs. But the real power and the most durable margins sit outside the rack. This map tells a more complete story- compute scale is bounded not by transistor density, but by the speed at which the ecosystem around the chip can adapt. 1. Equipment is the upstream moat- $ASML, $LRCX (Lam), and $AMAT determine how fast the world can add capacity. Lithography, etch, and metrology throughput decide yield, and yield defines time-to-scale. Every wafer shipped is a function of physics, tooling availability, and learning rates that compound only where equipment bottlenecks are cleared. 2. Packaging and HBM are the midstream moat- CoWoS, SoIC, and advanced 2.5D/3D integration are the new fabs. Performance and delivery are increasingly determined by packaging throughput, substrate supply, and HBM allocation. The constraint has shifted from wafer starts to package exits. Those who own this layer decide how fast “launched” products become revenue. Great opportunity for $INTC Intel Corporation 3. Power and thermal are the downstream moat Watts are the new wafers. The limiting factor for AI infrastructure is no longer GPU availability but power conversion, switchgear lead times, and cooling capacity. Schneider Electric, Eaton, and ABB are as critical to AI scaling as Nvidia or TSMC. Without grid and thermal readiness, silicon is wasted capital. 4. Networking and integration are the operational moat- Broadcom, Marvell, and Arista convert chip potential into real throughput. Network latency, congestion control, and thermal-aware integration determine usable performance at scale. Server OEMs like Supermicro win by cycle-time to configurability, how fast they can turn bill of materials into deployed compute. The next frontier is orchestration. The AI hardware stack is now a system-level problem that spans chip, rack, and grid. Capacity planning requires real-time coordination of packaging slots, HBM supply, transformer MW, cooling tons, and workload placement. The new differentiation is the sync between physics and infrastructure. My thoughts - By 2027, incremental AI capex shifts outside the rack. Packaging and power infrastructure outgrow server spend. - Hyperscalers will sign “capacity offtake” deals with OSATs and equipment vendors, similar to power purchase agreements in energy. - Tokens-per-watt replaces raw TOPS as the new board metric. - Winners integrate across the physical-digital boundary; chip, rack, and facility into one adaptive system. Inside the rack wins headlines. Outside the rack defines economics. #Semiconductors #AIInfrastructure #AdvancedPackaging #HBM #DataCenters #PowerAndThermal #ASML #LamResearch #AppliedMaterials #Networking #AgenticAI #TokensPerWatt

✍️ 𝗣𝗼𝘄𝗲𝗿𝗶𝗻𝗴 𝗔𝗜: 𝗔 $𝟮 𝗧𝗿𝗶𝗹𝗹𝗶𝗼𝗻 𝗖𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲, 𝗮𝗻𝗱 𝘄𝗵𝗮𝘁 𝗶𝘁 𝗺𝗲𝗮𝗻𝘀 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗨𝗦, 𝗠𝗘𝗡𝗔, & 𝗖𝗜𝗦. 🌎 🙋♂️ Through my ventures in data center investments and business development across the MENA, US, and CIS regions, including my work with AI-driven healthcare initiatives, I've seen firsthand the escalating demands for AI infrastructure. 📣𝙒𝙝𝙞𝙡𝙚 𝙗𝙞𝙡𝙡𝙞𝙤𝙣-𝙙𝙤𝙡𝙡𝙖𝙧 𝘼𝙄 𝙙𝙚𝙖𝙡𝙨 𝙢𝙖𝙠𝙚 𝙙𝙖𝙞𝙡𝙮 𝙝𝙚𝙖𝙙𝙡𝙞𝙣𝙚𝙨, 𝙖 𝙘𝙧𝙞𝙩𝙞𝙘𝙖𝙡 𝙘𝙝𝙖𝙡𝙡𝙚𝙣𝙜𝙚 𝙡𝙤𝙤𝙢𝙨: ‼️We're facing an $800 billion revenue shortfall for data centers, necessitating an estimated $2 trillion in investment by 2030 to maintain the current pace. It isn't just growth; it's a gold rush for computing power, the new most valuable commodity. 🪄 Consider these points: ✔️ AI's compute demand is doubling at twice the rate of Moore's Law, a pace of progress we've never seen before. ✔️ AI server growth is projected at a 41% CAGR, driving the overall data center market to a 23% CAGR. ✔️ To meet this demand, we need to invest $500 billion annually in data centers over the next decade. ✔️ The cost to construct a data center building has surged by 322% in just four years, before even adding a single chip or server. 📍 In healthcare AI—a sector I've focused extensively on in the MENA region—the infrastructure demands are particularly acute. Medical imaging, AI, and genomics processing require sustained high-performance computing, making reliable, cost-effective data center access critical for healthcare innovation. ♾️ This explosive growth is creating a significant energy bottleneck. Power demand from AI centers is set to quadruple in the next decade. By 2035, they could consume 1,600 terawatt-hours of power, equivalent to 4.4% of global electricity demand. 🔎 The AI revolution is still in its early stages. Addressing this $2 trillion challenge requires collaboration among investors, technology innovators, energy providers, and policymakers worldwide, from the US to the CIS and from Europe to the MENA region. 🖍️ 𝗔𝗱𝗱𝗿𝗲𝘀𝘀𝗶𝗻𝗴 𝘁𝗵𝗶𝘀 𝗰𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲 𝗿𝗲𝗾𝘂𝗶𝗿𝗲𝘀 𝗮 𝗳𝗼𝘂𝗿-𝗽𝗿𝗼𝗻𝗴𝗲𝗱 𝗮𝗽𝗽𝗿𝗼𝗮𝗰𝗵: ↗️ Alternative energy partnerships (nuclear, renewable microgrids), ↗️ Next-generation cooling technologies (liquid cooling, immersion cooling), ↗️ AI-optimized chip architectures that improve performance per watt, and ↗️ Strategic geographic distribution to leverage regional energy advantages. ⁉️ The future of AI depends on the physical infrastructure we build today. Which emerging markets do you see as most promising for sustainable AI infrastructure development? 🧐 How are you balancing immediate scaling needs with long-term sustainability commitments? 📶 Let's connect and discuss the future of AI infrastructure. #DataCenters #AI #Investment #Energy #PhysicalAI #AICenters #MENA #CIS #Healthcare #Data #Infrastructure

The UAE has 23M H100 equivalents. China has 400K. Geography isn't destiny in the AI race, strategy is. As regions scramble to build AI ecosystems, most focus on talent and funding. But the real bottleneck? Compute power infrastructure. The countries winning aren't just the obvious suspects. The new power dynamics are surprising: UAE ranks #2 globally in AI compute capacity (behind only the US) Saudi Arabia sits at #3 with 7.2M H100 equivalents China, despite 230 AI clusters, ranks only #7 in actual compute power India shows massive potential with 493K AI chips deployed Here's the 6-factor playbook regions are using to build compute dominance: 1. Power Infrastructure AI data centers require 3x more power than traditional facilities Grid stability for 24/7 operations Renewable energy integration Smart load balancing systems 2. Hardware Access Secure reliable access to cutting-edge AI chips from NVIDIA/TSMC Direct vendor partnerships Supply chain diversification Local assembly capabilities 3. Government Policy National backing provides essential scale like India's ₹10K crore investment Public-private partnerships Tax incentives & subsidies Data sovereignty frameworks 4. Physical Infrastructure Specialized data centers with advanced cooling for GPU clusters Liquid cooling systems High-speed fiber networks Modular expansion design 5. Talent Development Skilled workforce to design, deploy, and maintain AI infrastructure Technical certification programs University partnerships International talent attraction 6. Financial Capital Massive funding as global AI infrastructure hit $200B in 2025 Sovereign wealth funds Foreign direct investment Compute-as-a-Service models Small, strategic regions can leapfrog traditional tech powers through focused investments and smart partnerships. The UAE's rise proves that with the right energy infrastructure, government backing, and strategic partnerships, any region can become an AI superpower. Hyperscalers spending $500B annually by early 2030s, governments prioritizing AI sovereignty But there are semiconductor supply constraints, energy demands, talent competition The winners won't just build models, they'll build the infrastructure that powers the entire AI economy. 🔥 Want more breakdowns like this? Follow along for insights on: → Geopolitics of AI & global cooperation → Building AI infrastructure at scale → Strategic technology sovereignty → AI's impact on regional competitiveness

𝗧𝗵𝗲 𝗦𝗶𝗹𝗶𝗰𝗼𝗻 𝗦𝗵𝗼𝗰𝗸: 𝗪𝗵𝘆 𝘁𝗵𝗲 𝗪𝗼𝗿𝗹𝗱 𝗕𝗲𝗰𝗮𝗺𝗲 𝗢𝗯𝘀𝗲𝘀𝘀𝗲𝗱 𝘄𝗶𝘁𝗵 𝗦𝗲𝗺𝗶𝗰𝗼𝗻𝗱𝘂𝗰𝘁𝗼𝗿𝘀 Five years ago, semiconductors were a quiet, cyclical business which was rarely discussed outside engineering circles. Today, the same industry powers the global economy. Companies like NVIDIA, Broadcom, TSMC, and ASML together are worth over $8.5 trillion. Governments across the US, EU, Japan, and India are pouring hundreds of billions of dollars into new semiconductor strategies. This shift was driven by a mix of AI breakthroughs, geopolitical rivalry, and supply-chain dependence -turning silicon into the world’s most strategic resource. 𝗧𝗵𝗲 𝗙𝗶𝘃𝗲-𝗬𝗲𝗮𝗿 𝗦𝗵𝗶𝗳𝘁 (𝟮𝟬𝟮𝟬 → 𝟮𝟬𝟮𝟱)  • 𝗧𝗵𝗲 𝗔𝗜 𝗖𝗼𝗺𝗽𝘂𝘁𝗲 𝗕𝗼𝗼𝗺: Generative AI created a surge in demand for parallel computing—powered mainly by GPUs. AI compute became the new “digital oil,” pushing chipmakers to the center of the tech economy.  • 𝗦𝘂𝗽𝗽𝗹𝘆 𝗖𝗵𝗮𝗶𝗻 & 𝗚𝗲𝗼𝗽𝗼𝗹𝗶𝘁𝗶𝗰𝘀: The pandemic exposed how concentrated manufacturing is. Nearly all sub-10 nm capacity sits in Taiwan and South Korea, while China dominates packaging and materials. Owning silicon capacity now means owning economic resilience.  • 𝗧𝗵𝗲 𝗦𝗼𝘃𝗲𝗿𝗲𝗶𝗴𝗻𝘁𝘆 𝗣𝘂𝘀𝗵: Chips are now instruments of national power. Export limits on advanced devices and EUV lithography tools spurred major incentive programs—the US CHIPS Act, EU Chips Act, and others—to rebuild domestic manufacturing.  • 𝗖𝘂𝘀𝘁𝗼𝗺 𝗦𝗶𝗹𝗶𝗰𝗼𝗻: Cloud giants—Google, Amazon, Microsoft, Meta - design their own chips to optimize AI workloads. That shift boosts demand for design IP and specialized foundries such as TSMC. 𝗪𝗵𝗮𝘁’𝘀 𝗡𝗲𝘅𝘁 (𝟮𝟬𝟮𝟲 → 𝟮𝟬𝟯𝟬) The industry is on track to exceed $1 trillion in annual revenue by 2030—driven by AI, electrification, advanced packaging and edge computing.  • 𝗘𝗱𝗴𝗲 𝗔𝗜 𝗘𝘃𝗲𝗿𝘆𝘄𝗵𝗲𝗿𝗲 — Efficient NPUs will move AI from the cloud to devices: wearables, phones, robots, IoTs, drones, etc.  • 𝗘𝗹𝗲𝗰𝘁𝗿𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻 & 𝗘𝗻𝗲𝗿𝗴𝘆 — EVs and smart grids are accelerating demand for SiC and GaN power chips.  • 𝗣𝗮𝗰𝗸𝗮𝗴𝗶𝗻𝗴 𝗕𝗼𝘁𝘁𝗹𝗲𝗻𝗲𝗰𝗸 — Performance now hinges on chiplets, 3D stacking, #CPOs, and #HBM. Advanced packaging capacity is the new battleground. 𝗕𝗲𝘆𝗼𝗻𝗱 𝟮𝟬𝟯𝟬: 𝗧𝗵𝗲 𝗤𝘂𝗮𝗻𝘁𝘂𝗺 & 𝗣𝗵𝗼𝘁𝗼𝗻𝗶𝗰 𝗛𝗼𝗿𝗶𝘇𝗼𝗻 The next wave is already forming. #Quantum, #photonic, and #neuromorphic chips will bring massive leaps in speed and energy efficiency, reshaping how AI, computing, and communication systems are built. The silicon era was only the start of something much bigger. Murali Chirala Bala Joshi. Tarun Verma. Ish Kumar Bhargava Hrishi Sathawane BV Jagadeesh Krishna Yarlagadda Mahesh Lingareddy

The Silicon Treadmill: Why the Speed of Upgrades is Redefining Data Center Economics in 2026. 💻 We’ve officially shifted from 18-month cycles to a relentless 12-month (or even 9-month) hardware cadence. While semiconductor leaders like NVIDIA report record 73.6% gross margins, the rest of the infrastructure ecosystem is navigating a "margin paradox." Here is what is happening beneath the surface: 🧵 1. The Depreciation Conflict ⚖️ Hyperscalers are standardizing on 6-year depreciation schedules to protect earnings, yet top-tier training silicon often has a competitive life of just 1 to 3 years before becoming obsolete for frontier models. To bridge this gap, operators are using a "Value Cascade," repurposing older GPUs for the massive, cost-sensitive inference market to stretch asset utility. 2. The $25M Per Megawatt Fit-Out 🏗️ AI infrastructure is no longer standard real estate. Technical fit-outs for AI clusters now cost up to $25 million per MW—nearly triple the cost of traditional cloud facilities. With global construction costs rising significantly, the "AI Factory" requires unprecedented capital intensity. 3. Hitting the Thermal Wall 🧊 Air cooling is hitting its physical limit. As individual chip power (TDP) exceeds 1,000W, liquid cooling has moved from a specialized solution to the mainstream requirement, projected to account for 47% of deployments by year-end. 4. Power as the Sovereign Bottleneck ⚡ Location and land are no longer the primary site selection criteria—power availability is. With grid connection wait times exceeding four years in major markets, operators are forced to invest in "behind-the-meter" generation and long-term "Take-or-Pay" power contracts just to secure capacity. 5. The SaaS Margin Inversion 📉 Traditional SaaS enjoyed 80% margins, but AI-native application margins are hovering between 0-30%. Because every query burns tangible compute, CFOs are reclassifying AI infrastructure as a variable COGS rather than fixed OpEx, fundamentally changing how software is priced and sold. The winners in 2026 won't just have the best models—they will be the ones who can deliver the most "tokens per watt" at the lowest cost. Any strategies to boost margins? 👇 #AI #Semiconductors #DataCenters #CloudComputing #FinOps #Silicon2026 #AIInfrastructure

Every AI breakthrough we see today depends on a single, invisible & scarce resource: 𝗖𝗢𝗠𝗣𝗨𝗧𝗘 In the past, tech giants competed on apps & features. Today, the race is for 𝗶𝗻𝗳𝗿𝗮𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲. Real power now lies with whoever controls the flow of GPUs, wafers & energy to train the next generation of AI models. 𝗢𝗽𝗲𝗻𝗔𝗜 has been orchestrating this race through a trillion-dollar web of chips, compute & capital. But instead of buying billions in hardware outright, it convinced the biggest infra players to fund it: → 𝗡𝗩𝗜𝗗𝗜𝗔 provides billions in GPUs & compute credits → 𝗢𝗿𝗮𝗰𝗹𝗲 & 𝗠𝗶𝗰𝗿𝗼𝘀𝗼𝗳𝘁 commit long-term cloud capacity → 𝗦𝗼𝗳𝘁𝗕𝗮𝗻𝗸 sets as the Stargate joint venture financier → 𝗧𝗦𝗠𝗖 produces the advanced chips that power it all But even with over $1T in pledged capacity, the system still hits physical limits: → TSMC’s CoWoS (Chip-on-Wafer-on-Substrate) packaging is the main bottleneck in GPU supply chain → Gigawatt-scale data centers take up to 2 years to be built → Power grid upgrades lag behind compute demand When Sam Altman says compute is “terrible”, he implies AI’s growth is throttled by the speed of construction and silicon. And while the bottlenecks slow things down, they're also reshaping the tech landscape: → 𝗢𝗿𝗮𝗰𝗹𝗲 reinvented as an AI infra powerhouse → 𝗠𝗶𝗰𝗿𝗼𝘀𝗼𝗳𝘁, from provider to strategic partner → 𝗡𝗩𝗜𝗗𝗜𝗔, from chip maker to infrastructure financier → 𝗦𝗼𝗳𝘁𝗕𝗮𝗻𝗸 repositioned as AI’s backbone capital engine OpenAI’s flow of compute reveals a silent shift in tech value creation: from writing algorithms to owning infrastructure & controlling the supply chain of compute. Imho, data is no longer the new oil. Compute is.

Sunday AI Pulse 1. OpenAI signed a multiyear cloud deal with Amazon worth about $38 billion, locking in massive compute capacity and signalling continued hyperscaler competition over AI infrastructure. This is not just a vendor choice. It reshapes where large models run and who controls the physical stack. 2. Meta announced a sweeping plan to invest roughly $600 billion in U.S. infrastructure and jobs over the next three years, with a major focus on AI data centers. This underlines how big tech is shifting from model R&D to a race for physical capacity and nationwide deployment. 3. Microsoft expanded commercial programs and deals this week, from a $9.7 billion cloud contract tied to AI needs to a new Agentic Launchpad in the UK with NVIDIA to accelerate agentic AI startups. The pattern is clear. Cloud providers are bundling compute, go-to-market, and engineering support to turn models into businesses. 4. Big money is still flowing. PitchBook and coverage this week show venture capital and corporate budgets concentrating on AI infrastructure and enterprise AI, though the returns calculus is getting tighter as scrutiny on governance and deployment grows. Expect capital to chase both scale and defensible enterprise moats. 5. Small player to watch. Daylight, a Tel Aviv cybersecurity startup launched in 2025, secured a large preemptive funding package this week. Their focus on AI driven managed detection and response highlights how startups are emerging to solve AI-native security needs as models and infra scale. Early bets here are worth watching for enterprise risk management. My takeaways for leaders and builders 1. This week’s headlines are a reminder that AI is shifting from algorithmic novelty to industrial strategy. The competition is now about data center footprint, network partnerships, and supply chain for compute, not just model accuracy. 2. If you run product or engineering, your near term decisions should prioritize integration points. Where does your model run? Who owns the data path? How will you operate when a provider changes terms or capacity is constrained? 3. For founders and VPs thinking about hiring or fundraising, the playbook matters. If you are building vertical workflows and strong operating models around data and automation, you are building something that survives a shift in who controls the lowest layers. 4. Security and governance are urgent. As infra scales, so does attack surface and operational complexity. Expect new categories of startups and internal teams focused on AI reliability and detection. What stood out for you this week? Are you seeing the same infrastructure centric shift in your org, or is your focus still primarily on models and features? #AI #Infrastructure #EnterpriseAI #AIAgents #DataAndAI #Startups