How Openai is Diversifying Cloud Strategy

$4 billion. 250,000 GPUs. A cloud no one expected. CoreWeave just became OpenAI’s secret weapon... And maybe the fourth hyperscaler. This isn’t just about extra #GPUs or cloud spillover. It’s a signal. A shift in how the #AIinfrastructure game is being played and who gets to play. Here’s why this matters: 1. Infrastructure diversification is now strategy, not contingency OpenAI relies heavily on Microsoft. But scale, risk, and performance demands are growing faster than #Azure can deliver. CoreWeave gives OpenAI: – Redundancy across geos – Faster deployment timelines – GPU-optimized infrastructure It’s not a backup plan. It’s a second engine. 2. CoreWeave is the first hyperscaler built for AI, not retrofitted for it Crypto mining roots. GPU-native architecture. Air + liquid-cooled density. CoreWeave has: – 250,000+ NVIDIA GPUs live – Record-setting MLPerf scores on GB200 – $23B in #CapEx planned for 2025 It’s not just fast. It’s focused. 3. Alt-clouds are becoming essential infrastructure The next generation of AI will not run on general-purpose clouds alone. CoreWeave, Crusoe, Lamba Labs, they’re not fringe players anymore. They’re essential to anyone scaling models at the frontier. This $4B deal is more than revenue. It’s validation that purpose-built infrastructure will define the next phase of AI. And CoreWeave just locked in its place on the front lines. #datacenters

Yesterday, Reuters reported that OpenAI finalized a cloud deal with Google in May. This might look like routine tech news. It is not. This is a strategic inflection point in the AI infrastructure wars. OpenAI, whose ChatGPT threatens the core of Google Search, is now paying Google billions of dollars to power its growth. This was not a partnership of choice. It was a partnership of necessity. Since ChatGPT launched in late 2022, OpenAI has struggled to meet soaring demand for computing power. Training and inference workloads have outpaced what Microsoft’s Azure alone can support. OpenAI had to expand. Google Cloud was the solution. For OpenAI, the deal reduces its dependency on Microsoft. For Google, it is a calculated win. Google Cloud generated $43 billion in revenue last year, about 12 percent of Alphabet’s total. By serving a direct competitor, Google is positioning its cloud business as a neutral, high-performance platform for AI at scale. The market responded. Alphabet shares rose 2.1 percent on the news. Microsoft fell 0.6 percent. There are only a handful of true hyperscalers in the U.S. AWS, Azure, and GCP dominate, with Oracle and IBM trailing behind. The appetite for compute is growing faster than any one company can satisfy. In this new phase of the AI era, exclusivity is a luxury no one can afford. Collaboration across competitive lines is inevitable. -s

The race to build ever-smarter models is colliding with the physical limits of compute, power, and capital. What was once a contest of parameters and benchmarks is now a battle over who can assemble the biggest chip fleets, fastest buildouts, and most favorable power contracts. We’re witnessing the birth of the physical internet of intelligence. In the U.S., three companies lead the AI-industrial buildout: ▪️OpenAI: The Cloud Conductor. OpenAI is the most visibly advanced in the compute race, with 1M+ GPUs projected online by end-2025. Altman is less model builder, more a macro-scale orchestrator - navigating sovereign land deals, hyperscaler partnerships, and infrastructure contracts measured in gigawatts.  OpenAI signed a $30B/year deal with Oracle, locking in 4.5 GW of data center capacity, roughly two Hoover Dams. But Stargate, its much-hyped $500B venture with SoftBank and the Trump admin, has fizzled. No sites closed. One Ohio facility, still in planning, is all that remains. “Stargate” now loosely describes OpenAI’s infra push, but the real action is via Oracle and CoreWeave - not SoftBank. OpenAI doesn’t want to own the land. It wants to control the supply chain. ▪️xAI: Vertical Integration at the Edge of Debt. xAI is doing it Musk-style - building every layer in-house, from chips to power, and moving fast. Colossus 1 went up in 122 days with 100K GPUs. Colossus 2 is underway, targeting 1M. Musk claims 50M H100s in 5 years. But unlike OpenAI, xAI has no cloud partner. Every chip, pipe, and rack is self-financed. To fund this sovereign stack, Musk has layered on debt: $10B in equity/debt, $5B more in bonds backed by Grok and Colossus, $2B borrowed from SpaceX, and a new $12B chip leasing scheme with Valor Equity that turns capex into off-balance-sheet IOUs. xAI is burning $13B in 2025 with minimal revenue and a brand dinged by Grok’s behavior. But Musk’s build speed has believers. Jensen Huang calls him “superhuman.” ▪️Meta: Infrastructure Maximalism Without the Headlines. While others pitch investors, Meta writes checks. Zuck has committed $100B+ in capex by 2026. Its Hyperion campus alone will provide 5 GW, rivaling OpenAI’s Oracle footprint. Meta’s one of Nvidia’s biggest customers, with global buildouts underway. While xAI and OpenAI build for splash, Meta builds for integration. Its models power search, feed, WhatsApp, and Ray-Bans. Once open-weight and community-friendly, Meta is pivoting to productized, embedded AI. So: OpenAI is the global AI utility - cutting megadeals, hedging with partners, and betting that whoever routes the tokens, rules the future. xAI is the sovereign AI state - owning everything, borrowing against the future, and hoping that speed and vertical integration can outweigh its financial fragility. Meta is the invisible AI empire - embedding intelligence at the edge, controlling distribution, and building the most complete stack in the world. They’re laying railroads for our AI future.

OpenAI’s strategy that commands (or requires) $100B in funding… No startup in history has raised $100B or more in funding. OpenAI is the most capital-intensive companies ever built. The obvious answer to “where is all that money going?” is talent and compute. But, dig deeper and OpenAI’s latest deal activity tells a bigger story about what all this funding is… uh… funding. Across acquisitions, acqui-hires, investments, and partnerships, the company is building something far more ambitious than a better model. It is assembling the infrastructure, distribution, and transaction layers required to turn AI into a global economic platform. And it is doing so in the middle of a competitive shift. OpenAI owns consumer mindshare at unprecedented scale, while Anthropic has been steadily converting enterprise and coding wins, especially in regulated and high-stakes environments. OpenAI’s response is not just to improve model performance or tout safety, but to expand outward and shape the environment in which AI is deployed, purchased, and monetized. OpenAI is betting that the next wave of AI adoption and monetization will be won by whoever can industrialize intelligence and embed it into how work gets done and how money moves within an AI ecosystem. Its activity clusters around six interrelated bets: → Vertical scale in high-budget sectors like finance, healthcare, and government, where adoption cements long-term revenue and influence → Enterprise embedment, integrating directly into core software systems so OpenAI captures infrastructure spend rather than sitting on top as a feature → Developer gravity, building the tooling, analytics, evaluation, and monitoring layers that make OpenAI the default environment where AI products are created and refined → Industrial control of infrastructure, from data centers to chips to deployment capacity, reflecting a belief that the true bottleneck is physical and operational scale → AI-native commerce, where the default, conversational interfaces become transaction engines and capture value at the moment of intent → AI devices and new interfaces, where distribution shifts from screens and apps to ambient, voice-first, and always-available assistants OpenAI’s bet is that intelligence will become embedded in every workflow and every transaction, and that the company controlling the rails of deployment, distribution, and monetization will control the economics of the AI era. And, to win that may require $100B… or more. P.S. Want to compare this to Anthropic’s strategy? CB Insights Strategy Maps are available for any company.

OpenAI has signed a multi-year agreement with AMD to deploy 6 GW of Instinct GPUs, with the first 1 GW coming online in 2026 and an option for OpenAI to take up to a 10% equity stake in AMD. That scale is wild! Six gigawatts of compute power could: • Drive some of the largest hyperscale campuses ever built, or • Support tens of millions of AI inference workloads simultaneously. The message here runs deeper than hardware: 𝐃𝐢𝐯𝐞𝐫𝐬𝐢𝐟𝐢𝐜𝐚𝐭𝐢𝐨𝐧 𝐨𝐯𝐞𝐫 𝐝𝐞𝐩𝐞𝐧𝐝𝐞𝐧𝐜𝐲. OpenAI is reducing its reliance on NVIDIA and securing a second supply chain for chips. 𝐂𝐨𝐦𝐩𝐮𝐭𝐞 𝐢𝐬 𝐭𝐡𝐞 𝐧𝐞𝐰 𝐜𝐨𝐦𝐦𝐨𝐝𝐢𝐭𝐲. The AI race is now defined by who controls generation not of energy, but of compute. 𝐏𝐚𝐫𝐭𝐧𝐞𝐫𝐬𝐡𝐢𝐩𝐬 𝐚𝐫𝐞 𝐞𝐯𝐨𝐥𝐯𝐢𝐧𝐠. Equity alignment makes this more than a procurement deal, it’s a long-term infrastructure play. As AI demand surges, chip strategy is becoming national strategy. Hardware, energy, and talent are converging into one ecosystem and the next competitive edge won’t come from who builds faster, but who integrates deeper. Will this mark the start of a true two-horse race in AI infrastructure?

OpenAI ’s Cerebras deal is a signal that the AI infrastructure market is starting to price compute the way utilities price electricity: by megawatts and service levels. Yesterday, OpenAI said it’s partnering with Cerebras to add 750MW of low-latency compute to its platform, rolling out in phases through 2028. Here is what I saw in the announcement: 1) The unit of account just changed. 750MW is the headline, not “X number of chips.” OpenAI is telling the market that the bottleneck is now power-to-tokens conversion and guaranteed capacity delivery. 2) This is an inference-first move, aimed at behavior and benchmarks. OpenAI explicitly frames Cerebras as a way to make responses “much faster,” because when responses feel real-time, users stay longer and attempt higher-value work (agents, code, image generation). That is a revenue and retention play as much as a compute play. 3) “Portfolio compute” is now a strategy. OpenAI says the goal is to match “the right systems to the right workloads,” integrating Cerebras into its inference stack over time. OpenAI is building negotiating power and supply resilience by making its stack multi-architecture. 4) Cerebras just got an anchor tenant that changes its trajectory. Cerebras calls this the largest high-speed inference deployment and claims up to 15x faster responses than GPU systems for some workloads (their claim, but still telling). If OpenAI is willing to tune production workloads around this, it suggests parts of inference have become stable enough to justify specialization. 5) The second-order shock is to cloud positioning. When a model platform can buy capacity at this scale outside the usual hyperscaler path, it pulls the market toward “capacity contracts” with explicit delivery tranches and SLAs. That changes how procurement, financing, and risk get managed. 6) Beyond silicon supply chain, the real difference maker is execution. 750MW by 2028 means power interconnects, cooling, construction schedules, and operational reliability at scale. If any of that slips, capacity won’t show up in user experience. If you’re making decisions on AI infrastructure strategy (cost, supply risk, latency targets, vendor concentration), I and my colleagues, are happy to share deeper insights. Please reach out to your Gartner service team to engage with me or my colleagues Not a client? For access to free & compelling insights, check out https://gtnr.it/GExpert #AI #DataCenters #Compute #Semiconductors #EnterpriseAI

Delivering GPT5 - 200,000 GPUs for 'Planetary-Scale' AI OpenAI's GPT5 model is here, and many folks here on LinkedIn are offering first takes. So here's mine: Launching GPT5 required a LOT of data center infrastructure, and there's much more to come. Reports indicate OpenAI deployed 200,000 GPUs to support the launch of GPT5, and its compute infrastructure has grown 15X since 2024. The massive growth and future ambitions of OpenAI can be seen in its infrastructure journey. ChatGPT started out on the Microsoft Azure cloud, but soon needed even more capacity and began working with Oracle Cloud. Then came the announcement of the massive Stargate data center project, a deal to deploy capacity with CoreWeave, and most recently a deal to expand with Google Cloud. And OpenAI is actively hiring to build in-house data center capabilities. OpenAI's competitors are also scaling up their infrastructure - such as the Anthropic deployment across AWS' Project Rainier footprint, and as reflected in the massive CapEx expansion at Microsoft, Google, AWS and Meta. And more compute-intensive uses of AI are just beginning to scale. One example is generative AI video, which for most users consists of 10-second clips because it is so expensive in compute cycles. Meanwhile, ChatGPT has gone from launch to 700 million weekly users in less than three years. If usage of generative AI continues to grow at that pace, there will be much more data center infrastructure in our future.

OpenAI 🤝 Google Cloud The biggest news in AI infrastructure last week wasn't a new model launch. It was an update to a legal page (see link in the comments below). OpenAI is now officially using Google Cloud.! In my opinion this is a sign of how incredibly compute-hungry the AI industry has become. When you're operating at OpenAI's scale, a multi-cloud strategy isn't just an option - it is a necessity. This signals three major shifts for the industry: 1️⃣  The Rise of "Co-opetition": I believe this validates Google Cloud's strategy of being an open, neutral platform for the entire AI ecosystem. And for the market, it raises interesting questions about how specialised AI infrastructure is becoming the key differentiator. 2️⃣  Multi-Cloud is the New Default: This sets the blueprint for any company working at the frontier of AI. A sophisticated, multi-vendor, and multi-silicon (GPU, TPU) approach is now the standard for managing risk and securing capacity. 3️⃣  A New Strategic Layer for Enterprises: This introduces a new consideration for enterprise customers. as it makes deep diligence on a provider's operational maturity more critical than ever. Compute capacity is the ultimate enabler, and it isn't a zero-sum game. It's a sign of a maturing ecosystem where the sheer scale of AI requires a more sophisticated, collaborative approach across the industry. (Full disclosure: While my work at Google Cloud gives me a front-row seat to these shifts, this analysis is based purely on public information, and the thoughts are strictly my own.)