In partnership with

Find your customers on Roku this Black Friday

As with any digital ad campaign, the important thing is to reach streaming audiences who will convert. To that end, Roku’s self-service Ads Manager stands ready with powerful segmentation and targeting options. After all, you know your customers, and we know our streaming audience.

Worried it’s too late to spin up new Black Friday creative? With Roku Ads Manager, you can easily import and augment existing creative assets from your social channels. We also have AI-assisted upscaling, so every ad is primed for CTV.

Once you’ve done this, then you can easily set up A/B tests to flight different creative variants and Black Friday offers. If you’re a Shopify brand, you can even run shoppable ads directly on-screen so viewers can purchase with just a click of their Roku remote.

Bonus: we’re gifting you $5K in ad credits when you spend your first $5K on Roku Ads Manager. Just sign up and use code GET5K. Terms apply.

Use code GET5K now

Hey, Josh here. These stories from this last week are wild.

The $600 Billion Cloud Bet: How OpenAI Just Rewired the AI Economy

Listen, we need to talk about what just happened with OpenAI and cloud infrastructure, because it's not what it looks like on the surface.

The headline that dropped this week was clean enough: OpenAI signed a $38 billion deal with Amazon Web Services. Big number, sure. But here's the kicker—that's just one piece of a roughly $600 billion cloud commitment spanning AWS, Microsoft Azure, and Oracle.

Six. Hundred. Billion. Dollars.

To put that in perspective, that's more than the GDP of Poland. It's roughly the market cap of Visa. And it's all getting funneled into compute capacity for a company that, until recently, was primarily known for a chatbot that writes college essays and generates mediocre poetry.

What's actually going on here?

The Surface Story

OpenAI just restructured its entire cloud stack. The company agreed to buy $250 billion of Microsoft Azure capacity, locked in the $38 billion AWS deal for Nvidia's latest GPUs, and struck a separate multibillion-dollar arrangement with Oracle Cloud. These aren't yearly budgets—these are long-term commitments that essentially hard-wire OpenAI into the capital expenditure plans of three hyperscalers for the next decade or more.

The immediate market reaction told you everything: Amazon and Nvidia stock popped. Investors instantly repriced both companies higher, digesting the reality that decades-long AI workloads just got locked into a handful of vendors. Nvidia briefly hit a $5 trillion market cap this week, making it—for a moment—the world's most valuable public company.

Peeling Back the First Layer: OpenAI Isn't a Startup Anymore

Here's what this week confirmed: OpenAI is now an industrial-scale cloud utility customer. Not a promising AI lab. Not even a hot startup. They're operating at the scale of Netflix's streaming infrastructure, except instead of movies, they're serving up reasoning tokens and multimodal outputs to hundreds of millions of users.

Think about the physics here. Every ChatGPT query, every DALL-E image, every API call to GPT-4—all of that runs on someone else's hardware. OpenAI doesn't own data centers the way Google or Meta do. They rent at massive scale. And this week, they essentially pre-paid for the next era of that rental agreement.

The thing is, when you commit $600 billion to cloud capacity, you're not just buying compute. You're making a statement about your growth assumptions, your capital structure, and frankly, your ability to generate enough revenue to make those payments without going bust.

The Deeper Mechanism: How Cloud Deals Became AI Battlegrounds

Let's break down what's really happening with these hyperscaler relationships.

Microsoft has skin in the game. They're not just a vendor—they're an investor. Microsoft put roughly $13 billion into OpenAI and gets a significant cut of profits up to a cap. The Azure commitment isn't just a purchase order; it's part of a complex dance where Microsoft provides infrastructure and gets preferential access to OpenAI's models for their own products (Copilot, GitHub, etc.). It's symbiotic, but also creates weird incentives. Microsoft wants OpenAI to succeed, but not too much—not enough to become a direct competitor.

Amazon was the odd man out. AWS historically powered much of OpenAI's research infrastructure, but as the Microsoft investment deepened, OpenAI tilted heavily toward Azure. This $38 billion deal brings Amazon back into the fold, but with a specific angle: access to Nvidia's latest GPUs hosted on Amazon's infrastructure. Translation: OpenAI needs more than one vendor to avoid getting held hostage on pricing, and Amazon needs to prove to Wall Street they're not losing the AI infrastructure race.

Oracle is the wildcard. Oracle Cloud isn't typically mentioned in the same breath as AWS and Azure, but they've been quietly building out GPU capacity and positioning themselves as the "performance" option for AI workloads. For OpenAI, Oracle represents geographic diversity, specialized hardware access, and leverage in vendor negotiations. For Oracle, this deal is a lifeline—proof that they're relevant in the AI era and not just a legacy database company.

What This Reveals About Larger Forces

1. AI Is Now Systemic Market Risk

When Nvidia's market cap swings based on OpenAI's cloud deals, and when Bridgewater—one of the world's largest hedge funds—starts waving yellow flags about AI euphoria, you're watching a theme become a systemic risk.

The current market structure has an unprecedented level of concentration. A handful of mega-cap tech companies represent a disproportionate share of the S&P 500's gains. Nvidia alone is now worth $5 trillion, roughly 7% of the entire U.S. stock market. If anything breaks—chip export controls, AI regulation, a major model failure—the contagion risk is enormous.

Bridgewater's CIOs specifically warned investors are underpricing the risks to the AI-fueled rally, pointing to concentration in a few names and policy uncertainty around chips. They're basically saying: "Yeah, this looks like a super-cycle, but remember that every 'this time is different' narrative eventually gets tested."

2. The "AI Factory" Becomes a Product Category

Here's where it gets interesting. The same week OpenAI announced these deals, Nvidia was out in South Korea unveiling "AI factories" with Samsung, Hyundai, and SK Group. These aren't just loose GPU sales—they're packaged, industrial AI compute plus software, designed to power national-scale digital transformation.

The pattern: A national government teams up with its conglomerates and Nvidia to build shared AI infrastructure instead of each firm going it alone. South Korea's doing it. The U.S. Department of Energy just announced a similar plan with Nvidia and Oracle for America's largest AI supercomputer, framed explicitly as maintaining leadership over China.

This is sovereign AI infrastructure. It's industrial policy meets chipmaking meets cloud services. And it's become a template that's getting copied globally.

3. Geopolitics Is Baked Into the Stack

Remember when tech infrastructure was borderless? Yeah, that's over.

China's President Xi used the APEC summit to push for a "World Artificial Intelligence Cooperation Organization"—essentially a China-led alternative to U.S./EU AI governance frameworks. His pitch: AI should be a "public good for the international community," with China helping set the rules.

Meanwhile, the U.S. is expanding "Technology Prosperity" deals with Japan and South Korea that explicitly bundle quantum computing and AI collaboration—supply chains, research, talent pipelines, the whole thing. AI isn't just an industry anymore. It's a pillar of security alliances in the Pacific.

The world is splitting into AI blocs. One camp around U.S./EU norms and infrastructure. Another around China-centric "algorithmic sovereignty." OpenAI's cloud deals are part of that story—locking American AI leadership into American (and allied) cloud vendors.

The Energy and Infrastructure Crunch

Here's the part nobody wants to talk about: all this compute needs power. Lots of it.

A report from the Conference Board of Canada this week warned that AI-driven data center build-out is straining electricity and water systems. Proposed Alberta data centers could consume thousands of megawatts and require tens of thousands of liters of water per day. Some projects are already getting blocked.

Natural Resources Canada is positioning AI as both a tool for decarbonization and a driver of new energy demand. That's a polite way of saying: "We need AI to solve the energy crisis, but AI is also making the energy crisis worse."

For OpenAI, this is a real constraint. You can't just scale compute infinitely if the grid can't support it. Power availability is becoming a gating factor for AI growth, not just chips and capital. The $600 billion cloud bet assumes those electrons will be available. That's not guaranteed.

The Funding Follow-Through: Where the Money Actually Went This Week

While OpenAI was making headlines with hyperscale deals, venture capital was quietly flowing into vertical AI plays that tell a more nuanced story about where defensible moats actually exist.

Legora, a Swedish legal AI startup, raised $150 million at a $1.8 billion valuation—one of Europe's higher-valued vertical AI players. They're automating document review, contract workflows, and legal research. High billable hours, clear ROI, obvious defensibility.

Graph AI raised $3 million to apply graph-based AI to pharmacovigilance—detecting adverse drug event patterns in real time. This is narrow, high-stakes, and deeply technical. Not sexy, but critical.

The Prompting Company raised $6.5 million to help brands get mentioned inside ChatGPT and other AI apps. Think of it as "AI shelf placement"—pay to show up in AI assistant responses. It's early-stage SEO for the LLM era, and someone's betting $6.5 million it becomes a real market.

The pattern: Investors aren't chasing yet another general-purpose LLM. They're going into narrow, high-margin vertical plays where AI delivers obvious, quantifiable value. Law, pharma safety, AI-native marketing, SME productivity tools.

This matters because it suggests the next wave of AI value creation isn't at the foundation model layer—it's in the applications that sit on top of those models and solve specific, expensive problems.

What Happens Next?

OpenAI's $600 billion cloud bet is a forcing function. It locks in a cost structure that requires massive, sustained revenue growth. ChatGPT Plus subscriptions and API revenue are substantial, but nowhere near enough to service commitments of this magnitude.

So what are the options?

1. Enterprise goes huge. OpenAI needs to become the default AI backend for every Fortune 500 company. That means competing directly with Microsoft, Google, and Amazon on enterprise sales—awkward, given those are also their cloud vendors.

2. Consumer scales exponentially. ChatGPT needs to become as essential as Google Search. That means not just users, but monetizable engagement—subscriptions, ads, commerce. The launch of Sora in Asia this week (Thailand, Vietnam, Taiwan) is part of that play. AI video is going consumer, fast.

3. New revenue streams emerge. AI hardware deals, licensing foundation models to governments, charging for API access at scale to thousands of vertical AI startups. OpenAI becomes infrastructure for infrastructure.

4. The capital structure gets creative. Maybe those cloud commitments get restructured. Maybe OpenAI goes public sooner than expected to raise growth capital. Maybe one of the hyperscalers ends up acquiring outright, though regulatory scrutiny makes that tricky.

The stakes are absurd. If OpenAI executes, they lock in a decade of AI dominance and validate the entire hyperscaler capex boom. If they stumble—if growth slows, if a competitor leapfrogs them technically, if regulation kneecaps their business model—it's not just an OpenAI problem. It's an Amazon problem, a Microsoft problem, an Nvidia problem. It cascades.

The Bigger Picture: AI as Critical Infrastructure

What we're watching unfold isn't just a bunch of cloud contracts. It's the industrialization of AI—the moment a technology stops being experimental and starts being weight-bearing infrastructure for the global economy.

Geoffrey Hinton, speaking at the University of Toronto's "Who's Afraid of AI?" conference this week, argued we need to design "maternal AI" systems that care about humans once they surpass us. Fei-Fei Li emphasized shared responsibility and AI's potential to transform education. These are important conversations.

But alongside the philosophical questions about alignment and ethics, there's a harder, more immediate reality: AI is now a critical load on grids, water systems, and supply chains. It's a board-level governance topic with disclosure expectations. It's a pillar of national security alliances. It's systemic market risk.

The $600 billion cloud bet is OpenAI saying: "We're going all-in on this future." And dragging Amazon, Microsoft, Oracle, Nvidia, and by extension, every investor holding those stocks, along for the ride.

The question isn't whether AI is transformative. That's settled. The question is whether the infrastructure—technical, financial, regulatory, physical—can actually support the scale of transformation we're betting on.

And whether OpenAI, specifically, can generate enough value to justify commitments that make most countries' annual budgets look small.

Here's the kicker: We're about to find out.

Links and Signals

• Nvidia hitting $5T the same week as these deals isn't coincidence—it's the market pricing in decades of AI workload lock-in

• Oracle's inclusion in the deal is a quiet signal: even "legacy" vendors can matter in AI if they move fast on specialized infrastructure

• Watch for power availability and grid capacity to become a recurring theme in 2025 AI earnings calls

• The vertical AI funding (legal, pharma, marketing) is where the next wave of billion-dollar outcomes likely hides—not in foundation models

• If you're not thinking about AI as geopolitical infrastructure yet, you're already behind