Investors point to a rare mix that doesn’t usually show up at the same time: a deficit near the mid‑single digits as a share of GDP and a central bank that is easing rather than tightening, which is the inverse of the late‑90s backdrop when Washington ran a surplus and the Fed was lifting rates. In 2025, multiple independent trackers and official data peg deficits in the ballpark of 5.6%–6.1% of GDP, while the Fed has already cut once and signaled more reductions, a setup that can keep risk appetites stretched even as fundamentals whisper caution. Back in 1999, the U.S. posted a roughly $124 billion surplus near 1.4% of GDP and the Fed delivered rate increases into a hot economy, which is a very different policy mix than today’s. That’s why late‑cycle surges can still appear—fast, glowing, and hard to fade—before a final top asserts itself.
At the same time, the AI build‑out is entering a phase that feels thrilling and a little hazardous. The numbers are gigantic, the timelines are aggressive, and the equity market is rewarding promises at a breathtaking clip. Read the headlines, then read the footnotes—because underneath the excitement sits a set of contracts, letters of intent, and staged deployments that stretch well into 2026 and beyond.
▪️ OpenAI has inked what’s been described as a $300 billion, multi‑year cloud compute agreement with Oracle, framed inside the broader Stargate effort to stand up industrial‑scale AI infrastructure across the U.S. with power targets measured in gigawatts. Reporting and industry analysis point to a plan that adds roughly 4.5 GW of capacity via Oracle, with five new U.S. sites announced to push the pipeline toward 7 GW and keep a longer‑term 10 GW ambition in play. Even boosters acknowledge the financing strain: estimates suggest Oracle may need to lean heavily on debt markets to support the outlays tied to the contract schedule. In short, the scope is national in scale, the intent is clear, and the delivery is paced over several years rather than several quarters.
▪️ Markets reacted in a flash—Oracle’s shares posted their biggest one‑day jump in decades, mid‑to‑high 30s in percentage terms, as backlog disclosures and the OpenAI pact landed, and Larry Ellison briefly topped the global wealth rankings during the surge. That move followed upbeat commentary about record remaining performance obligations and a wave of AI‑linked contracts, the kind of forward pipeline that equity traders love in a momentum phase. The wealth‑ranking shuffle was fleeting, but it captured the mood: one contract cycle can reprice hundreds of billions across a cluster of names, almost overnight.
▪️ On chips, Oracle has been vocal about sourcing Nvidia hardware for its AI regions, and in parallel OpenAI and Nvidia announced a sweeping partnership to deploy at least 10 GW of Nvidia systems—with Nvidia indicating it intends to invest up to $100 billion in OpenAI as each gigawatt comes online. Timelines point to the first gigawatt landing in the second half of 2026 on Nvidia’s Vera Rubin platform, translating to millions of GPUs over the life of the plan and extending the runway for Nvidia’s top‑line story. Nvidia’s stock response underscored the theme: even incremental clarity on multi‑year capacity can tack on well over $100 billion in market value in a single session during an AI‑led tape.
▪️ Then came AMD’s turn: OpenAI agreed to deploy up to 6 GW of AMD Instinct GPU capacity over multiple generations, starting with an initial 1 GW milestone in 2026, and received a warrant covering up to 160 million AMD shares that vests alongside deployment and price targets. The structure aligns incentives—compute rollouts and share‑price thresholds unlock tranches—and the market response was immediate, with AMD spiking more than thirty percent as investors priced in “tens of billions” in potential revenue from the arrangement. It’s a classic late‑cycle tell: the challenger secures a marquee partner on ambitious terms, and the equity crowd rushes to extrapolate the out‑years.
▪️ On the ground, the concrete is starting to pour—but much of the vaunted capacity is still pre‑operational, with staged rollouts tied to power, permitting, and next‑gen silicon that won’t be live until 2026. Multiple Stargate sites have been announced, with at least one flagship campus publicized as under construction and tours showing early structural progress, yet broad, multi‑gigawatt output remains a future tense story. Meanwhile, the equity market has already moved—Oracle’s jump alone vaulted it toward the trillion‑dollar conversation, while Nvidia’s single‑day gains have reached roughly $170 billion—so it doesn’t take much arithmetic to see how “hundreds of billions” in market value can reprice ahead of concrete results.
That is the picture—dazzling headlines about capital commitments and compute, strong price action, and a policy backdrop that can keep risk buoyant longer than skeptics expect. Yet the echoes of 1999 are hard to ignore: back then, profits were deferred to a later chapter, rates were heading up, and Washington’s surplus offered a cushion; now deficits are large, rates are edging down from high levels, and the race is to build first and scale later. Beautiful? Sure, in the way a firework is beautiful—brilliant, loud, and brief—and that means the unwind, if and when it comes, can be just as dramatic.