Nvidia's Two-Million-GPU AWS Order Puts Hard Numbers Behind the Hype

AWS commits to 2M additional Nvidia GPUs for 2027-28, easing demand concerns. Q2 revenue surged 106% to $96.2B, but stock fell 4.1% on Friday.

The gap between Nvidia's soaring revenue projections and the skepticism that has trailed its stock all summer narrowed considerably this week. Amazon Web Services has committed to installing two million additional Nvidia GPUs across its data centers in 2027 and 2028, a deal that gives the chipmaker something its critics have said it lacked: concrete, multi-year demand commitments from a financially muscular cloud operator.

The AWS order lands on top of an earlier pledge for more than a million units and arrives in a week when Nvidia's earnings already dominated market conversation. For a company whose growth narrative has increasingly been questioned as circular — Nvidia has poured roughly $50 billion into AI labs that subsequently buy its own chips — the hyperscaler deal offers a counterweight. It demonstrates that established, deep-pocketed cloud providers are willing to lock themselves into Nvidia's roadmap years in advance.

The Quarter That Keeps Raising the Bar

Nvidia's fiscal second quarter delivered revenue of $96.2 billion, up 106 percent year over year, with the data center segment alone climbing 117 percent to $89 billion. Management guided for current-quarter revenue of $108 billion, plus or minus 2 percent, and signaled roughly 70 percent growth for fiscal 2028 — a figure Morgan Stanley noted sits well above its own 52 percent estimate and the roughly 40 percent market consensus.

The customer base is shifting as fast as the growth rate. Established hyperscalers lifted their contribution 102 percent to $48.7 billion, but a newer class of buyers — AI-native clouds, industrial firms, sovereign projects and specialists like CoreWeave and Nebius — expanded even faster, adding 138 percent to reach $40.3 billion. That diversification spreads the concentration risk inherent in relying on a handful of tech giants for a supplier's future.

The Price of Growth

This expansion carries a literal cost. Memory-chip shortages have pushed Nvidia to raise prices on its Vera Rubin and Grace Blackwell AI servers by 15 to 17 percent for deliveries from 2027 onward. Memory now accounts for roughly a quarter of a rack's cost, according to industry estimates, while the GPU share has fallen from over 80 percent to about 50 percent. New memory fabrication capacity is not expected online until 2029 or 2030, and TSMC, Nvidia's key manufacturing partner, plans price increases of 10 to 15 percent across all process nodes.

Management has itself flagged memory constraints as the limiting factor on growth. The Vera Rubin processor, whose deliveries begin this autumn, is seen as the next catalyst — and if AWS's additional capacity runs on that architecture, it would further cement Nvidia's position as the preferred supplier for AI infrastructure.

A Financing Machine in Chipmaker's Clothing

The company's transformation into something beyond a semiconductor vendor has become impossible to ignore. CFO Colette Kress put the share of expected fiscal 2028 business coming from financed customers at about 25 percent. Nvidia is building financing platforms exceeding $500 billion with partners including Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR. It has also committed up to $105 billion in leasing pledges for OpenAI's Ohio campus, alongside a $1.5 billion investment in SB Energy.

Critics call this circular financing — Nvidia funding the very demand it books. Supporters argue these capital bridges are the only way many AI clouds can order the computing capacity they need. Both camps have a point, and the debate is unlikely to resolve with a single quarter's numbers.

Market Mood: Reassured, Then Wary

The stock's reaction to the earnings print told its own story. After an initial surge, gains faded. By Friday's close, shares sat at €187.78, down 4.1 percent on the day. Over seven trading days the stock is still up 2.1 percent, and on a 30-day view it has gained 13 percent. That leaves it roughly 7.3 percent below its 52-week high of €202.50 from May, but comfortably above the €139.78 low from September last year and its 200-day moving average.

The annualized 30-day volatility of 45 percent reflects how violently investors are swinging the stock in both directions — hardly surprising given the torrent of news around financing deals, capacity expansion and growth forecasts. Analyst targets have moved higher: Truist lifted its price objective to $346, JPMorgan to $320 and Bernstein to $400, against a market average just above $320. Yet insiders have sold more than $664 million worth of shares this year, much of it from a single director — routine for a company this size, but a data point that keeps the summer's underlying question alive: Nvidia is growing at a pace that regularly surprises even seasoned market participants, and that very speed generates doubts no single earnings report can fully put to rest.