Nvidia’s most recent quarterly earnings, covering its fiscal first quarter of 2027 and reported in late May, delivered revenue of $81.62 billion, beating Wall Street’s average estimate of $78.86 billion, with profit roughly tripling year over year. It’s the kind of beat that’s become almost routine for Nvidia over the past two years, but the number worth paying attention to isn’t the headline revenue figure, it’s what the company’s forward order book says about demand still to come.
Data center revenue alone reached $39.1 billion for the quarter, up 69 percent year over year, following $62.3 billion in the prior quarter, a 75 percent year-over-year jump. Data center now makes up roughly 87 percent of Nvidia’s total revenue, and the company’s full fiscal year 2026 data center revenue is estimated in the $170 billion to $190 billion range. Nvidia has, for practical purposes, become a data center infrastructure company that happens to also sell gaming graphics cards.
The more telling figure sits in Nvidia’s supply-related commitments, the forward orders customers have locked in beyond the current quarter. Those commitments went from $50.3 billion at the end of the prior quarter to $95.2 billion by the end of the most recent one, nearly doubling in three months. A backlog of that size and growth rate is a stronger signal of sustained demand than any single quarter’s earnings beat, because it reflects customers, hyperscalers and increasingly sovereign AI infrastructure projects, committing capital years ahead of delivery rather than reacting to current-quarter results.
What makes the backlog especially notable is that it’s growing despite Nvidia being effectively locked out of selling its most advanced chips to China, historically one of its largest markets. The export-control history here is tangled: in April 2025 Washington required a license for H20 chip exports to China, forcing Nvidia to take a $4.5 billion charge on unsellable inventory, before granting limited licenses that August that generated only about $60 million in H20 revenue. In January 2026, the Trump administration eased restrictions further, allowing H200 chip sales to “approved” Chinese customers subject to a 25 percent government surcharge, with the first licenses for limited H200 shipments granted that February, though no revenue has yet been booked under that program. Nvidia’s newest and most capable architecture, Blackwell, remains entirely off-limits: despite Jensen Huang personally lobbying and joining the US delegation at a recent US-China summit, the export ban on Blackwell’s B100, B200, and B300 chips survived those talks intact. That a backlog this large is building even with the China market functionally closed to Nvidia’s best chips says something about how much demand exists everywhere else.
Not every analyst treats the backlog figure as an unambiguous green light. A signed supply commitment is not yet collected revenue, and a meaningful share of Nvidia’s forward book now consists of enormous individual deals, tens of billions of dollars each, with hyperscalers and sovereign buyers, concentrations large enough that any single customer delaying or restructuring a commitment could move the backlog figure noticeably. The number is real, but it’s a commitment, not a guarantee.
This has been a rocky year for individual frontier AI labs, which makes the backlog’s resilience more striking rather than less. Google delayed Gemini 3.5 Pro for a full architectural rebuild and lost roughly $225 billion in market value in the fallout, Anthropic briefly lost export clearance on its flagship model, and OpenAI has had its own release slow-walked by government oversight demands. Despite all of that visible turbulence at the model layer, and despite Nvidia’s own China exposure being squeezed by export controls, the underlying demand for the compute those models run on shows no sign of cooling, if anything, Nvidia’s numbers suggest it’s accelerating.
This matters well beyond Nvidia’s own stock price. Every AI infrastructure investment thesis built around the Philippines this year, including the reported $10 billion US-backed AI hub taking shape near Clark Freeport Zone and the separate wave of domestic data center construction now underway from PLDT, ST Telemedia, and others, ultimately depends on being able to secure GPU allocation at a price that makes the underlying project economically viable. Nvidia’s backlog data is the clearest global signal available that competition for that same compute isn’t easing, it’s intensifying, and that it’s intensifying among buyers who, unlike China, aren’t blocked from the newest chips at all, which means Philippine buyers are competing directly against the same hyperscaler and sovereign demand driving the backlog rather than against a market Nvidia is deliberately holding back.
Whether that global compute crunch eventually eases as more fabrication capacity comes online, or gets worse as sovereign AI ambitions multiply across more countries at once, is the open question that will determine how the Philippines’ own AI infrastructure bets play out over the next two to three years, and it’s a question being decided almost entirely outside the Philippines’ control, in trade negotiations and chip-fabrication timelines that have nothing to do with Philippine policy at all.
Share this article