Encouraging Signs for the AI Industry Off the Back of Nvidia Earnings

2026-09-02

Key Takeaways

Nvidia reported $96.2 billion of revenue for the second quarter of fiscal 2027, up 106% year over year and 4.5% ahead of consensus.
Data centre revenue reached $89.0 billion, up 117% year over year and 93% of the company total.
Management guided to roughly 70% revenue growth for fiscal 2028 — about $690 billion — against a prior Wall Street estimate near 44%, and described that figure as supply-constrained rather than demand-constrained.
Customers outside the big cloud providers now account for 45% of data centre revenue, at $40 billion and growing 138% year over year.
Supply and capacity commitments jumped from $119 billion to $279 billion in a single quarter, with total future commitments around $366 billion.
Vera Rubin entered production shipment in August and is expected to make up about 20% of third-quarter data centre revenue.
Jensen Huang told investors: “AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable.”
The clearest negative is margin: memory cost inflation pushes gross margin from 75.0% now to a 71–72% trough in the fourth quarter before recovering to 72–73%.

Nvidia’s Jensen Huang and Olaf the robot. (YouTube screenshot)

Nvidia’s August 2026 results gave the AI industry the thing it needed most: evidence that spending on compute is being matched by revenue on the other side. Revenue of $96.2 billion for the quarter, data centre sales of $89.0 billion, and guidance of roughly 70% growth for the next full year all landed above expectations. The share price rose 8.7% the following day, the largest single-session gain since April 2025.

The encouraging part is not the size of the numbers but their composition. Growth came increasingly from customers who are not the four largest cloud providers, the company gave an unusually long forward guide, and management framed the constraint as supply rather than appetite. Those three details answer the question that has hung over the sector for two years — whether AI infrastructure spending is being absorbed by paying end demand, or simply circulating between a small group of balance sheets.

The Headline Numbers

Metric	Q2 FY2027	Change
Total revenue	$96.2 billion	+106% year over year, +18% quarter over quarter
Data centre revenue	$89.0 billion	+117% year over year
Edge computing revenue	$7.2 billion	+27% year over year
GAAP diluted EPS	$2.46	+128% year over year
Non-GAAP diluted EPS	$2.22	Beat $2.09 consensus
Gross margin	75.0%	In line with guidance
Free cash flow (six months)	$69.9 billion	—
Q3 FY2027 revenue guidance	$108 billion ±2%	~89% year-over-year growth

Encouraging Sign One: The Customer Base Is Widening

For most of the current cycle, a handful of hyperscalers carried Nvidia’s growth, which made the whole sector look like a bet on four purchasing departments. That has changed. Hyperscale customers contributed $49 billion in the quarter, up 13% sequentially. The category Nvidia labels alternative compute infrastructure and enterprise — specialist GPU clouds, corporates and sovereign programmes — contributed $40 billion, up 25% sequentially and 138% year over year.

Non-hyperscale buyers now represent 45% of data centre revenue and are on course to become the larger half. Sovereign AI programmes and enterprise deployments have longer procurement cycles and different funding sources than cloud capital budgets, so a customer mix spread across both is more resilient than one concentrated in either. For anyone tracking the best-performing AI stocks of 2026, this diversification is the most consequential line in the release.

Encouraging Sign Two: Supply, Not Demand, Is the Limit

The fiscal 2028 guide of about 70% growth arrived roughly 26 percentage points above the consensus estimate, and management was explicit that the figure describes what Nvidia can build rather than what customers want. Chief financial officer Colette Kress pointed to demand consistent with growth closer to 100%, with the company able to supply only part of it.

A supply ceiling is a far better problem for an industry than a demand ceiling. It means order books, not inventory, set the pace — and it explains why hyperscalers keep raising capital expenditure. Combined 2026 capex across the five largest US technology companies is running at roughly $660–690 billion, close to double the $380 billion spent in 2025, with Amazon around $200 billion, Alphabet $175–185 billion, Microsoft above $120 billion, Meta $115–135 billion and Oracle near $50 billion.

Encouraging Sign Three: Inference Is Producing Revenue

Huang’s framing was blunt: “AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue. And demand is accelerating.” Kress added that “the surge in AI demand is driving a global infrastructure build-out” across hyperscalers, labs, enterprises and sovereign initiatives.

Those statements matter because the economics of inference differ from the economics of training. Training is a capital project with an uncertain payoff. Inference is a metered service with a per-request price, which is why tokens have become the billing unit of generative AI. A quarter in which inference workloads drive record networking revenue — up 18% sequentially to an all-time high — describes a sector selling output rather than buying capacity on faith.

Encouraging Sign Four: Commitments Nearly Doubled

Nvidia’s supply and capacity commitments rose from $119 billion to $279 billion in three months, bringing total future commitments to roughly $366 billion. That is Nvidia putting cash behind foundry allocation, memory contracts and packaging capacity years ahead of shipment.

Read one way, it is the strongest possible statement of confidence from the company with the clearest view of order flow. Read another way, it converts a flexible cost base into fixed obligations, and it is the number that would hurt most if demand cooled. Both readings are correct; the size of the commitment simply raises the stakes on the forecast.

Encouraging Sign Five: The Product Cadence Held

Vera Rubin began production shipments in August and is projected to account for around 20% of third-quarter data centre revenue, which Nvidia described as its fastest product ramp to date. Hitting a schedule on a new platform in a supply-constrained market is a meaningful operational result, particularly after packaging difficulties earlier in the Rubin Ultra programme. The generational story running from Blackwell Ultra to Vera Rubin is now on its published timetable.

The Part That Is Not Encouraging

Gross margin is heading down. Memory pricing has reached what management called extreme conditions, and the guidance path runs from 75.0% now to 74.0% next quarter, a 71–72% trough in the fourth quarter, and 72–73% through fiscal 2028 as contracted price increases take effect. HBM supply is concentrated among a small number of producers, and the pricing power currently sits with them rather than with their customers — a dynamic set out in more detail in South Korea’s role in AI memory chip production.

China is effectively gone. Less than 1% of second-quarter data centre revenue shipped to Chinese customers, and guidance assumes zero going forward. That is prudent forecasting rather than a shock, but it removes a market that once contributed a meaningful share of sales.

Execution risk scales with everything else. Committing $366 billion against a supply chain that must roughly double output leaves little room for a packaging delay, a memory shortage or a power connection that arrives late. And a 70% growth guide issued a year in advance is a hostage to fortune if enterprise adoption plateaus.

What This Means for the Wider AI Industry

Nvidia sells to almost everyone building AI systems, which makes its order book the sector’s most reliable leading indicator. On that measure the picture is of demand outrunning manufacturing across more customer types than a year ago, with inference workloads generating revenue rather than consuming venture funding.

The constraints have moved accordingly. Chips are no longer the scarce item in the way they were in 2024; memory, power and construction schedules are. That shifts attention toward electricity supply, where AI’s power requirements now determine how quickly ordered hardware can actually be switched on, and toward the question of how long the current growth rate can hold before the market reaches the scale analysts project when the AI market approaches $5 trillion.

One quarter does not settle the argument about whether AI infrastructure spending is proportionate. It does show that, as of August 2026, the buyers are more numerous, the revenue is broader-based, and the bottleneck is a factory rather than a customer.

If you are interested in this topic, we suggest you check our articles:

Sources: Nvidia Q2 FY2027 press release (SEC), Nvidia Investor Relations, Nvidia Q2 earnings call highlights, Kiplinger, CNBC, Futurum Group

Written by Alius Noreika