Nvidia Q3 Fiscal 2027 Earnings Explained: $115.8B Revenue, Vera Rubin Ramp, Networking and What It Means for AI Infrastructure Buyers

Nvidia Q3 Fiscal 2027 Earnings Explained: $115.8B Revenue, Vera Rubin Ramp, Networking and What It Means for AI Infrastructure Buyers

Nvidia Q3 Fiscal 2027 Earnings Explained: $115.8B Revenue, Vera Rubin Ramp, Networking and What It Means for AI Infrastructure Buyers

Nvidia just booked $115.8 billion in one quarter, $7.8 billion above its own midpoint guidance, and the stock still fell. The nvidia q3 fiscal 2027 earnings landed on 17 November 2026, five days after Jensen Huang resigned as CEO and one day after Washington tightened AI chip export rules. Most coverage treats the print as a market story about a beat, a CEO exit and a sell-off. For people who actually buy GPUs, racks and fabric, the same release reads very differently. It is a disclosure of how much supply Nvidia has locked up, how firmly it is pricing, how fast Vera Rubin is replacing Blackwell, and how much of every cluster dollar now goes to networking. You will leave with a verified metrics table, the derived numbers that matter for capacity planning, and a procurement playbook for the next three quarters.

What this covers: verified Q3 numbers and their sources, the Vera Rubin ramp, networking attach, supply commitments and pricing power, power constraints, risks from leadership change and export controls, and a buyer checklist.

Disclaimer: this is a systems and infrastructure analysis, not investment advice.

Context and Background

Nvidia’s fiscal year ends in late January, so fiscal 2027 runs roughly from February 2026 to January 2027. The third quarter ended on 25 October 2026. The sequence of the year tells the story: revenue was $81.6 billion in Q1, $96.2 billion in Q2 and $115.8 billion in Q3, with $130 billion guided for Q4. That is a business adding roughly $15–20 billion of quarterly revenue every three months.

The company also changed how it reports. The Q3 press release shows two market platforms, Data Center and Edge Computing, instead of the older Gaming, Professional Visualization and Automotive lines. Inside Data Center, the CFO commentary splits revenue by customer type: Hyperscale versus AI Clouds, Industrial and Enterprise (ACI&E). That is more useful for buyers than the old split, because it shows who is absorbing supply. It is also less comparable with older posts, including our earlier Nvidia earnings analysis focused on Blackwell Ultra, which worked from the pre-change segment view.

Three events framed this print. First, on 12 November 2026 Nvidia announced that Jensen Huang had resigned as President, CEO and board member, effective immediately. CFO Colette Kress became interim CEO while remaining CFO, and Harvey C. Jones became the company’s first Chair of the Board. The company gave no reason, and on the call it declined to discuss the circumstances or what it called an ongoing investigation. Second, on 16 November the Bureau of Industry and Security (BIS) issued a 132-page interim final rule restricting advanced AI chip sales to overseas third-party data centers linked to countries of concern. Third, the stock had already fallen sharply in the week before results.

The incumbents around Nvidia matter too. Hyperscalers keep growing their own silicon, as covered in our analysis of hyperscaler custom AI silicon. AMD is pushing its MI400 generation. Yet the Q3 numbers show that Nvidia’s share of the accelerator dollar is not shrinking fast enough to change procurement dynamics this cycle. The primary sources are Nvidia’s Q3 fiscal 2027 press release and the Q3 FY27 CFO commentary.

The thesis of this article is simple. The headline beat is the least interesting number in the release. The important disclosures are a $390 billion supply commitment book, a guided gross-margin trough, a networking line now above $21 billion a quarter, and price evidence showing older GPUs re-contracting higher. Together they say that buyers should plan for scarcity and firm pricing through at least the first half of calendar 2027, not for relief.

Nvidia Q3 Fiscal 2027 Earnings: The Numbers That Matter for Infrastructure Buyers

Nvidia Q3 fiscal 2027 earnings showed revenue of $115.8 billion, up 103% year over year and 20% sequentially, with Data Center at $107.6 billion. Networking exceeded $21 billion, Vera Rubin contributed nearly $20 billion in its first production quarter, gross margin was 74.8% GAAP, and Q4 is guided to $130 billion plus or minus 3%.

The table below is the snippet-ready summary. Every figure comes from Nvidia’s own release, CFO commentary or earnings call. Derived values are marked as such.

Metric Q3 FY27 value Change Source
Total revenue $115.8B +20% q/q, +103% y/y Press release
Q3 guidance midpoint (given in August) $108.0B ±2% beat by $7.8B Q2 FY27 outlook
Data Center revenue $107.6B +21% q/q, +110% y/y Press release
Hyperscale revenue $60.1B +23% q/q, +71% y/y CFO commentary
AI Clouds, Industrial and Enterprise $47.5B +18% q/q, +195% y/y CFO commentary
Networking revenue over $21B not disclosed Earnings call
Rubin platform revenue nearly $20B first full quarter Earnings call
Edge Computing revenue $8.2B +14% q/q, +42% y/y Press release
China Data Center Hopper shipments $2.4B not in Q4 guide CFO commentary
GAAP gross margin 74.8% 75.0% in Q2 Press release
GAAP net income $65.3B +105% y/y Press release
Inventory $39.7B $31.6B in Q2 CFO commentary
Supply and capacity commitments $390B $279B in Q2 CFO commentary
Free cash flow $43.8B n/a Press release
Q4 FY27 revenue outlook $130.0B ±3% includes an extra week Press release
Q4 FY27 gross margin outlook 71.0–72.0% guided trough Press release

Nvidia Q3 fiscal 2027 earnings revenue decomposition from total revenue to data center, networking and Rubin

Figure 1: How Nvidia’s $115.8 billion Q3 FY27 revenue decomposes, from total to platform, customer type and product line.

Figure 1 shows the revenue tree. Total revenue splits into Data Center at $107.6 billion and Edge Computing at $8.2 billion. Data Center is shown two ways: by customer, as Hyperscale and ACI&E, and by product, as networking above $21 billion and an implied compute residual below $86.6 billion. The compute branch then shows the two items management called out, Rubin at nearly $20 billion and licensed Hopper shipments to China at $2.4 billion. The compute residual is our derivation, not a reported figure.

The Vera Rubin ramp absorbed almost all sequential growth

The cleanest derived insight in the quarter comes from two numbers. Data Center grew by $18.6 billion sequentially, from $89.0 billion to $107.6 billion. Management said the Rubin platform generated nearly $20 billion in its first production quarter, with shipments starting in August. Because Q3 ran from late July to 25 October, almost all Rubin revenue landed in this quarter.

The arithmetic implies that everything else in Data Center was roughly flat. Blackwell Ultra, older Blackwell, Hopper and non-Rubin networking together moved from about $89 billion to about $88 billion. That fits Nvidia’s own wording that Blackwell Ultra continued to grow, if China Hopper and other legacy lines shrank, but the release does not give that split. The key point for buyers is the transition speed. Management said hyperscalers brought Vera Rubin into production in about one month, half the time Grace Blackwell needed.

This is the fastest product transition Nvidia has disclosed. It matters because generational overlap is when pricing gets interesting. During a fast transition, the outgoing part can either get cheaper, as supply frees up, or hold price, if demand still exceeds supply. The Q3 evidence points firmly to the second case, as the pricing section below shows.

Networking is now a $21-billion-a-quarter business inside the data center

Networking revenue above $21 billion means networking is at least 19.5% of Data Center revenue. Put differently, for every five dollars of Nvidia data center revenue, about one dollar is NVLink switches, NICs, DPUs, Ethernet and InfiniBand switching, or optics. At a run rate, that is more than $84 billion a year, which by itself would make it one of the largest networking businesses in the industry.

The mechanism is architectural, not a sales tactic. A Vera Rubin NVL72 rack contains 72 Rubin GPUs and 36 Vera CPUs, connected by NVLink 6 at 3.6 TB/s per GPU and 260 TB/s across the rack switch fabric, per Nvidia’s preliminary specifications. Each GPU also gets a ConnectX-9 SuperNIC with 1.6 Tb/s of scale-out bandwidth, for 28.8 TB/s per rack. The NVLink switch trays sit inside the rack, so buying a rack-scale system buys Nvidia scale-up networking by default. Scale-out is where buyers still have choice, which is covered in our NVLink vs InfiniBand vs RoCE comparison for collectives.

Management also said co-packaged optics (CPO) scale-out switching is now in volume production. It cited a Lambda deployment of Photonics CPO switches connecting more than 10,000 GB300 GPUs. Versus pluggable optics, Nvidia claimed 5x better power efficiency, 15x higher reliability and 10% lower latency. Those are vendor claims without a published methodology, but the direction matters: optics is moving from a third-party transceiver line item into Nvidia’s switch bill of materials.

The customer mix moved toward neoclouds and enterprise

Hyperscale revenue was $60.1 billion, or 55.9% of Data Center. ACI&E was $47.5 billion, or 44.1%. ACI&E grew 195% year over year against 71% for Hyperscale. In plain terms, the fastest-growing buyers are not the big four clouds but GPU clouds, sovereign projects, industrial users and enterprises.

That shift has two consequences. First, Nvidia is less dependent on any single hyperscaler’s capex decision. Second, it is more exposed to the credit quality of smaller buyers, which is exactly what the market worried about after the call. Management said frontier labs would contribute about 15% of Data Center revenue this year. Sovereign AI revenue more than tripled year over year, and financial-services on-premises revenue reached nearly $5 billion on a trailing twelve-month basis.

Margins dipped on purpose, and Q4 is the guided trough

GAAP gross margin slipped from 75.0% to 74.8%, which the CFO commentary attributes to the initial Rubin ramp. The Q4 outlook steps down to 71–72%. On the call, Kress said Nvidia expects Q4 to mark the bottom of gross margins, before improving to a 72–73% range in fiscal 2028.

For a buyer, this is a useful signal. A margin dip during a ramp reflects early yields, new memory costs and new system configurations, not price cuts. Nvidia is absorbing ramp costs rather than passing on discounts. If you expected Rubin list pricing to soften because margins are falling, the guidance says the opposite: margins recover next year while volume grows.

Deeper Analysis: Supply Lock-Up, Pricing Power, Power Limits and Guidance

This section moves from the reported numbers to what they imply for capacity planning. Where we derive a number, we show the arithmetic. Where we use an assumption, we label it illustrative.

A $390 billion supply book is a moat made of contracts

Nvidia’s total supply and capacity commitments rose to $390 billion from $279 billion, an increase of $111 billion in one quarter. The CFO commentary attributes the increase primarily to multi-year memory supply procurement. The schedule matters more than the total: $69 billion falls in the remainder of fiscal 2027, $203 billion in fiscal 2028, $92 billion in fiscal 2029, and the rest later.

Vera Rubin ramp supply chain from memory, wafers and packaging through Nvidia commitments to hyperscalers and neoclouds

Figure 2: The supply path behind the Vera Rubin ramp, from component commitments to rack integrators and end buyers, with power as a parallel constraint.

Figure 2 maps that supply chain. Memory suppliers, foundry wafers and advanced packaging and substrates feed Nvidia’s $390 billion commitment book. Committed parts flow into $39.7 billion of inventory, then to rack integrators and original design manufacturers (ODMs), and finally to hyperscalers, neoclouds and enterprise or sovereign buyers. Power, land and shell sit beside the chain as an independent constraint on when racks can be energized.

Why does this matter to a buyer? HBM, the stacked high-bandwidth memory on every Rubin GPU, is the scarcest input in the system. Each Rubin GPU carries 288 GB of HBM4, per Nvidia’s preliminary specs, and an NVL72 rack carries 20.7 TB. When the largest buyer of HBM signs multi-year commitments at this scale, it reserves capacity that competing accelerators would otherwise bid for. The memory side of that dynamic is covered in our analysis of the HBM memory supercycle.

Management also named capacity agreements with Kinsus and Samsung Electro-Mechanics for substrates and SPIL for advanced packaging, adding to earlier partners Amkor, IBIDEN, KYEC and Unimicron. Kress said on the call: “We’ve secured memory. We’ve secured advanced packaging.” The inventory build, up $8.1 billion in a quarter, is the physical evidence. This is supply being pre-positioned for Rubin, not unsold product.

The implication is uncomfortable for alternative-silicon strategies. A team that plans to switch part of its fleet to another accelerator in 2027 should ask that vendor a direct question: how much HBM4 have you contracted for our delivery window? The answer often shapes delivery dates more than the chip’s benchmark.

Pricing power: the evidence is in the rental market

The most practical numbers on the call were not about Nvidia’s own prices. They were about what its customers charge. Management said Hopper GPUs have re-contracted at up to twice their original price under multi-year agreements. It also said B300 rental rates have risen more than 60% since introduction. At the same time, it said token prices have fallen nearly 60% since May.

Those three statements fit together through one mechanism: falling cost per token drives more usage, which keeps older GPUs fully booked. Management said its compute “remains fully utilized across every hyperscaler and neocloud we serve.” When two-year-old silicon re-prices upward, the usual depreciation curve for GPU capacity breaks. Buyers who budgeted for Hopper and early Blackwell hourly rates to fall by half within a year should revisit that assumption.

These are management statements, not audited market data. They likely describe specific contracts, and spot prices vary widely by provider and term. For independent per-hour price points, see our November 2026 cloud GPU pricing survey across hyperscalers and neoclouds. The direction, though, is consistent with a 74.8% gross margin and a $130 billion guide.

Networking attach: what you are really buying

The network share of spending is the number infrastructure teams most often underestimate. Using the reported numbers, networking was at least 19.5% of Nvidia’s Data Center revenue in Q3. That is Nvidia’s share alone; it excludes third-party optics, cabling, structured fiber and storage networks.

Vera Rubin NVL72 networking layers from NVLink scale-up through ConnectX-9 and Spectrum-X or InfiniBand scale-out

Figure 3: The networking layers that make up Nvidia’s networking revenue in a Vera Rubin deployment, from in-rack NVLink to scale-out fabric and storage.

Figure 3 shows where that revenue comes from. Inside the rack, 72 Rubin GPUs and 36 Vera CPUs connect through NVLink 6 switch trays at 3.6 TB/s per GPU. Each GPU also has a ConnectX-9 SuperNIC at 1.6 Tb/s for scale-out traffic. Those NICs connect to Spectrum-X Ethernet or Quantum InfiniBand leaf switches, then to spines that can use co-packaged optics. BlueField-4 DPUs handle storage and north-south traffic.

The practical lesson is that the scale-up layer is fixed once you choose NVL72. The real procurement decision is the scale-out fabric. Spectrum-X is Nvidia’s Ethernet platform and competes directly with Broadcom-based and Arista or Cisco fabrics. InfiniBand remains Nvidia’s lowest-friction option for large training jobs. The trade-offs are covered in our Spectrum-X AI Ethernet fabric deep dive and our RoCE v2 vs InfiniBand vs Ultra Ethernet comparison.

A worked example helps. Suppose a buyer plans a cluster and budgets 10% of hardware cost for networking, which is common in older CPU-era plans. If Nvidia’s own mix is near 20% of data center revenue, the buyer’s real network share, after third-party optics and cabling, is likely higher still. On a $500 million deployment, the gap between a 10% and a 20% network budget is $50 million. These shares are illustrative; the actual mix depends on fabric choice, rail design and oversubscription ratio.

Power is now the binding constraint, not chips

Management flagged power and land as critical bottlenecks for data center buildout. It also cited an Australian project targeting 2 gigawatts of land, power and shell by the end of calendar 2027. On the call, it said a 1-gigawatt AI factory could generate annual revenue approaching $50 billion for its operator, and that reaching first token one quarter earlier could add more than $10 billion.

Treat the $50 billion figure as a vendor sales argument rather than an independent estimate. It still shows how Nvidia now frames the purchase: value per gigawatt and time to first token, not cost per GPU. Vera Rubin’s own pitch follows that logic. Nvidia’s product page claims up to 10x more inference tokens per megawatt and one-tenth the cost per million tokens versus GB200 NVL72, for specific reasoning workloads. On the call, management said Vera Rubin delivered 10x more tokens per watt than GB300 in MLPerf Endpoints. These are vendor projections and vendor-run benchmarks; wait for independent MLPerf results on your model class before treating them as planning figures.

A simple illustrative model shows why power dominates. Assume a Vera Rubin NVL72 rack draws on the order of 200 kW at the facility level including cooling overhead; this is an assumption for illustration, not a published spec. A 100 MW hall then fits about 500 racks, or 36,000 GPUs. If a power upgrade slips two quarters, those racks sit idle or never ship to you. Power readiness, not GPU allocation, is the gating item for most 2027 projects. Cooling design matters as much, as covered in our liquid cooling guide for AI data centers, and power delivery is shifting too, as explained in our 800V DC AI data center power architecture analysis.

Reading the Q4 FY2027 guidance correctly

Nvidia guided Q4 fiscal 2027 revenue to $130 billion plus or minus 3%, a range of about $126.1 billion to $133.9 billion. The headline sequential growth is 12.2%. However, the release notes that Q4 includes an additional week, making it a 14-week quarter.

A rough normalization helps. Scaling $130 billion by 13/14 gives about $120.7 billion on a 13-week basis, which is roughly 4% above Q3. Revenue does not scale linearly with calendar days, because shipments are lumpy, so treat that as illustrative. The point is that the extra week flatters the sequential comparison. Buyers reading “12% growth” as “supply loosening” should look at the normalized number, which suggests a business still constrained by supply.

The guide also assumes no Data Center compute revenue from China. Management said the company is no longer shipping licensed Hopper data center products to China. That removes about $2.4 billion of Q3 revenue from the Q4 base. The rest of the guide therefore implies that non-China demand grows faster than the headline suggests.

Adding the year together gives a derived fiscal 2027 total. Q1 at $81.6 billion, Q2 at $96.2 billion, Q3 at $115.8 billion and a $130 billion Q4 midpoint sum to about $423.6 billion. Management said it expects fiscal 2028 revenue to grow approximately 75%, up from a prior 70% outlook. Applied to that base, the implied fiscal 2028 figure is roughly $740 billion. That is our arithmetic, not a company number, and it depends entirely on Q4 landing at the midpoint.

Set against hyperscaler capex

The demand side can be cross-checked against cloud spending. Based on the hyperscalers’ late-October 2026 earnings, as compiled from their releases, calendar 2026 capex guidance was about $220 billion for Amazon, about $170 billion for Microsoft, $195–205 billion for Alphabet and $135–145 billion for Meta. Together that is roughly $720–740 billion, and each signalled higher spending in 2027. Definitions differ, since some figures include finance leases, so the totals are approximate.

Nvidia’s Q3 Data Center revenue annualizes to about $430 billion. That does not mean 58% of hyperscaler capex goes to Nvidia. Capex includes buildings, power and land, and Nvidia also sells to non-hyperscalers, which made up 44% of Data Center revenue in Q3. The comparison simply shows that Nvidia’s revenue is consistent with the capex commitments its largest customers have made. The broader economics are covered in our hyperscaler capex and AI compute economics analysis.

Decision matrix for infrastructure buyers

The numbers support different moves for different buyer types. The matrix below summarizes our reading.

Buyer type What Q3 signals Recommended move Main risk
Hyperscale or large lab Rubin ramp fast, margins at trough Secure 2027 Rubin and Rubin Ultra allocations now Overcommitting before power is ready
Neocloud Nvidia financing more selective Secure own financing before relying on Nvidia guarantees Guarantees lapse if projects slip
Enterprise on-prem Enterprise revenue to double in FY28 Start with Blackwell Ultra or cloud Rubin, plan on-prem later Long lead times, facility upgrades
Inference-heavy startup Token prices down, GPU rents up Rent capacity on multi-year terms with price caps Re-pricing at renewal
Sovereign or regulated buyer Sovereign revenue tripled, export rules tighter Clear export licensing early 90-day licensing window for some countries

Trade-offs, Gotchas, and What Goes Wrong

The first risk is governance. Nvidia’s CEO left without a stated reason, and the board said it does not intend to reinstate him. The release and the call stress continuity, and the roadmap remains Vera Rubin Ultra next year and Feynman in 2028. Still, product cadence at Nvidia has historically been driven from the top. Buyers signing multi-year plans should treat roadmap dates beyond 2027 as less certain than they were a month ago.

The second risk is neocloud credit. Nvidia disclosed $118 billion of maximum gross exposure on guarantees, including $105 billion of credit support for SB Energy’s PORTS-Pike campus in Ohio, reported at about 4.25 GW. Financing commitments to neoclouds fell to $32 billion from $36 billion. On the call, Kress said that if a project is delayed, “our guarantee doesn’t exist anymore,” as reported by TechCrunch and others. On 18 November, CoreWeave closed down 4.13% and IREN about 3.6%, while Nvidia fell 2.89%, per CNBC. If your capacity sits with a smaller provider, check how its build is financed.

Vera Rubin ramp procurement decision flow for AI infrastructure buyers

Figure 4: A procurement decision flow for the Rubin transition, covering timing, power readiness, fabric choice, alternative silicon and contract terms.

Figure 4 turns these risks into a decision path. If you need capacity within six months, reserve Blackwell Ultra or Hopper in the cloud. If not, check whether your facility is ready for Rubin’s power and cooling. If it is not, book Rubin cloud capacity first; if it is, negotiate on-prem allocation and lock the fabric design. Every path ends with a price check against alternative silicon and a contract with exit and delay clauses.

The third risk is export policy. The 16 November BIS rule adds due-diligence duties for data centers serving firms linked to countries of concern. It also requires licenses for sales to Indonesia, Malaysia, Singapore, Thailand and Brazil, with 90 days for pending deals. Regional buildouts in Southeast Asia may face delays.

Finally, comparability is weaker than it looks. The new segment structure, the extra week in Q4 and vendor-run benchmarks all make easy comparisons misleading. Read the primary documents rather than summary headlines.

Practical Recommendations

Plan for firm pricing, not relief. The combination of a $390 billion supply book, rising rental rates for older GPUs and a margin trough that management expects to reverse argues against waiting for discounts. If you need capacity in the first half of calendar 2027, the cost of waiting is likely higher than the cost of committing.

Separate the GPU decision from the fabric decision. NVL72 locks in NVLink, but scale-out is still open. Run a structured comparison of Spectrum-X, InfiniBand and third-party Ethernet on your actual job mix, collective patterns and operations skills.

Make power the first workstream. Treat energization dates as the critical path and tie GPU delivery schedules to them contractually. A rack that arrives before its power is a stranded asset.

Hedge counterparty risk. If your capacity comes from a neocloud, ask about its financing sources, any Nvidia guarantees and what happens if construction slips.

Use the transition. Rubin’s fast ramp creates a window for Blackwell Ultra capacity with shorter lead times. For inference workloads that fit on GB300, that may be the best price-to-availability point this year. The architecture is covered in our GB300 NVL72 Blackwell Ultra architecture guide.

Checklist for the next procurement cycle:

  • Confirm facility power, cooling and floor loading against Rubin rack requirements before signing.
  • Budget networking at a realistic share of total cost, not a CPU-era 10%.
  • Ask every accelerator vendor for contracted HBM supply in your delivery window.
  • Add delay, substitution and price-cap clauses to multi-year capacity contracts.
  • Review export-license exposure for any non-US site or customer.
  • Benchmark on your own models before relying on vendor tokens-per-watt claims.

Frequently Asked Questions

What were Nvidia’s Q3 fiscal 2027 revenue and earnings?

Nvidia reported revenue of $115.8 billion for the quarter ended 25 October 2026, up 103% year over year and 20% sequentially. GAAP net income was $65.3 billion and GAAP diluted earnings per share were $2.70. Non-GAAP diluted EPS was $2.68. Data Center revenue was $107.6 billion and Edge Computing revenue was $8.2 billion. The release came out on 17 November 2026, beating the company’s own August guidance of $108 billion plus or minus 2% by about $7.8 billion at the midpoint.

How much did Vera Rubin contribute in Q3 fiscal 2027?

Management said on the earnings call that the Rubin platform generated nearly $20 billion in revenue in its first production quarter, calling it the fastest product ramp in Nvidia’s history. Production shipments began in August. Since Data Center revenue grew $18.6 billion sequentially, Rubin accounted for almost all of the quarter’s growth. Nvidia also said hyperscalers brought Vera Rubin into production in about one month, half the time required for Grace Blackwell. Vera Rubin NVL72 is generally available on Oracle Cloud Infrastructure, and Google Cloud’s A5X instances are generally available.

How big is Nvidia’s networking business now?

Nvidia said networking revenue exceeded $21 billion in Q3 fiscal 2027. That is at least 19.5% of Data Center revenue and implies an annual run rate above $84 billion. The figure covers NVLink switching, ConnectX SuperNICs, BlueField DPUs, Spectrum-X Ethernet, Quantum InfiniBand and, increasingly, co-packaged optics switches, which management said are now in volume production. Nvidia did not disclose a detailed split between those product lines, so any breakdown by protocol is an estimate rather than reported data.

What is Nvidia’s Q4 fiscal 2027 guidance?

Nvidia guided Q4 fiscal 2027 revenue to $130 billion plus or minus 3%, with GAAP and non-GAAP gross margins of 71% to 72%. GAAP operating expenses are expected at about $11.0 billion. The quarter includes an additional week, so sequential growth looks higher than the underlying trend. The outlook assumes no Data Center compute revenue from China. Management said it expects Q4 to be the gross-margin trough, with margins improving to 72–73% in fiscal 2028 and revenue growing about 75% that year.

Why did Nvidia stock fall after strong Q3 earnings?

On 18 November 2026, Nvidia closed down 2.89%, per CNBC, despite a revenue beat and higher guidance. Coverage pointed to several factors: the board’s statement that Jensen Huang will not return as CEO, the guided gross-margin dip, new export controls, and Nvidia’s more selective stance on financing neoclouds. CoreWeave and IREN also fell. Market moves have many causes and this article does not offer investment advice; the infrastructure-relevant signal is the tighter financing stance toward smaller GPU clouds.

Should infrastructure buyers wait for GPU prices to fall?

The Q3 evidence argues against waiting. Management said Hopper GPUs re-contracted at up to twice their original price, B300 rental rates rose more than 60% since introduction, and compute is fully utilized across the clouds it serves. Nvidia also expanded supply commitments to $390 billion, mostly memory, and expects margins to recover in fiscal 2028. Prices may ease for specific older SKUs, but buyers who need capacity in the first half of 2027 should plan for firm pricing and secure terms now.

Further Reading

By Riju — about

Comments

No comments yet. Why don’t you start the discussion?

Leave a Reply

Your email address will not be published. Required fields are marked *