Nvidia’s second quarter of fiscal 2027 delivered $96.2 billion in revenue and sharpened one message for enterprise technology buyers: capacity remains tighter than demand. For CTOs, infrastructure leaders, and AI platform owners, the issue raised by the Aug. 27, 2026 earnings coverage is not whether enterprises want more GPUs, but how long supply constraints will shape budgets and stack choices — and what memory inflation does to 2027 hardware costs.
What changed
CIO Dive reported that Nvidia generated $96.2 billion in revenue in the second quarter of fiscal 2027, more than doubling year over year, with data center revenue reaching $89 billion, up 117%. A SiliconANGLE guest column by ZK Research principal analyst Zeus Kerravala put the third-quarter guide at $108 billion.
On the duration of the squeeze, CIO Dive reported that Nvidia expects the memory bottleneck to remain through the end of fiscal year 2028. Kerravala’s column quotes CEO Jensen Huang saying on the call that supply constraints will persist “at least through the end of fiscal year ’28,” and that when asked to rank the bottlenecks he declined to name one: “Our entire supply chain is challenged.”
CIO Dive reported that Huang told investors supply constraints factor into fiscal 2028 revenue projections, which are expected to grow 70% year over year. Kerravala’s column adds what matters most to buyers: CFO Colette Kress presented that 70% as a supply-constrained figure, while customer forecasts, she said, “point to our growth doubling next year.” The gap between roughly 100% demand and roughly 70% deliverable supply is the quarter’s central fact for anyone planning capacity.
Why B2B teams should care
CIO Dive reported that enterprises should not expect significant GPU price drops within the next year to two years, citing Forrester principal analyst Naveen Chhabra, and cited a Gartner projection that global semiconductor revenue will exceed $1.3 trillion in 2026, with DRAM prices up 125% and NAND flash up 243%.
Nvidia is absorbing only part of that. Per Kerravala’s column, it lowered its gross margin outlook to 74% in Q3, bottoming at 71% to 72% in Q4 and settling at 72% to 73% in fiscal 2028, and described “extreme pricing conditions in memory” with increases that “exceeded our prior expectations and are headed even higher into next year.” If the world’s largest memory buyer is passing part of that through, buyers with less leverage should assume every server, storage array, and edge device with memory costs more in 2027 — and that AI systems take the hit twice, on the accelerator and on the host.
Who is affected
Enterprise buyers compete for capacity with hyperscalers, neoclouds, and sovereign customers. CIO Dive reported that Kress named hyperscalers a major growth driver, and that Nvidia announced an expanded AWS collaboration to deploy 2 million additional GPUs across AWS infrastructure.
CIO Dive reported $49 billion from hyperscalers and $40 billion from enterprise, AI cloud, and industrial customers; Kerravala’s column reports that second bucket grew 138% year over year and is expected to reach roughly half of the data center business. That is the same capacity-lock-in dynamic visible in large pre-committed compute deals.
The quarter also touches networking and CPU buyers, not only accelerator buyers. Per the SiliconANGLE column, networking revenue rose 18% sequentially, Spectrum-X Ethernet grew 2.6x year over year, the Grace CPU surpassed $5 billion on a trailing-12-month basis, and Nvidia expects CPU revenue to more than double in fiscal 2028.
What teams should check now
- Extend the planning horizon. Test capacity plans, reservation strategy, and deployment timing against management’s statement that constraints last through fiscal 2028; Huang noted that land, power, and shell capacity are “often two to three years out.” Contract certainty is now worth more than a few points of discount.
- Rework finance assumptions that GPU or AI server pricing will ease soon; Forrester’s one-to-two-year window is the planning baseline.
- Build memory inflation into 2027 capital plans, not just AI line items. The DRAM and NAND increases Gartner projects hit general-purpose refresh cycles too, so right-sizing memory footprints on existing fleets has direct budget value.
- Reassess architecture scope. Kress said Nvidia is “not just selling the best chips” but “a full-stack AI factory platform.” Check whether budgets cover only accelerators or also networking and CPU, and benchmark those on your own workloads.
- Re-baseline capacity models for agents. Huang said an agent consumes “probably 15 to 100 times” the compute of a human using the same model, and that agentic usage crossed over human-prompted usage last month. Add per-agent budgets and token-level chargeback.
- Rank workloads by business value before chasing capacity. CIO Dive attributed to Chhabra the view that CIOs should focus on measurable business value from AI, not more GPUs — which depends on the cost visibility many AI programs still lack.
What remains unclear
- Not yet confirmed: enterprise pricing for GPUs, AI servers, or related platform components by customer segment.
- Not yet confirmed: how Nvidia handles allocation across hyperscalers, enterprise accounts, AI cloud providers, and industrial customers.
- Not yet confirmed: which enterprise subsegments inside the reported $40 billion bucket show the strongest or weakest demand.
- Not yet confirmed: how much of the memory cost increase Nvidia passes into system pricing versus absorbs; the gross margin path signals direction, not per-configuration impact.
- Not yet confirmed: the reported $12.9 billion Nvidia acquisition of Hugging Face, which Kerravala’s column attributes to The Information. Nvidia did not announce it on the earnings call.
What to watch next
Management commentary on the Vera Rubin ramp will matter more to enterprise buyers than share-price reaction. Kerravala’s column reports Vera is in full production, with shipments to Oracle Cloud Infrastructure and, this quarter, to AWS. The variables to track are supply updates through fiscal 2028, memory pricing, the gross margin trough, and whether customer spending continues to support the fiscal 2028 outlook. One number frames the rest: the column cites Nvidia’s revenue opportunity per gigawatt of data center capacity rising from roughly $18 billion with Hopper to about $40 billion with Vera Rubin. Read from the buyer’s side, that is the cost of a megawatt of AI capacity climbing with each generation.
