NVIDIA's Market Segmentation and CPU Strategy for 2027 Q1

Segmentation change: what drove it and the underlying philosophy

Management said the segmentation change was primarily motivated by the business having “evolved and grown to such a large scale” that segmenting it helps “have a better understanding of how our business works.” ¹² The philosophy is also explicitly tied to reflecting “our current and future growth drivers” in a “new reporting framework.” ³

In the new framework, NVIDIA organizes the market into two market platforms—data center and edge computing—and within data center it reports two submarkets: Hyperscale and ACIE (which management describes as incorporating AI clouds, industrial, and enterprise). ³

Management further rationalizes the factoring by noting that AI is diverse in terms of applications and governing environment—e.g., where it runs (hyperscale cloud vs AI natives vs enterprise/industrial/on-prem vs edge) and how it is governed (public cloud vs enterprise/on-prem vs sovereign constraints). ⁴ The split is intended to produce “the simplest factoring of it” even though the real business is “more complex than the 2 large parts.” ⁵⁶

---

Rationale for the two data-center submarkets (Hyperscale vs ACIE)

1) Hyperscale: where growth is tightly linked to very large public-cloud compute demand

Management describes Hyperscale as including revenue from public cloud and the world’s largest consumer Internet companies, and it corresponds to the dominant “few” hyperscalers. ³⁷

Operationally, Hyperscale is characterized as the “easiest go to market” because it is concentrated (NVIDIA calls out “5 or 6” hyperscalers). ⁷ It is also described as having compute-centric economics where “Compute is revenues” and “Compute is profit,” and Hyperscale CapEx is characterized as “trillion dollars” and expected to grow. ⁶

2) ACIE: AI-native/industrial/enterprise/sovreign “diverse data centers” with different constraints and integration needs

Management argues ACIE addresses “growth opportunities in diverse AI purpose built data centers, and AI factories across industries and countries.” ³ It includes “AI natives,” enterprise on-prem, industrial on-prem, and sovereign AI. ⁷³

The rationale is that these environments often cannot be run the same way as hyperscale public cloud due to governance, security, and deployment constraints—management lists examples such as regulatory reasons, confidential computing, and national security, and it states “different data centers have to be built differently.” ⁵

Management also emphasizes the “stack” and integration element: it says NVIDIA is unique because it “builds all of the technology components” in an end-to-end, codesigned way, but it also “open[s] the platform” so it can be integrated across environments—even though some customers prefer a complete solution because they “would like to buy it and operate it.” ⁵

Finally, management stresses that ACIE is far more fragmented: instead of hyperscaler concentration, it is described as involving hundreds/thousands of companies (and potentially hundreds of thousands over time), with many smaller installations. ⁸

---

Competitive differences: hyperscale vs AI-native/edge (how they compete and what’s “different”)

Competitive difference 1: Hyperscale demands scale and fast provisioning, but ACIE demands integrated, rapid “time-to-token” and assembly constraints

For AI natives/ACIE-like builders, management says they do not build or design their own chips, and they also “cannot really assemble … unrelated parts together into an AI factory.” ⁹ They have “very low” patience for “time to first token,” and they require an “architecture that has a great deal of offtake so that it runs every model” with “customers from everywhere.” ⁹

Management then positions NVIDIA’s advantage as:

• offering “every component” (and otherwise relying on partners) so the system is “fully integrated” and “works together,” ⁹
• being “the most rentable of any computing platform” and emphasizing properties like “most performant,” “easiest to put together,” “best TCO,” and “easiest to finance.” ⁹

By contrast, hyperscalers are portrayed as more straightforward to engage because they are fewer, with mature data-center capability and a direct compute roadmap. ⁷⁶

Competitive difference 2: Partner ecosystem and breadth of frontier-model coverage matters differently by segment

Management explicitly claims broad hyperscale presence (“in every hyperscale cloud”) supporting “core data processing and machine learning workloads,” plus demand for NVIDIA users in public cloud services. ¹⁰ It also claims platform breadth (“only platform that runs every frontier AI model”) attributing this to the addition of Anthropic to existing partners. ¹⁰

When management discusses inference share gains, it links share growth to hyperscale inference demand and the growth in “frontier model” companies and to securing additional capacity for Anthropic across clouds. ¹¹ While the excerpts do not quantify ACIE inference share separately, they do state that the second category is fragmented and “requires … a fairly … integrated platform solution,” reinforcing that competitive advantage in ACIE is tied to integration and turnkey functionality. ¹¹

Competitive difference 3: AI-native/edge environments require responsiveness and on-site deployment, limiting “cloud-like” deployment

For physical AI / “edge” contexts within the overall framework, management states that in many industrial settings “there is no choice but to put the computer where the context is, where the action is,” and it cites latency/reliability needs (“respond … reliably quickly, every single time”) making remote cloud impractical for certain industrial processes. ⁸

Even though this statement is in the context of the “second category” (ACIE) and the broader framework, it supports the idea that competitive differences are driven by deployment constraints: hyperscalers can be served through large standardized cloud capacity, whereas AI-native/edge/industrial require localized compute readiness and integrated stacks. ⁸⁵

---

How management’s CPU trajectory affects both segments going forward

What the CPU trajectory is (Vera and CPU purpose-built for agentic AI)

Management highlights a “surprising CPU number” in connection with agentic AI needs and positions it as a new opportunity: Vera, described as “the world’s first CPU purpose built for agentic AI,” and it says Vera “opens a brand new $200 billion TAM.” ¹⁰

It also provides concrete performance/positioning claims for Vera:

• “up to 1.5x faster performance per core,” ¹²
• “2x performance per watt,” ¹²
• “4x density per rack compared to x86 based alternatives,” ¹²
• built “on custom arm cores and codesigned end to end with Rubin GPUs and NVLink.” ¹²

Segment impact: Hyperscale

Management indicates the CPU opportunity is directly tied to hyperscale inference/training infrastructure economics: it states “agentic AI and reinforcement learning represents new growth opportunities for CPUs,” and then links Vera’s arrival to meeting an inflection. ¹² It further says it has “visibility to nearly $20 billion in total CPU revenue this year,” and it explicitly frames that in the context of becoming “the world leading CPU supplier.” ¹²

Additionally, management claims “every major hyperscaler and system maker is partnering with us to get it deployed,” and it gives a production timeline: “commence production shipments of VeraRubin in the second half of this year starting in Q3.” ¹² That implies hyperscalers are an immediate adoption vector for CPU alongside GPU roadmaps because they control large-scale deployment and the “compute is revenue” paradigm. ⁶¹²

Segment impact: ACIE (AI-native/enterprise/industrial/sovreign)

The excerpts do not provide a numeric CPU revenue split between Hyperscale and ACIE. However, the qualitative linkage is clear:

1. ACIE customers are described as needing fully integrated, turnkey systems because they cannot assemble unrelated parts into working AI factories and they have low tolerance for time-to-first-token. ⁹⁵ A purpose-built CPU that is codesigned end-to-end with NVIDIA GPUs and networking (Vera with Rubin GPUs and NVLink) would fit that integration advantage proposition. ¹²⁵

2. Management also frames ACIE as including on-prem/sovereign/industrial environments where “different data centers have to be built differently” due to regulatory/security/confidentiality constraints. ⁵ In such environments, having an end-to-end stack (compute components plus software) can be particularly important, reinforcing that CPU adoption is likely to be bundled into ACIE “buy and operate” solutions rather than assembled from disparate third-party components. ⁵⁹

3. Management explicitly says NVIDIA provides the “entire solution” that makes it “much easier” for builders in ACIE-like contexts to build AI systems. ⁷ While the excerpt doesn’t say “Vera goes to ACIE at X%,” it establishes the mechanism by which a CPU designed for agentic workloads would be demanded by ACIE builders who need integrated systems. ⁷¹²

Put differently: CPU trajectory amplifies NVIDIA’s role in “time-to-production” and system economics across both segments

Management ties CPU growth to agentic AI workload needs and positions CPU as part of the “right economic metric” for customers’ AI factories (lifetime cost, tokens per watt/dollar, utilization, uptime, time to production, software durability, asset life). ¹² Because both Hyperscale and ACIE are competing for faster deployment and profitable throughput (Hyperscale via scale compute economics; ACIE via integration and low tolerance for time-to-first-token), the CPU trajectory is intended to strengthen NVIDIA’s competitiveness in both segments. ¹²⁹⁶

---

Direct answer to the question (condensed)

• Segmentation change driver: The business has grown in scale and complexity enough that a new reporting framework is needed to better reflect “current and future growth drivers,” and management believes segmentation helps understanding. ¹²³
• Rationale for two data-center submarkets:
  • Hyperscale is concentrated, public-cloud/customer-scale dominated, and easier go-to-market; it is tied to hyperscalers’ massive CapEx and the compute economics (“compute is revenues”). ⁷³⁶
  • ACIE is fragmented across AI natives, enterprise/industrial on-prem, and sovereign deployments with different governance and deployment constraints, where integrated end-to-end stacks matter. ⁷⁵³⁸
• Competitive differences (hyperscale vs AI-native/edge):
  • AI natives/ACIE cannot assemble unrelated parts and have low tolerance for time-to-first-token, so NVIDIA’s integrated, codesigned platform approach is positioned as differentiating; hyperscale is simpler to engage due to concentration and scale. ⁹⁷
  • Industrial/edge contexts require on-site responsiveness, further emphasizing system readiness and integration rather than generic cloud-like deployment. ⁸
• CPU trajectory impact on both segments going forward:
  • NVIDIA’s Vera is described as purpose-built for agentic AI, with a claimed multi-dimensional performance/watt/density advantage and a large new CPU TAM; hyperscale partnerships and adoption are indicated by “every major hyperscaler” partnering and Q3 shipments of VeraRubin. ¹⁰¹²
  • While ACIE CPU numbers are not provided in the excerpts, the described ACIE customer requirements (integrated full-stack, inability to assemble unrelated components, and low time-to-first-token tolerance) create a strong pathway for Vera-style purpose-built CPU adoption as part of turnkey “AI factory” deployments. ⁹⁵¹²