A16Z Found the AI Retention Unlock. It's Not the Model — It's the Workload Match.

On September 29, 2026, Samsung Electronics and five of its affiliates — Samsung C&T, Samsung SDS, Samsung SDI, Samsung Life Insurance, and Samsung Fire & Marine Insurance — committed a combined $1 billion to Helix Digital Infrastructure, the KKR- and Nvidia-backed AI infrastructure venture launched in June 2026. The move pushes Helix's total committed capital past $11 billion and signals that the global race to build AI-native infrastructure has entered a new phase: the phase where the companies closest to the hardware supply chain — memory makers, chip manufacturers, power providers, and conglomerate builders — want a direct stake in the infrastructure layer that will define AI economics for the next decade.

This is not a passive financial bet. Samsung's investment structure — six entities, a $500 million anchor from Samsung Electronics plus $500 million distributed across manufacturing, logistics, financial, and technology affiliates — mirrors a strategic alignment rather than a fund allocation. It is Samsung saying: wherever Helix builds a data center, Samsung wants to be the memory supplier, the logistics partner, the construction arm, the energy storage provider, and the financing backstop.

For enterprise AI strategy teams, the Samsung-Helix deal is a signal event. The infrastructure layer beneath AI is now controlled by a small number of deeply capitalized players with interlocking interests, and the availability and pricing of AI compute is going to be shaped by those relationships over the next three to five years.

What Helix Digital Infrastructure Actually Is

Helix is not a traditional data center operator. It was founded in June 2026 by KKR to solve a problem that every hyperscaler has been raising for two years: AI infrastructure is not a software problem. It is a physical infrastructure problem at a scale that no single entity can finance, permit, and build fast enough.

Hyperscale data centers for AI inference and training require four things that are genuinely hard to assemble simultaneously:

• Power: at gigawatt scale, requiring partnerships with utilities, grid operators, and increasingly private power generation
• Land: in locations with power access, water for cooling, fiber connectivity, and regulatory stability
• Connectivity: fiber-optic networks that can move model weights and inference traffic at the latency requirements AI workloads impose
• Capital: the full-stack build-out for a hyperscale AI campus costs $2–5 billion before the first GPU rack is installed

No single company does all four well. KKR's thesis with Helix was to assemble the capital, the operating expertise, and the strategic partnerships to do all four simultaneously — at the speed hyperscalers need to maintain their competitive positions.

The founding investor base reflects this thesis directly. KKR brought capital markets expertise and project finance capability. Nvidia brought GPU supply certainty and architectural guidance as a cornerstone strategic partner. The Kuwait Investment Authority (KIA) brought sovereign capital willing to take long-duration infrastructure risk. Vistra, as preferred power partner, brought the utility-scale electricity generation relationships that are the hardest constraint in AI infrastructure buildout.

Why Samsung Deployed Six Companies, Not One

The most telling detail in the Samsung-Helix announcement is not the $1 billion commitment. It is the structure of how Samsung committed it.

Samsung Electronics invested $500 million — the anchor and the headline figure. But the remaining $500 million came from five distinct Samsung affiliates, each representing a different piece of the value chain:

• Samsung C&T: Samsung's construction and trading arm — the entity that builds large-scale physical infrastructure projects worldwide
• Samsung SDS: Samsung's IT services and logistics unit — the entity that designs, operates, and manages data centers domestically
• Samsung SDI: Samsung's battery and energy storage division — the entity that supplies energy storage systems now essential to data center power resilience
• Samsung Life Insurance and Samsung Fire & Marine Insurance: long-duration capital — entities capable of financing infrastructure at the 15–20 year investment horizon large AI campuses require

This is not a financial investment. This is Samsung locking in preferred position across the Helix supply chain. When Helix builds a 500-megawatt AI campus in the United States or Southeast Asia, Samsung wants to be the memory supplier, the construction contractor, the operations partner, the energy storage provider, and the financing backstop simultaneously.

The investment structure converts a $1 billion check into a multi-decade supply chain relationship. It is the kind of strategic move that KKR's team will market to future data center customers: their infrastructure carries guaranteed supply chain support from one of the world's largest electronics conglomerates. For hyperscalers comparing AI campus build-out proposals, that supply chain guarantee is worth more than any single discount on construction cost.

Adam Selipsky's AWS Playbook, Applied One Layer Down

When Adam Selipsky was CEO of Amazon Web Services, the service grew from a collection of compute primitives to the world's most valuable enterprise infrastructure business by applying a consistent playbook: identify the constrained workload, build the reliability and scale guarantee that enterprises require, price at a level that makes the build-versus-buy equation obvious, and compound the moat through operational data and expertise.

Selipsky's Helix thesis appears to follow the same structure, applied one layer down the stack.

1. Identify the constrained workload. The constrained workload is AI infrastructure at hyperscaler scale — the integrated data center, power, and fiber combination that hyperscalers need to build but cannot build fast enough on their own. Every major cloud provider has acknowledged a multi-year supply gap between the AI compute capacity they need to deploy and the infrastructure they can realistically build with internal resources and standard construction timelines.

2. Build the reliability guarantee. Helix's investor base — KKR, KIA, Nvidia, Vistra, Samsung — functions as the reliability guarantee. No hyperscaler signing a 10-year data center contract with Helix faces meaningful counterparty risk when the counterparty is backed by those entities. The investor list is itself a product feature.

3. Price at the build-versus-buy inflection point. Large language model training and inference infrastructure is capital-intensive enough that even the largest hyperscalers — Microsoft, Google, Amazon — find it economically rational to outsource portions of their AI campus buildout to specialized operators. Helix's pricing model is designed to sit just below the internal cost of hyperscaler self-build, creating a clear economic case for outsourcing.

4. Compound the moat through operational data. Every Helix campus generates operational data — power efficiency, cooling performance, hardware failure rates, network utilization patterns — that makes future campuses cheaper and faster to build and operate. That operational data moat compounds over time in ways that a hyperscaler building one or two campuses per year cannot replicate.

The Power and Connectivity Problem Nobody Talks About

The public narrative around AI infrastructure is dominated by GPU availability. GPUs are visible, countable, and tied to Nvidia's supply dynamics in ways that produce regular headlines. But ask any hyperscaler chief infrastructure officer what actually constrains AI compute deployment in 2026 and the answer is not GPUs. It is power.

Training a frontier AI model at scale requires between 50 and 150 megawatts of power during peak compute runs. Inference infrastructure for a production AI product with millions of daily active users requires comparable power continuously — not just during training. And power at that scale, in the geographies where land and cooling costs are manageable, does not materialize quickly. Utility-scale power connections take three to seven years to permit, fund, and build in the United States under current regulatory frameworks.

This is the constraint that makes Vistra's role as Helix's preferred power partner so strategically important. Vistra is one of the largest power generators in the United States, with a portfolio spanning natural gas, nuclear, and renewable generation. Their existing relationships with grid operators, their generation assets in power-stressed markets like Texas and Illinois, and their ability to co-locate generation with data center campuses gives Helix a power sourcing advantage that most infrastructure developers do not have.

Fiber connectivity operates on a similar constraint timeline. Moving AI model weights — which can exceed 100 gigabytes for frontier models — and inference traffic at the latency requirements production AI products impose requires fiber infrastructure that does not exist in many geographies where land and power costs make AI campuses economically attractive. Helix's integrated approach — building or leasing fiber alongside power and data center — addresses this constraint at a system level.

As Signal documented in Anthropic's $11.6 billion Akamai edge infrastructure deal, the edge inference problem is driving a new generation of infrastructure investment decisions. The question is not just how much compute exists globally, but where it is located relative to the users and applications that need it. Helix's full-stack approach is designed to answer that question at every layer simultaneously.

The Infrastructure Value Chain: Where Margin Is Moving

The Samsung-Helix deal is best understood as part of a broader realignment of who captures value in the AI infrastructure stack. The value chain runs from raw materials to model output, and the margin distribution along that chain is shifting measurably.

The implication for enterprise buyers is significant: the margins in AI are migrating toward the physical infrastructure layers — power, land, fiber, data center operations — and away from the compute abstraction and model layers where the pricing war is most intense. The entities assembling positions in the physical layers now are doing so at a moment when the margin structure still favors them. That window will narrow as infrastructure capacity catches up with AI demand, but the supply gap in 2026 is measured in years, not quarters.

Signal's analysis of Nvidia's $12.9 billion Hugging Face acquisition documented the same dynamic one level up: vertical integration across the AI stack — from hardware to model distribution — is the defining strategic move of the current era. Helix, backed by the semiconductor giant that controls GPU supply and now the conglomerate that controls memory and construction, is executing a comparable vertical integration play at the infrastructure layer.

What Enterprise AI Procurement Teams Need to Do Right Now

For enterprise teams evaluating AI infrastructure strategy, the Samsung-Helix deal has three direct implications that are actionable in the next 90 days.

First, cloud provider infrastructure dependencies matter more than pricing sheets. The price floor for AI compute over the next three to five years will be set by the entities that control power, land, and fiber — not by the model providers or cloud hyperscalers alone. When evaluating long-term cloud AI commitments, enterprise procurement teams should ask which infrastructure players their cloud provider depends on, and whether those players have the supply chain depth to deliver on multi-year capacity commitments.

Second, geographic AI compute positioning needs to happen before it becomes a constraint. Helix is in active development for AI campus projects across North America and, with Samsung's involvement, likely across Southeast Asia and South Korea. The locations of those campuses will determine the latency and data sovereignty profile of AI inference for the next decade. Enterprise teams with regulatory requirements around data residency or latency-sensitive AI applications should track Helix's campus pipeline and factor it into their cloud provider selection criteria.

Third, the build-versus-buy equation for private AI infrastructure is shifting. As Signal's analysis of enterprise AI agent sprawl documented, 81% of CIOs lack full oversight of their AI agent deployments — partly because those deployments run across a patchwork of public cloud regions with inconsistent performance and cost characteristics. The emergence of dedicated AI infrastructure players like Helix creates a new option: long-term dedicated capacity agreements with infrastructure-layer providers that offer predictable performance and cost profiles.

The Samsung-Helix deal is one data point in a structural shift. The physical layer of AI infrastructure is being consolidated around a small number of deeply capitalized, vertically integrated players. The enterprise teams that understand that consolidation early — and factor it into their procurement strategy — will have better leverage and better outcomes than those who treat AI infrastructure as an undifferentiated cloud bill.

Takeaway: Samsung's $1 billion investment in Helix is a supply chain integration play, not a financial bet. It positions Samsung across construction, operations, energy storage, and long-duration financing in the AI infrastructure stack — ensuring preferred partner status wherever Helix builds. For enterprise AI strategy teams, the key insight is that the binding constraints on AI compute over the next three to five years are physical — power, land, fiber — and the players who control those layers are assembling their positions now. Helix, with $11+ billion in committed capital and Adam Selipsky's AWS playbook, is one of the entities that will define AI infrastructure economics for the decade. Track the campus pipeline. Factor it into your cloud procurement. The infrastructure layer is where AI's structural economics are being determined.