Samsung Puts $1B Behind KKR's Plan to Build AI's Power Grid
Samsung Puts $1B Behind KKR's Plan to Build AI's Power Grid
Samsung and five affiliates committed $1 billion on Tuesday to Helix Digital, the KKR-built AI infrastructure company led by ex-AWS chief Adam Selipsky. With Nvidia, Kuwait's sovereign wealth fund, and US power giant Vistra already aboard, Helix is becoming the industry's biggest bet that the AI race will be won on electricity, not just chips.
Published September 29, 2026 · By Daniel Rao
![]()
Illustrative: an aerial view of a hyperscale data center campus in Utah. Helix Digital, launched by KKR in June, has not yet disclosed the sites of its first campuses. Image: حمزة مستقيم / CC BY-SA 4.0.
Samsung just made its biggest-ever wager on the AI infrastructure boom — and it isn't buying chips. The Korean conglomerate announced Tuesday that it will invest $1 billion in Helix Digital Infrastructure, the AI infrastructure company launched in June by private-equity giant KKR and led by former Amazon Web Services CEO Adam Selipsky.
The commitment, reported by Reuters on Tuesday, builds on more than $10 billion of capital already committed to Helix's strategy. That makes the combined war chest one of the largest purpose-built AI infrastructure vehicles ever assembled by private equity — and it signals that the industry's power brokers have decided the binding constraint on AI's future is no longer the GPU. It is the grid.
What makes Samsung's check unusual is who is signing it. The $1 billion is not coming from Samsung Electronics alone. Samsung Electronics is contributing $500 million through its capital fund, and five affiliates — Samsung C&T, Samsung SDS, Samsung SDI, Samsung Life Insurance, and Samsung Fire & Marine Insurance — are covering the rest. Six subsidiaries, one check, and a strategy that reads as much like supply-chain positioning as it does portfolio investing.
What Helix actually is
Helix Digital debuted on June 11, 2026, with more than $10 billion in committed capital and a founding-investor roster that reads like a roll call of the AI buildout's heaviest hitters: KKR, the Kuwait Investment Authority, Nvidia, and Vistra, the US power generation company with a portfolio of roughly 50 gigawatts across 18 states. The company is led by Adam Selipsky — who ran AWS through its period of highest revenue growth before departing in 2023 — as CEO and chair, with Waldemar Szlezak from KKR's infrastructure team as chief investment officer, and roughly 170 dedicated professionals from KKR's global infrastructure business behind it.
The business model is deliberately end-to-end. Helix develops and operates hyperscale data centers, but it also invests directly in power generation — both baseload and flexible sources — plus transmission and distribution networks and fiber-optic connectivity. The point is to sell hyperscalers an integrated package: compute capacity with the electricity already attached.
"Large users of digital infrastructure have an urgent need to reduce complexity and unlock new capacity," Selipsky said at the June launch. "Helix combines significant long-term capital with the capabilities and expertise to deliver holistic AI infrastructure solutions with speed and scale." Vistra will serve as the preferred power provider for Helix investments, while Nvidia contributes both anchor capital and a strategic technology partnership around its DSX AI factory-aligned infrastructure — aiming, in Nvidia's phrasing, for "lowest total cost of ownership and accelerating time to first token."
The power bottleneck everyone keeps naming
Helix exists because of a bottleneck that has become the AI industry's open secret: the grid cannot keep up. Data centers that once drew tens of megawatts now draw hundreds, and utility interconnection queues in the United States stretch for years. Projects stall not for lack of GPUs or land, but for lack of electrons. Hyperscaler capital expenditure — Microsoft, Amazon, Alphabet, and Meta combined — was disclosed at roughly $700 billion for 2026 earlier this year, per Bloomberg. A fraction of that spend is worthless if the power doesn't arrive.
The market has responded by inventing a new asset class. So-called "neoclouds" like CoreWeave and Nebius rent out AI compute as a service; power-first developers like Crusoe build data centers directly at the source of cheap energy. Helix is the private-equity version of that idea at maximum scale: instead of building one project at a time, assemble the capital, the power partnerships, and the chipmaker relationships up front, and offer hyperscalers a shortcut around the interconnection queue.
KKR has been assembling this playbook for years. The firm has invested more than $75 billion globally across digital infrastructure and power, including roughly $9 billion in Korea since 2009. Helix is the financialized, institutional version of that idea — Wall Street buying the power problem wholesale rather than solving it project by project.
Why Samsung is spreading the check across six companies
The most interesting detail of Tuesday's announcement is not the $1 billion figure but its plumbing. Samsung Electronics covers half; Samsung C&T (construction and engineering), Samsung SDS (IT services), Samsung SDI (batteries), and two insurance arms split the rest. That distribution is not about diversification. It is about downstream contracts.
Samsung is the world's largest memory-chip maker, and every hyperscale data center that breaks ground is an order book: high-bandwidth memory for the accelerators, solid-state drives for storage, and — through SDI — battery energy storage for the power side. Samsung C&T builds data centers; Samsung SDS can sell the IT-services layer on top. By putting its construction, battery, and services arms directly into the deal, Samsung is positioning itself not just as an investor in Helix's buildout but as a vendor to it.
There is a timing dimension, too. Samsung has been fighting to qualify its HBM4 memory with Nvidia — supply is expanding, and Q2 guidance pointed to sequential improvement, but the qualification has been slower than projected. A direct pipeline into the industry's fastest-growing data-center buildout is worth more than any single purchase order.
Related: Nvidia's $150B Buyback: the Biggest in US Corporate History — our report on how the AI chip giant is converting the infrastructure boom into shareholder returns.
Nvidia's quiet role: from chip vendor to ecosystem architect
Note who keeps showing up in these deals. Nvidia is not just selling GPUs to Helix; it is a founding investor and strategic partner, shaping how Helix's campuses get built — "maximizing tokens per watt," in the company's words. This is the same pattern visible across the industry: Nvidia's Open Agent Safety Platform for AI security, its deep partnerships with neoclouds, and now a seat inside the private-equity vehicle designed to build the power plants behind the data centers.
The chipmaker's incentives are straightforward. Every megawatt of Helix capacity is future GPU demand, and every bottleneck Helix clears is one less reason for a hyperscaler to diversify away from Nvidia's stack. Jensen Huang has spent 2026 arguing that AI risk should be solved through engineering rather than slowed down — and then selling the engineering to everyone involved. The Helix partnership is the infrastructure version of that play.
The US angle: onshoring the AI factory
For a US audience, the geography matters. Helix is a US-based company, its preferred power partner operates across 18 US states, and the model it is scaling — compute co-located with secured domestic power — is the physical counterpart to the policy push reshoring AI infrastructure to American soil. The company has not yet disclosed any sites for its data centers, which is the obvious next shoe to drop: the first announced campus will tell us where KKR, Nvidia, and now Samsung believe the US grid still has room to grow.
Compare the alternatives. Oracle's $18 billion New Mexico campus has hit force-majeure territory. Crusoe is building Google's Texas campus at the source of West Texas wind. Helix is the financialized, institutional version — Wall Street buying the power problem wholesale rather than solving it project by project.
Our analysis: follow the affiliates
What follows is analysis. The headline is "$1 billion," but the story is the cap table. When a conglomerate splits a check across six subsidiaries — including its construction company, its battery maker, and its IT-services arm — it is not making a passive investment. It is buying a customer. Samsung's affiliates each map to a line item in a hyperscale data center's bill of materials, and Helix's buildout, whenever the first sites are announced, will now carry a built-in Samsung procurement pipeline.
The deeper read is about where value accrues in the AI stack. Two years ago the money chased models; last year it chased chips. The Helix raise — and Samsung's decision to chase power and concrete alongside it — suggests the industry's center of gravity has moved downstream to the two inputs that cannot be synthesized: land with power attached, and the memory that fills the racks. Chips get the headlines. Electrons get the term sheet.
Watch for two things. First, Helix's first site announcements — the campuses will reveal whether the power-first model can actually beat the interconnection queue, which is the entire thesis. Second, whether Samsung's affiliates start disclosing Helix-related orders in earnings calls. If the HBM qualification with Nvidia lands and the Helix pipeline converts, Tuesday's $1 billion will look less like infrastructure investing and more like the most expensive sales meeting in Samsung's history.
Sources: Reuters (September 29, 2026); Samsung Global Newsroom (September 29, 2026); Data Center Dynamics (Helix launch, June 2026); KED Global (affiliate breakdown, September 29, 2026); Crypto Briefing and Fresh Blog News (background on Selipsky, Szlezak, founding investors, September 29, 2026); Bloomberg via GetAI Decoded (hyperscaler capex context, April 2026).
Comments
Post a Comment