1. NVIDIA – Korean Tech Company Partnership Announcement
• Core Source
“SK Hynix has been and will continue to be NVIDIA’s largest memory partner. This is a long-standing partnership and friendship.”
“LG is fantastic when it comes to power supply, design, and construction, and both companies are working together on the future data center architecture (AI Factory).”
“The partnership with SK Hynix and SK Telecom could bring hundreds of billions of dollars in business opportunities to Korea in the future.”
“Naver is pushing ahead with the construction of the largest AI factory by a domestic company in partnership with NVIDIA… with plans to expand from 55MW next year to 100MW in the same year and 200MW by 2028.”
“Korea is a country with both AI and manufacturing capabilities, and the combination of the two is robotics.”
• Expected Impact
The most direct benefit NVIDIA secured from this Korea visit is stable long-term supply of advanced memory. The long-term technology partnership with SK Hynix is structured not as a simple procurement contract but as a co-development arrangement covering NVIDIA’s full product line — from the Vera Rubin AI supercomputer and Vera CPU to the RTX Spark PC and Jetson Thor robotics platform. According to industry estimates cited in the source material, SK Hynix is currently responsible for an estimated 60–70% of Vera Rubin HBM4 volume, and Jensen Huang directly stated that “memory shortages will persist for years.” The agreement is interpreted as encompassing not just HBM4 but also the next-generation HBM4E (Vera Rubin Ultra, late 2027).
NVIDIA has designated Korea as a core hub for building its Physical AI ecosystem. The fact that a single visit resulted in partnerships with an exceptionally broad range of companies — Hyundai Motor, LG, Doosan, Naver, SK Telecom, Krafton, and NCSoft — reflects NVIDIA’s strategy of vertically expanding its platform influence from AI semiconductors → data centers → robotics and gaming. In particular, the intent to use Korea to demonstrate the structure in which “manufacturers build the robot hardware while NVIDIA supplies the AI brain” — through Isaac and Omniverse as robotics industry standards — is unmistakable.
The AI Factory collaboration with Naver and SKT also carries significant weight in terms of expanding the global reference base for NVIDIA’s DSX platform. Naver plans to build GW-scale AI infrastructure on an NVIDIA DSX basis, starting from 55MW next year, scaling to 200MW by 2028, and SKT is entering the DSX-based AI cloud business, expanding from Korea across Asia. Both companies have agreed to jointly target sovereign AI infrastructure markets beyond Asia-Pacific into Europe and the Middle East. From NVIDIA’s perspective, this is a structure for pioneering global expansion pathways for the DSX full-stack platform through Korean partners, going beyond mere GPU sales.
2 Google Pays SpaceX $920M Per Month to Secure AI Computing Capacity… Total Contract Value Up to $29.4 Billion
• Core Source
“Official disclosure of cloud service agreement signed with Google LLC on June 5, 2026”
“Computing provided: approximately 110,000 NVIDIA GPUs + CPUs, memory, and related components”
“$920 million per month — payments from October 2026 through June 2029”
“$920 million/month × 32 months (~June 2029) = maximum total contract value of approximately $29.4 billion”
“Anthropic signed a contract for $1.25 billion per month and Google for $920 million per month, with the total value of both contracts reaching approximately $70 billion.”
• Expected Impact
Through this contract, SpaceX has secured an AI infrastructure leasing revenue stream of an exceptional scale for a single contract. Under the disclosed terms, Google will pay SpaceX $920 million per month from October 2026 through June 2029, in exchange for access to approximately 110,000 NVIDIA GPUs, CPUs, memory, and related infrastructure. The total contract value reaches a maximum of approximately $29.4 billion.
The structural significance of this contract lies in SpaceX’s repositioning from a rocket and satellite company to an ‘AI infrastructure company and neo-cloud company’. Adding the previously announced $1.25 billion per month contract with Anthropic, SpaceX’s annual recurring revenue (ARR) from its AI infrastructure leasing business is estimated at approximately $26 billion, with the total value of both contracts reaching around $70 billion. This figure is widely regarded by the market as the core revenue basis for SpaceX’s planned $75 billion IPO.
The contract also includes risk provisions. If SpaceX fails to deliver the agreed GPU quantities by September 30, 2026, Google may, after a one-month grace period, either immediately terminate the contract or accept the capacity actually delivered and have the monthly fee reduced on a pro-rata basis. After December 31, 2026, either party may terminate the contract with 90 days’ advance notice. Intellectual property rights over AI models, content, and related data are retained by Google.
The fact that even Google — one of the world’s largest owners of AI computing with its own cloud and TPU chips — has resorted to leasing external GPU clusters is a powerful market signal that the pace of AI computing demand is outstripping even the top hyperscalers’ ability to self-supply. Before and after signing this contract, Google pursued what was its first equity raise in 20 years and its second since going public, at a scale $10 billion larger than the SpaceX IPO, with the stated purpose of funding computing capital expenditure. It is also notable that Berkshire Hathaway participated in that equity raise.
Ultimately, this contract demonstrates that GPU clusters have begun to function as strategic assets beyond mere facilities. The fact that SpaceX has secured both Alphabet (Google’s parent company) and Anthropic — entities that are in competitive relationship with each other — as AI infrastructure customers further attests to SpaceX’s standing as an independent platform provider in the AI infrastructure space.
3. Meta Exploring Equity Raise of Hundreds of Billions of Dollars
• Core Source
“Meta Platforms has been grappling with how to finance its AI-related capital expenditure, which is expected to reach $145 billion this year and grow even larger next year.”
“The Financial Times reported, citing sources, that after Alphabet expanded its equity raise from a planned $80 billion to $85 billion and observed a positive market response, Meta has been in discussions to pursue something similar.”
“Meta Platforms is also reportedly exploring the use of mandatory convertible preferred shares — the same structure Alphabet adopted — as a way to cushion the market impact.”
“Meta is pursuing a project to build an ultra-large data center in Louisiana with a total investment of $200 billion. This project is expected to create more than 10,000 jobs in the region in addition to expanding AI computing infrastructure.”
“Meta stated through a spokesperson that the equity raise report is purely speculative and that the company will focus on raising capital in a flexible manner.”
• Expected Impact
The Financial Times report triggered a 5.51% single-day drop in Meta’s share price. The core of the report was that Meta is exploring a multi-hundred-billion-dollar equity raise to fund its AI capital expenditure expansion. Meta immediately denied it as “purely speculative,” but the market reacted differently. With corporate bond issuance already rising and anxiety over free cash flow building up, the equity raise rumor served as a trigger amplifying dilution fears and balance sheet concerns.
Notably, the structure under consideration is not a plain equity issuance but a ‘Mandatory Convertible Preferred’ structure. This is the same mechanism Alphabet used for its $85 billion raise — shares that convert into common stock after a set period. Its advantage lies in cushioning the immediate dilution impact while still securing large-scale capital upfront, which is why it has emerged as the preferred fundraising instrument among hyperscalers competing in AI infrastructure investment.
Behind Meta’s exploration of large-scale external capital lies the explosive expansion of its investment commitments. This year’s AI capex budget stands at $145 billion and is expected to exceed that figure next year. At the same time, a $200 billion ultra-large data center project is being pursued in Louisiana. This is in itself a signal that Meta is acknowledging it has entered an investment cycle that cannot be sustained on internal cash generation alone.
The timing is also noteworthy. According to the source material, Meta is seen as feeling “pressure to front-run capital markets before liquidity gets absorbed by the upcoming mega-IPOs of SpaceX, OpenAI, and others.” In other words, this exploration goes beyond a simple financing need — it carries the character of proactive capital positioning to avoid being left behind in the AI investment race.
4. Musk Discusses ‘Terafab’ Project at ASML Private Conference
• Core Source
“Elon Musk is set to attend ASML’s private technology conference via video link to discuss the ‘Terafab’ project. ASML regards the project as a very serious endeavor.”
“ASML has arranged for Musk to present the Terafab project directly to its own employees.”
“There are currently zero large-scale computer memory fabs in the United States. Zero.”
“Even taking the most optimistic projections from memory manufacturers and logic chip makers combined, supply would still fall far short of anticipated demand.”
“ASML is a company to be treasured and supported. It is arguably the finest company in Europe.”
• Expected Impact
Terafab is a next-generation fab plan designed to self-supply the semiconductor needs of Tesla, SpaceX, and xAI. The fact that ASML arranged for Musk to present the project directly to its own employees — and that Musk attended via video link — suggests that collaboration between the two parties has moved beyond mere exploratory discussions. ASML has characterized the project as “a very serious technical challenge” and is reportedly exploring the possibility of cooperation related to building ultra-high-speed chip production lines.
The backdrop of this project is the structural gap in the U.S. semiconductor supply chain that Musk has publicly raised himself. In his interview with JPMorgan’s Jamie Dimon, he stated flatly that “there are currently zero large-scale computer memory fabs in the United States.” Micron’s Idaho facility won’t begin full-scale production until 2028, and the New York facility not until 2029–2030, with even those amounting to “only a tiny fraction of the total memory needed.” In other words, Terafab reads not merely as a self-supply strategy, but as Musk’s proposed solution to the semiconductor supply shortage that is a core bottleneck in U.S. AI infrastructure expansion.
The connection with ASML carries structural significance as well. ASML is effectively the monopoly supplier of EUV lithography equipment — the absolute bottleneck in leading-edge semiconductor manufacturing. Musk’s direct description of ASML as “arguably the finest company in Europe” is not mere flattery; it signals his recognition that collaboration with ASML is a prerequisite for realizing Terafab. Since securing EUV equipment is a precondition for establishing any new fab, this private event can be evaluated as the first public touchpoint for formalizing the potential partnership between the two parties.
Realistic constraints, however, remain clear. ASML’s EUV tools are in severely constrained supply, and building a new fab requires years of construction and tens of billions of dollars in capital. Musk himself noted that Micron’s Idaho plant won’t be operational until 2028 and the New York facility until 2029–2030, and Terafab is likely to be a similarly long-horizon project. Ultimately, the real significance of these discussions lies not in near-term supply relief but in the long-term positioning toward building an independent advanced semiconductor manufacturing ecosystem within the United States.
5. U.S. Crane (Former TMI) Nuclear Plant Restart
• Core Source
“The U.S. NRC and the DOE’s Office of Energy Dominance Financing published the draft Environmental Assessment (EA) reviewing Constellation Energy’s license application to restart the Christopher M. Crane Clean Energy Center.”
“The draft EA concluded that the proposed federal action would have no significant impact on any environmental resource area and no significant effect on the quality of the human environment (draft FONSI).”
“The NRC plans to conduct a 30-day public comment period from June 8 to July 8, after which it will proceed with the final EA and final decision process.”
“The Federal Energy Regulatory Commission (FERC) approved Constellation’s waiver request to transfer the grid interconnection rights of an existing fossil fuel power plant to the Crane nuclear facility for the purpose of restarting it.”
“Constellation stated that it could maintain its 2027 restart target regardless of whether the waiver was approved, but acknowledged that without approval, the plant’s ability to supply power to the grid following restart would have been constrained.”
• Expected Impact
The passage of the NRC’s draft EA and FERC’s approval of the grid interconnection waiver represent the simultaneous removal of two critical regulatory barriers to the Crane nuclear restart. The Crane Clean Energy Center — formerly operated as Three Mile Island Unit 1 (TMI) — was shut down in 2019 for economic reasons, and Constellation Energy has since been pursuing its restart. The NRC’s draft FONSI conclusion effectively clears the final hurdle in the environmental permitting process, while FERC’s approval of the PJM interconnection rights transfer establishes the legal foundation for the plant to actually supply power to the market once restarted.
The backdrop to this story is the surge in AI data center power demand. According to statements by DOE Associate Deputy Secretary Alex Fitzsimmons cited in the source material, the DOE assessed that “reliability of dispatchable power sources is deteriorating due to premature retirements and rising power demand,” and that “gas turbine supply shortages mean it could take years to secure replacement capacity, necessitating individual reviews of plant closures.” The DOE’s position is that meeting peak demand requires dispatchable sources such as nuclear, gas, and coal — not wind and solar. In this context of accelerating AI infrastructure buildout, nuclear power is emerging as a strategic asset beyond a mere energy source, and the Crane restart is the most concrete illustration of that trend.
Should Crane’s 2027 restart be realized, it would mark the first instance of a closed U.S. nuclear plant being brought back online — a development of considerable symbolic weight. This is not simply the return of a single plant; it could serve as a precedent for exploring the restart potential of nuclear facilities that were prematurely shut down across the United States for economic or other reasons. The DOE has already ordered the continued operation of OUC’s 465MW coal unit in Florida citing data center power demand, making clear that the federal government is actively moving to leverage existing generation assets to secure AI infrastructure power supply. Constellation Energy, as well as the broader nuclear operations and restart supply chain, stands to benefit over the medium to long term.
Comment [0]