On the evening of August 17 Eastern Time, the three major US stock indexes closed mixed, while AI hardware names bucked the trend and pushed higher. Nvidia CEO Jensen Huang formally announced that OpenAI has committed to deploying Nvidia AI infrastructure at massive scale by 2030, with existing and planned deployments totaling approximately 12 gigawatts, with the opportunity to expand to roughly 16 gigawatts. At current scale, this partnership could correspond to approximately $600 billion (about 4 trillion yuan) in Nvidia compute infrastructure revenue by 2030.

At the same time, Nvidia announced a partnership with SB Energy to lock in land, power, and data center shell construction capacity at the PORTS-Pike technology park in Ohio, with initial AI factory capacity of 4.25 gigawatts and an option for Nvidia to use the remaining 3.75 gigawatts. OpenAI will serve as the tenant, with the data center planned to use Nvidia’s full-stack DSX AI factory platform—including GPUs, CPUs, networking, and infrastructure software—with phased deployment expected between 2028 and 2030.

Market Reaction: Memory and Optical Networking Names Surge

Buoyed by AI infrastructure expansion expectations, US-listed chip stocks rallied broadly. As of 22:30 Beijing time, the Philadelphia Semiconductor Index was up 1.96%, Applied Materials gained more than 4%, ON Semiconductor rose over 2%, and ASML ADRs and Texas Instruments each added more than 1%.

Memory-related stocks were particularly strong. Kioxia ADRs soared more than 14%, SanDisk jumped over 8% to its highest level since July 14; Western Digital gained more than 6%, Micron Technology rose over 5%, SK Hynix ADRs climbed more than 4%, and Seagate Technology advanced over 3%. During the Asia-Pacific trading session earlier that day, Kioxia shares had already surged more than 15% by the close of trading in Japan.

Optical networking stocks also rallied in tandem, with Marvell Technology up nearly 6%, Coherent gaining more than 5%, Lumentum rising over 4%, and Corning and Applied Optoelectronics each up more than 3%.

The strength in memory chips is backed by tangible capital expenditure expansion. According to South Korean media outlet Etoday, Samsung Electronics and SK Hynix invested a combined 43.198 trillion won (approximately $30.2 billion) in semiconductor facilities during the first half of 2026, up 35.1% from 31.98 trillion won a year earlier. Both companies are running at 100% capacity utilization, underscoring the intensity of current AI memory demand.

Ohio Deal Details: Nvidia’s LPS Strategy

The deal between Nvidia and SB Energy marks the first time Nvidia has directly moved to lock in land, power, and shell capacity—the three physical cornerstone resources known in the industry as LPS. SB Energy will build, own, and operate the data center, signing a 20-year lease with OpenAI. Nvidia will invest $1.5 billion in SB Energy, joining existing investors SoftBank Group and OpenAI.

Nvidia estimates that each generation of systems deployed at PORTS-Pike represents roughly 1.5 million Nvidia GPUs, capable of generating $150 billion to $200 billion in revenue. The campus has a 20-year lifecycle, sufficient to support multiple equipment refresh cycles.

“AI is becoming infrastructure—the foundation of intelligence for every industry—and land, power, and shell have become critical in the AI era,” Huang said.

Financing Restructuring: Guarantee Scale Sharply Reduced

Notably, Nvidia and OpenAI made significant adjustments to their financing arrangements. According to The Wall Street Journal, the revised structure cuts Nvidia’s potential guarantee from $250 billion to under $120 billion, with Nvidia initially guaranteeing only about half of the planned construction scale.

The adjustment came after investors raised questions about Nvidia using its own balance sheet to support infrastructure spending. Nvidia made clear that its credit support does not cover the entire campus cost, nor does it backstop all of OpenAI’s lease obligations. Instead, it is limited to specific portions of lease and power payments, plus a specific residual value commitment. The guarantees will take effect in phases as data centers come online between 2028 and 2030.

Huang previously denied questions about “circular financing.” He emphasized that AI compute has become essential infrastructure for a new era, and that building a financing platform reflects “real and long-term demand” for AI infrastructure—not bubble demand driven purely by capital speculation.

In response to questions about whether compute equipment could be quickly redeployed if OpenAI were to vacate the site in the future, Nvidia’s position is that its compute equipment is highly versatile and easily transferable, with capacity that can be redirected to cloud service providers, enterprises, or other AI startups across the global ecosystem. The CUDA ecosystem makes these devices high-value assets that can be leased and financed at any time, and the value of the site is not tied to a single customer or a single generation of equipment.

The Debate Over Compute Financialization and Its Risks

Nvidia’s series of moves is not an isolated event. On August 10, 2026, Nvidia announced a partnership with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to establish an independent compute financing platform aimed at raising more than $500 billion in third-party capital for AI infrastructure buildout. Nvidia has the option to provide backstop support of 25% (i.e., $125 billion) for potential transactions.

The core logic of this model: companies hungry for compute no longer need to pay the full price of expensive chips upfront. Instead, they pay a monthly compute usage fee through the financing platform—like “renting an apartment.” This “lease” can be packaged into financial products and sold to investors, allowing compute to circulate like bonds.

However, skepticism has followed. Michael Burry, one of the real-life figures behind “The Big Short,” pointed out that much of current and future chip demand is not driven by real end customers, but rather by “circular stacking” through off-balance-sheet financing.

The Bank for International Settlements (BIS), in its report “The AI Investment Race,” noted that the scale and speed of the AI investment boom bears striking resemblance to historical manias, and that near-term downside risks cannot be ignored. If AI compute returns fall short of expectations, risks could rapidly transmit from chip inventories and SPV residual value chains to the broader financial system.

Huang, however, pushed back on social media platform X, citing the A100 GPU launched in 2020 as an example. He noted that six years later, it remains active in training, fine-tuning, and inference, with its economic life extending to nearly a decade. One-year rental prices for H100 GPUs rose from approximately $1.70 per GPU-hour in October 2025 to about $2.35 per GPU-hour in March 2026. AI neocloud provider CoreWeave recently signed a contract extending A100 leases through 2029, further challenging the notion that AI chips are “here today, gone tomorrow.”

Market Outlook

Nvidia will report earnings on August 26, with the market expecting earnings per share of $2.07 (versus $1.04 a year earlier) and revenue of $91.91 billion (versus $46.74 billion a year earlier).

Looking at the broader US equity market, Evercore ISI strategist Julian Emanuel said the S&P 500 could reach 9,000 within 12 months, representing roughly 16% upside from current levels. He believes that traditional bull-market killers—recession, surging long-term yields, and extreme investor euphoria—have not yet appeared. The AI boom, low corporate leverage, and diversified investor positioning will provide further upside momentum.

Capital.com senior analyst Kyle Rodda said the biggest headwind facing the market remains geopolitical uncertainty, though relatively limited military activity in the Middle East has somewhat reduced market volatility in recent weeks.