Technology Sector
AI infrastructure ambitions collided with regulatory and market headwinds today. Nvidia doubled down on power capacity for data centers while OpenAI paused its largest training run after a security breach, and energy bottlenecks are forcing 30–50% of data-center projects to delay or cancel. Separately, Samsung and Alibaba disappointed investors despite strong fundamentals, while regulators tightened oversight of automation and AI systems.
Nvidia has taken a minority stake in Cloverleaf Infrastructure, a data center power developer founded in 2024, in a deal expected to be worth several hundred million dollars. Cloverleaf acts as a middleman between utility companies and data center operators, securing long-term power contracts and preparing shovel-ready sites before construction begins. The company has delivered projects totaling more than 7 gigawatts of powered land since its founding and holds a pipeline exceeding 10 gigawatts. Under the partnership, Cloverleaf will adopt Nvidia's DSX platform—a reference design that integrates site selection, power, cooling, computing, and facility decisions earlier in the project lifecycle. Nvidia is embedding itself upstream in the data center supply chain, locking in power capacity years before chips are deployed. The move follows Nvidia's $1.5 billion investment in SB Energy this week and its earlier 20 percent stake in Lancium, signaling a deliberate strategy to control the physical infrastructure bottleneck that now constrains AI data center buildout more than chip availability itself. J.P. Morgan Securities advised Cloverleaf on the transaction, with Kirkland & Ellis as legal counsel.
The FDA opened a public comment period on a discussion paper outlining how it might regulate generative AI-enabled medical devices, seeking feedback from manufacturers, clinicians, and researchers by October 19. The agency's Digital Health Center of Excellence proposes a two-axis risk framework and a "competency-based" premarket evaluation modeled on how physicians are trained and credentialed, rather than how traditional software is tested. Under this approach, the FDA would assess the final user-facing device—not foundational models or subcomponents—through non-clinical benchmarking (testing clinical knowledge, safety behavior, and generalizability, potentially using synthetic data and virtual patient avatars) and clinical confirmation (prospective studies, retrospective patient data analysis, shadow deployment in live workflows, or patient actors). The agency acknowledged that generative AI devices pose unique regulatory challenges: they accept open-ended inputs, produce variable outputs, can "hallucinate" convincingly, and may evolve after launch in ways that degrade performance unpredictably. The paper also explores whether the FDA should accept greater premarket uncertainty in exchange for stronger postmarket monitoring—a shift from traditional device regulation. The framework is not new policy but an invitation for early input; the FDA has authorized numerous AI-enabled devices to date, though most do not use generative AI. Acting Commissioner Kyle Diamantas framed the effort as aligning with the Trump administration's priority to accelerate AI-powered medical products to market while maintaining safety.
Samsung Electronics said Friday its shareholder returns this year could reach 110 trillion won ($79.54 billion), more than five times the previous high of 20.3 trillion won in 2020. The package includes 30 trillion won in cash dividends in the third quarter and 15 trillion won in share buybacks already executed for employee stock bonuses; the board will decide remaining payouts in January 2027, choosing among cash dividends, buybacks, and share cancellations. The announcement came as Samsung's chip profits surged more than 250-fold in the second quarter to 89 trillion won, driven by AI demand, and the company committed to returning 50% of free cash flow accumulated over its 2024–2026 policy period to shareholders. Shares fell 5.2% in early Monday trade despite the announcement, though they had gained 3.5% on Friday when the plan was disclosed. Analysts noted the phased approach—with the final decision deferred to January—may have disappointed investors who expected the full package immediately, and that a larger buyback component would have created more direct demand for the stock. Samsung and SK Hynix together are set to hold $263 billion in net cash by year's end, more than double Nvidia's estimated $102 billion and exceeding the combined cash of the other six Magnificent Seven U.S. tech companies.