AI Infrastructure

The substrate maker plans NT$53.7 billion in capex to unlock high-bandwidth packaging as industry faces potential 42 percent supply shortfall by 2028

Oracle's founder cancels planned share offering amid 23 percent decline and heavy debt load from infrastructure spending

Oracle and OpenAI's New Mexico data center faces legal challenges and delayed air quality permits, prompting a major renewable energy procurement to support full carbon-free operations by 2031

Oracle's five-year computing deal with OpenAI lifted remaining performance obligations to $523B-$638B, but data center execution risk and debt burden cloud the growth narrative.

The IT distributor's HIVE segment posting 50% YoY growth as hyperscalers build out capacity; Street now sees $285 price target vs. $166 prior.

New semi-custom platform on TSMC N3P doubles socket performance versus N3, reshaping the CPU economics for hyperscalers building custom silicon

Governor Greg Abbott reversed support for AI data centers citing power demand gaps, while Trump called rejection of the industry a mistake.

Bloom Energy's $69.4 billion market cap enters the S&P 500 on Sept. 21, attracting passive capital to its fuel cell infrastructure bet—but dilution risk and project execution remain critical watch points.

The British AI infrastructure firm is pursuing convertible notes and Nvidia financing ahead of a potential September IPO, betting signed customer leases as proof of demand.

Despite Asia's loan market drought, ByteDance attracted $30+ billion in orders for its financing package, underwriting a $70 billion 2026 capex plan that reflects winner-take-most dynamics in AI infrastructure.

Virginia's Loudoun County hosts 250 data centers generating nearly half its tax revenue, offering a rare case study in how local governments can profit from the AI infrastructure boom.

The French chip materials maker is securing two years of forward revenue through multi-year contracts backed by customer cash deposits, but a smartphone business trough threatens consolidated profitability.

Meta is testing robots from vendors like Kinova and ABB to handle tasks such as server power cycling and cable swaps, aiming to reduce operational expenses.

Earnings reveal Nvidia's strategic pivot toward specialized systems and storage optimization as gigawatt-scale data centers become the new competitive battleground.

Nine companies chased the AI routing startup. Stripe won by recognizing that metering every enterprise's AI consumption is worth more than processing its payments.

Formal consensus at $92.3B masks trader expectations of $107B–$110B Q3 guidance; JPMorgan projects Q2 at $94–$95B, but gross margin and capital allocation may matter more than the beat.

Forced off Nvidia's most advanced accelerators, Chinese labs built leaner training methods and inference architectures—now pairing that software edge with domestic silicon closing the hardware gap.

Custom silicon outperforms Nvidia Blackwell on efficiency metrics, but volume production delayed to 2027 as Nvidia's next architecture looms.

Intel detailed its next-generation Xeon 7 at Hot Chips 2026, revealing a 16-chiplet design and centralizing memory I/O—a structural shift aimed at latency-sensitive AI workloads where GPUs don't pencil out.

The Santa Barbara startup's quantum dot comb laser consolidates multiple light sources into one chip, cutting data-movement power by up to 40 percent as hyperscalers face optical networking bottlenecks.

Two-to-four-year lead times for grid-scale power transformers mean hyperscalers who ordered equipment years ago now hold a structural competitive advantage over rivals still waiting.

By taking minority stakes in power and land developers, Nvidia converts open competitive bidding into guaranteed sales by tying facility design to its chips.

The accelerator extension of Vera Rubin NVL72 hits fastest recorded throughput for Gemma 4 31B, with Nebius first to deploy it commercially.

Stripe acquires the AI model router to embed token optimization into its economic infrastructure, positioning itself as the intermediary between businesses and 400+ models from 80 providers.

The three-day San Francisco conference draws 10,000 founders and investors Oct. 13-15; early pricing ends Aug. 21.

The London startup's platform routes workloads to specialized chips, cutting inference costs by 70 percent in early customer deployments.

CFO Jason Child says demand for Arm's new data center chips could exceed $2 billion in fiscal 2027-2028, but TSMC capacity and memory availability are the real bottleneck

CEO Dario Amodei and six co-founders will hold a special voting class before a Wall Street debut expected as early as late September, insulating them from outside shareholder pressure.

By withholding 2027 capacity from long-term sales, the AI compute provider is charging double its standard rate for short-duration contracts and running auctions that clear 15-20 percent above historical highs.

Samsung's 4nm and 5nm wafer prices jumped 10-15% in July, marking the first time the loss-making foundry division could raise prices instead of compete on cost

CFO Jason Child says Arm expects more than $2 billion in customer demand for its AI chips across fiscal 2027 and 2028, as the company races to secure fabrication capacity.

The Moses Lake silicon anode plant is targeting 250 GWh of annual capacity, and the customer pitch no longer centers on cars.

The flaw scores 9.4 out of 10 on the CVSS scale and lets an attacker run arbitrary code on a developer's machine through a malicious website visit.

Taiwan industry reports peg this year's foundry price increases at 15-20 percent, with further hikes expected in 2027 as AI-driven wafer demand keeps utilization rates elevated.

BHP projects global copper demand growing from 34 million tonnes per year today to more than 50 million tonnes by 2050, with data centers alone adding roughly 3 million tonnes.

The two industrial firms are jointly developing a solid-state transformer rated to 36 kV AC that cuts grid-to-rack power conversion to a single step.

Selective AI investors are concentrating in infrastructure names; Marvell now projects 40 percent full-year revenue growth to $11.5 billion as data center demand outruns earlier forecasts.

Azure's 33 percent constant-currency growth and a debt-to-equity ratio that beats four major software rivals give Microsoft a structural edge its competitors lack.

Released Aug. 13 alongside the V4 Pro model, DeepSeek Harness ships full source code under MIT and lets developers swap models, tools and agent loops in configuration alone.

The startup is applying futures-style hedging to AI inference costs as enterprise budgets strain under volatile token and GPU pricing.

China's largest foundry posted a 93.7% utilization rate and tripled shareholder profit to $479.2 million in Q2, then announced another price increase for Q3.

The AI infrastructure provider posted a $626 million net loss as its revenue backlog hit $104 billion, with $25 billion in new third-quarter commitments not yet included.

The world's largest printed circuit board maker is pouring capital into server and optical communication products as AI demand drives record monthly revenue.

The company's Aegaeon system allows 213 Nvidia GPUs to perform the work of 1,192, cutting hardware requirements and improving cloud margins.

AI startup Runlayer alleges HR firm Rippling copied its Model Context Protocol gateway after a year-long product trial.

Stripe is in talks to acquire AI model marketplace OpenRouter for roughly $10 billion, a deal that has triggered a surge of inbound interest for competitor startups.

DeepMind's 2016 Go win produced a single stone placement that launched a decade of AI research and a commercial infrastructure race now worth hundreds of billions.

Elon Musk said during SpaceX's first earnings call that all future AI infrastructure will run on Nvidia chips, citing the Vera Rubin architecture.

South Korean memory giants posted their strongest daily gains in more than a year as sold-out HBM production through 2026 gives both companies rare earnings visibility.

The data center giant targets investment-grade buyers as credit spreads on technology infrastructure debt widen and duration risk mounts.