📁 Market

GameStop's CEO calls Sony's disc-less console decision 'totally irrelevant,' noting physical software accounts for just 12% of the business. The figure marks the relentless digital shift but obscures a deeper question: handing control to someone else's servers comes at a cost. For on-premise AI, the argument is identical.

OpenAI CFO Sarah Friar introduces a scorecard to measure AI ROI across four dimensions: useful work, cost per successful task, dependability, and return on compute. The framework shifts the conversation from theoretical potential to verifiable business metrics. For on-premise operators, the ability to track value per GPU cycle becomes a competitive factor.

US lawmakers push to ban memory chips from China, even through allied supply chains, citing national security risks. The move could reshape procurement of critical components for on-premise AI hardware.

Runta raised $20 million to rein in autonomous AI agents, with a16z leading the round. Analysis: why 'parenting' agents points to a new control infrastructure that closely concerns those running on-premise LLMs.

The Kimi K3 team launched their model by openly admitting it’s not matching the top proprietary models. A rare move in a hype-driven sector, this honesty could shift trust dynamics in enterprise markets and influence self-hosted deployment evaluations.

ASML signals intent to raise prices for its Low-NA EUV lithography tools, moving beyond the existing productivity-based model. The Dutch company aims to capture the full value of all the advantages its systems deliver, not just improvements in wafers per hour. This shift could increase costs for advanced chipmakers and, downstream, for AI hardware, potentially impacting total cost of ownership calculations for on-premise infrastructure.

The Japanese company ACSL is betting on Taiwan to expand its drone supply chain, and its commitment to TADTE 2027 signals a strategic shift. This move goes beyond airframes: the value chain for autonomous systems now hinges on the ability to manufacture and train AI on-premise, reflecting a broader push for technological sovereignty.

The UK’s development bank commits €25 million to EQT’s health-focused fund. The move underscores a policy drive for medical AI, which will intensify demand for sovereign, on-premise compute infrastructure in healthcare.

The Taiwan-Japan AI tech forum aims to strengthen partnerships that could reshape the semiconductor supply chain for artificial intelligence. The goal: a more resilient regional production pipeline that also benefits organizations choosing on-premise deployment.

TSMC's CoWoS packaging capacity, critical for AI GPUs, remains extremely tight. This constraint slows the availability of essential hardware. However, OSAT partners are intensifying their efforts to expand production, signaling a potential easing of bottlenecks in the medium term. The situation highlights the complexity of the supply chain and the challenges for those planning large-scale AI deployments, especially in on-premise contexts.

SoftBank is preparing a record $60 billion bond sale to back its OpenAI investment, underscoring the immense capital appetite of frontier AI. The move raises questions about the sustainability of a model that concentrates resources in a few cloud players, prompting organizations to assess on-premise alternatives for cost and data control.

Z.ai, the Chinese startup behind GLM, projects $1bn in annual sales while giving away its most powerful LLMs free. A model that challenges paywalled APIs and shifts value toward concrete deployment, with direct implications for those choosing on-premise.

Montage Technology, a Chinese chip designer, raises first-half profit forecasts on explosive AI-driven demand for memory interfaces, but simultaneously discloses a search by South Korean prosecutors at its local office, highlighting the escalating IP battles in the semiconductor supply chain.

Soaring demand for high-bandwidth memory (HBM) used in AI training and inference is sparking unprecedented competition. Automakers, increasingly reliant on specialized chips for autonomous driving and smart manufacturing, are scrambling to shield their supply chains with long-term procurement strategies and direct investments in production capacity.

SK Group Chairman Chey Tae-won floated a 'memory as a service' model for SK Hynix at Computex 2026, with Nvidia CEO Jensen Huang in attendance. It signals a shift toward flexible hardware consumption, with potential impacts on TCO and data sovereignty for AI deployments.

The Taiwanese manufacturer's massive investment reshapes the semiconductor supply chain. For those running LLMs on-premise, future GPU availability hinges on this move, amid geopolitical tensions and explosive compute demand.

As Taipei reassures about domestic advanced chip leadership, greenlighting TSMC's US expansion redraws the supply map. For those building on-premise AI infrastructure, this signals a future of fragmented supply chains, recalculated TCO, and geopolitical knots that become project variables, not background noise.

The stagnation of the electric vehicle market after tax credits expired mirrors the challenges of deploying LLMs on local hardware. High GPU costs and energy bottlenecks threaten digital sovereignty plans. A lesson for those designing self-hosted stacks.

Rumors of 5- and 10-trillion-parameter models fuel the suspicion that top labs' edge comes not from exclusive algorithms but from the ability to train models at unrivaled scale. With DeepSeek V4 and Kimi K3 breaking the trillion-parameter barrier, the battleground shifts to hardware and on-premise deployment.

The US telecom giant is shedding 274 corporate stores to independent franchises and cutting 3,000 jobs. The move fast-tracks AI-driven customer service automation and raises questions about sensitive data handling in cloud environments.