All articles tagged with #gpus
AMD is reportedly planning a 10% price increase across its GPUs, motherboard chipsets, and possibly CPUs as wafer costs rise with TSMC. Because AMD relies on TSMC for fabrication, higher wafer pricing is passed on to partners and consumers, with details on how the changes will be communicated still unclear and expected to be defined in Q4.
AI firms are reportedly purchasing bulk GeForce RTX 5090 GPUs for AI server workloads, bypassing NVIDIA’s data-center accelerators; photos shared by HKEPC show pallets of new cards being readied for AI rigs, and retail prices have surged past $5,000—well above the $1,999 launch price. Some images may be edited, but the situation mirrors a crypto-era squeeze as AI demand climbs, with NVIDIA not stepping in to throttle or restrict the consumer card’s AI use.
Nvidia CEO Jensen Huang congratulated OpenAI on Astra and proclaimed that “AGI has arrived,” noting Astra was trained on Nvidia GPUs as OpenAI promotes Astra as the most capable model; critics question the AGI definition while OpenAI defends Astra’s capabilities. Nvidia also reported a huge quarterly revenue and, with its data-center GPUs driving demand, signals 400,000 more GPUs coming online as AI workloads expand.
Nvidia is shifting its AI moat from chip leadership to financing, backing OpenAI's Ohio data center with up to $105 billion and partnering with Wall Street firms to pursue about $500 billion in GPU financing, all backed by a stronger balance sheet, a dividend boost, and a large stock buyback. The move broadens Nvidia's equity bets across the AI ecosystem to accelerate infrastructure buildout amid growing competition from Google and AMD, with analysts viewing the strategy as a durable extension of the AI investment cycle.
Nvidia will provide more than $105 billion in financing to back the PORTS‑Pike Technology Campus in Ohio, a new OpenAI data center with initial 4.25 gigawatts of compute (potentially expanding to 8 GW) powered entirely by Nvidia GPUs and leased to OpenAI for 20 years; Nvidia also invests $1.5 billion in SB Energy, and when combined with prior deals could total up to $600 billion in AI infrastructure activity, while Nvidia says the arrangement is about long-term capacity and denies circular investing.
Nvidia signs MOUs with six major asset managers to create compute financing platforms that could raise up to $500 billion to fund AI hardware, borrowing against GPUs and related infrastructure and potentially guaranteeing part of the deals. This moves AI capex off corporate books, easing hyperscalers’ spending while raising questions about GPU depreciation, credit risk, and the circularity of financing the AI boom.
SpaceX has completed its $60 billion takeover of Cursor (Anysphere), gaining Cursor’s GPU compute capabilities and engineers to accelerate SpaceXAI’s Grok-based AI tooling; Cursor even announced Firetiger joining the day before closing, signaling a rapid integration. The move expands SpaceX’s shift from rockets to a software-led AI business aimed at competing with Anthropic and OpenAI, while questions linger on pricing, unit economics, and how the Cursor brand will endure as compute becomes central to value.
Nvidia said it secured about $500 billion from Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to fund AI infrastructure, including new data centers and chip-manufacturing capacity. The funding treats AI compute as a distinct asset class and will back Nvidia’s own projects as well as those built by its partners, aiming to meet soaring demand from major tech companies for AI workloads.
AI data centers run GPUs nearly 24/7 in hot, high-use conditions, cutting their lifespans to about 1–3 years (up to 5 with lower load), far shorter than gaming GPUs. In gigawatt-scale centers with hundreds of thousands of GPUs, replacement cycles contribute to supplier shortages and price pressures for consumer GPUs. Google’s TPUs are cited as longer-lasting at 7–8 years under heavy use, and cooling innovations like Nvidia’s liquid cooling aim to extend uptime, offering a path to stabilize prices as data-center hardware improves.
Nvidia shares rose about 2.2% after SpaceX said it will build its AI platform exclusively on Nvidia GPUs, signaling a major ongoing customer for Nvidia’s chips and underscoring Nvidia’s central role in AI infrastructure as SpaceX expands its AI capacity. The deal size wasn’t disclosed, but SpaceX aims to secure a large portion of Nvidia’s GPU supply next year, which could translate into sustained GPU orders and a lifting of Nvidia’s market position.
SpaceX will exclusively buy Nvidia GPUs, Elon Musk announced on the earnings call, citing Nvidia's Vera Rubin architecture as the best AI computer and signaling SpaceX could become a meaningful share of Nvidia's revenue next year. Nvidia has long supplied GPUs to SpaceX and Tesla. SpaceX posted Q2 revenue of $7.8B, up 92% year‑over‑year, driven by AI infrastructure and Starlink, while Musk reiterated a goal of $100B in annual revenue by year‑end; however, high AI spending contributed to a $541M quarterly loss and shares fell after hours. The exclusivity underscores Nvidia’s deep partnership with Musk’s ventures, while reducing diversification risk by constraining other chip suppliers.
Nvidia has boosted GPU prices by 20–30% across its lineup—the third hike this year—as surging AI demand tightens supply; the ripple effects extend to DRAM and retailers, potentially affecting PC and console pricing and elevating the cost of upgrading gaming rigs amid a broader affordability crunch.
Nvidia unveiled revenue-sharing agreements with fast-growing AI startups, offering token credits to power development in exchange for a share of future product and cloud revenues. Initial partners Sharon AI (up to 40,000 Nvidia GPUs) and Firmus Technologies (Batam data center targeting 170,000 GPUs and 360 MW) will provide the compute behind the program, highlighting the escalating importance of access to GPUs for AI startups.
The piece argues Nvidia is shifting away from frequent PC-gaming upgrades toward AI/data-center profitability, exemplified by a $5.4B GPU sale to Valor that funds xAI and is booked as revenue while shifting risk. It traces a long-term pattern where hardware gains have slowed and prices risen, aided by rising wafer costs and EUV/node costs, pushing upgrade cycles from years to five or more. Nvidia has leaned on software like DLSS and frame generation to simulate gains and monetize through enterprise markets, rather than delivering regular hardware leaps for gamers. The author suggests Nvidia will continue serving gaming but is unlikely to return to the old, rapid upgrade cadence, leaving room for competitors if the AI/data-center focus remains dominant.
SpaceX disclosed a deal with Google to lease 110,000 Nvidia GPUs and related hardware for AI workloads from Oct 2026 to Jun 2029 at $920 million per month, with capacity ramping starting Sept 2026 and a one-month grace period if not ready; either side can terminate with 90 days’ notice after Dec 31, 2026. The arrangement follows SpaceX's prior GPU rental to Anthropic and underscores its expanding AI-infrastructure push ahead of an anticipated IPO valued around $1.8 trillion, with Grok AI competing against Claude and Google's Gemini and OpenAI's ChatGPT.