Chips, GPUs, processors, and the physical infrastructure powering the digital world.
60 articles
The new customer lineup answers the demand question that has shadowed ASML's $400 million next-generation machines, giving the Dutch equipment maker three of the industry's biggest chipmakers and signaling a costly shift to larger photomasks.
With domestic memory inside and a homegrown processor at its core, the foldable doubles as a supply-chain statement and a test of how far Xiaomi's premium brand can stretch.
The N1X silicon pairs Blackwell graphics with Grace-class CPU cores and up to 128GB of unified memory, and the first laptops and desktops were shown at IFA 2026 before an October release.
Shown at IFA 2026, the workstation combines a Ryzen Threadripper PRO 9995WX with 288 gigabytes of HBM3E and up to 2 terabytes of DDR5, aimed at running trillion-parameter models locally.
The premium that early buyers are paying buys a router built around reliability rather than peak speed, with the Wi-Fi 8 specification still technically unfinished.
Hyperscale demand is driving the chipmaker's growth, with custom accelerator shipments more than tripling year over year and data center networking doubling, while non-GAAP profit and free cash flow both climbed sharply in the quarter.
The price gap between negotiated and open-market memory has widened to a multiple of several times, and it is squeezing device makers, laptop buyers and the phone aisle rather than the data centers that caused the squeeze. Contract deals stretching years into the future are now the rule, leaving the spot market to whoever cannot commit.
Revenue climbed 58 percent year over year to $47 billion while per-share earnings more than tripled, and Dell raised its full-year revenue target by $25 billion on demand for AI systems that now dwarfs what it can ship in a single quarter.
Nvidia is investing $3.5 billion in the Taiwanese chip designer MediaTek through convertible bonds, a bet that helping cloud giants assemble their own chips will keep its interconnect central to AI data centers. The investment turns a would-be rival trend into a toll revenue engine.
After Samsung's July increase, TSMC is expected to raise foundry prices 10 to 15 percent across all nodes, with 3nm already up 15 percent and N2 and N5 set for further hikes by early 2027.
Nvidia reported fiscal Q2 2027 revenue of $96.2 billion on August 26, more than double a year earlier, with data center revenue of $89 billion and a Q3 guide of roughly $108 billion, while warning that surging memory costs are compressing gross margins.
NVIDIA shares jumped 6% on August 27 after the company reaffirmed a 70% data-center growth forecast for fiscal 2027 and confirmed that Rubin GPU samples are now in select hyperscaler labs, ahead of the Q4 2026 launch.
Intel at Hot Chips 2026 confirmed Crescent Island ships with 32 Xe3P cores, 256 XMX engines, and up to 480GB of LPDDR5X in a 350W air-cooled PCIe card, with customer sampling targeted for H2 2026.
Bloomberg reported on August 23 that Nvidia has notified hyperscaler customers that Grace Blackwell and Vera Rubin systems shipping in early 2027 will see price hikes above 15%, as HBM4 supply constraints push up AI server bill of materials.
Nvidia disclosed a strategic equity stake in DSX on August 22, the third land-and-power deal of August 2026 following investments in Cloverleaf and Lancium. The three investments give Nvidia control over an estimated 6.4 GW of data-center-ready power capacity across Texas, New Mexico, and Utah.
Bloomberg reports customers including Microsoft, Google and Oracle have been told Vera Rubin and Grace Blackwell system prices will jump more than 15% on early-2027 shipments — Samsung is also raising 4nm foundry prices by up to 15%, with Chinese customers absorbing the largest hikes.
Mercury Research data shows AMD's x86 share climbed to 34.1 percent in Q2 2026 — a fresh all-time high — on the same day TSMC confirmed 1.6-nanometer development was complete and disclosed that it has begun outsourcing CoWoS packaging work to Intel's Malaysian plant.
SK Hynix announced a 3.3% treasury share retirement worth roughly $10 billion on August 22 while TSMC's C.C. Wei told investors on August 21 that chip supply cannot meet AI-fueled demand 'for years.' The two messages together redraw the memory-versus-foundry bottleneck map for 2027.
Google secured warrants to buy 59 million Marvell shares as part of a custom AI silicon deal aimed at cutting reliance on Broadcom. The deal, reported on August 20, is the third major non-NVIDIA AI silicon announcement in two weeks and accelerates the disaggregation of the AI accelerator stack.
ByteDance and Tencent have each received around 10,000 Nvidia H200 AI chips, according to Financial Times reporting on August 18 and follow-ups on August 19. The small-batch shipments run at roughly 13 percent of the legal license ceiling and mark the first concrete H200 deliveries to mainland China since the January 2026 approval.
Supply-chain reporting on August 14–17 confirmed that NVIDIA will skip TSMC's N2 process for its post-Rubin 'Feynman' generation and go straight to the A16 1.6nm node with backside power delivery, 3D chiplets via SoIC, custom HBM4E, and co-packaged optics. TSMC is accelerating AP7 and AP8 construction to support a potential 50,000-wafer SoIC run rate by end-2027.