AI Compute & Accelerator Silicon

UBS now expects artificial intelligence capital expenditure to reach nearly $1 trillion in 2026 before climbing to around $1.4 trillion in 2027, with surging memory costs accounting for most of the increase. The bank's updated estimates put total AI capex at $998 billion this year, almost double the $506 billion recorded in 2025, and forecast spending of $1.447 trillion next year. Memory is emerging as the main source of that growth, with UBS estimating memory spending will climb from $71 billion in 2025 to $367 billion this year and $923 billion in 2027, while other AI-related costs are estimated at $631 billion in 2026 before declining to $525 billion next year. That means higher memory costs will account for about 60% of the increase in AI capex this year and more than the entire net increase in 2027, and across the two years UBS calculates that roughly 90% of the nearly $1 trillion increase in AI capital expenditure between 2025 and 2027 will come from higher memory spending. Memory represented about 14% of total AI capex in 2025, a share the bank estimates will rise to 37% this year and reach 64% in 2027, and UBS said price-driven increases would add relatively little to real U.S. gross domestic product, instead representing a transfer of income and profits toward memory producers in Asia.

Qualcomm has joined Advantech and other partners in the new WEDA edge AI ecosystem, an alliance aimed at supporting scalable deployments across varied hardware by standardising tools and reference designs for industrial and enterprise use cases. Qualcomm is also participating in up to 29 cross-sector agreements between US and Vietnamese firms announced during a recent state visit. The company develops and licenses wireless technologies that underpin mobile devices and connected equipment, and the moves tie its core wireless IP to new industrial and enterprise computing uses. Whether the WEDA ecosystem projects and Vietnam partnerships turn into disclosed design wins or revenue targets in segments like industrial IoT and AI edge devices over the next few reporting periods, rather than staying at the memorandum of understanding stage, remains to be seen.

Microchip Technology has been selected by AnalogAI to supply its memBrain SAGE neuromorphic hardware IP for use in new edge AI processors. The collaboration centers on compute-in-memory technology targeting training and inference for adaptive robots, drones and vehicles, with AnalogAI planning to integrate memBrain SAGE into real-world, low-power AI systems designed for deployment at the network edge. Microchip Technology is a US-based semiconductor supplier with a roughly $38.5b market cap focused on smart, connected and secure embedded control solutions. The design win supports the edge-computing catalyst in the company's narrative, though analysts flag execution risk from competition, including edge-AI platforms from Nvidia and Qualcomm.

Micron Technology and Intel CEOs cautioned that memory chip shortages and higher prices could persist through 2027, with Micron's leadership indicating on a recent call that supply constraints may only start to ease meaningfully from 2028 onward. Intel's CEO echoed the outlook for extended tightness in DRAM and NAND availability, pointing to heavy AI and data center demand. Micron Technology designs and produces memory and storage hardware used in everything from smartphones and PCs to data centers, so long running tightness in DRAM and NAND supply directly touches the products it sells into these markets. As one of the larger US based chip manufacturers by scale, with a reported market value of about $1.1 trillion, its comments on supply conditions can influence how investors think about capacity planning across the wider semiconductor sector. The clearest test of this read will be how Micron's long term customer agreements and utilization plans look through 2027, especially whether the company keeps reporting high take or pay coverage across its AI oriented memory output as new fabs and its 512GB DDR5 modules move toward volume production in the second half of 2027.

Jim Cramer said Broadcom CEO Hock Tan told him demand for AI compute infrastructure remains extremely strong and durable, with the custom chip designer holding more orders than almost anybody other than Jensen Huang. Cramer's remarks on CNBC center on whether Broadcom can keep capturing custom AI chip orders, and the third quarter earnings released on September 2nd support the growth narrative, with revenue up 86%, AI semiconductor revenue up 221%, and fiscal year 2026 guidance implying 186% annual AI revenue growth. Tan reaffirmed that Broadcom could pull in $115 billion in annual AI chip sales in 2027 and $230 billion in 2028. Still, the fiscal fourth quarter guide left investors wanting more, with $34.8 billion in revenue missing analyst estimates of $35.03 billion and gross margin guided at 73%, a five point annual drop due to a higher mix of XPU sales. Estimates suggest 71% of Broadcom's fiscal 2027 and 2028 XPU deployment could rely on OpenAI and Anthropic, meaning a huge portion of orders might come from firms now calling for a slowdown in AI development. In Q2, 170 out of the 1,006 funds tracked by Insider Monkey held a stake in Broadcom, a slight drop from 163 out of 1,022 funds in Q1, with notable exits including Third Point and Two Sigma Advisors, and the stock trades at a forward P/E ratio of 18 versus NVIDIA's 23.42.

S&P 500 earnings are expected to increase by +24% from the same period last year in the third quarter, the 8th straight quarter of double-digit earnings growth for the index, according to Zacks Investment Research. Earnings are expected to be above the year-earlier level for 14 of the 16 Zacks sectors, with 5 sectors expected to enjoy double-digit growth: Aerospace up +159.3%, Energy up +111.9%, Tech up +41.9%, Basic Materials up +31.2%, and Transportation up +15.1%. The Conglomerates sector is the only one expected to have lower earnings in Q3 relative to the same period last year, down 35.4%, while Consumer Staples earnings are expected to be flat. Excluding the Energy sector, Q3 earnings growth for the S&P 500 drops to +20% from +24%, and excluding the Tech sector, growth for the rest of the index drops to +14.4%. Nvidia's Q3 earnings are expected to increase +90% year-over-year on +91.2% higher revenues, while Micron's year-over-year earnings and revenue growth rates are expected to be +938% and +348.6%, respectively, and Tech sector earnings growth gets cut by slightly more than half once contributions from Nvidia and Micron are excluded. The Q3 earnings season will get the spotlight when the big banks report on October 13th, but the reporting cycle actually got underway with the September 10th quarterly releases from Oracle and Adobe, followed by homebuilder Lennar as the third S&P 500 member to report such Q3 results, with an additional six index members on deck this week including Costco, AutoZone and Darden. Total Q3 earnings for the three S&P 500 members that have reported results already are up +22.6% from the same period last year on +14.9% higher revenues, with 33.3% beating EPS estimates and 66.7% besting revenue estimates.

Humanoid robots are making meaningful progress in autonomy, dexterity and repetitive real-world tasks, but widespread commercial deployment remains a work in progress, according to a report published Wednesday by Citrini Research. After a week meeting robotics labs, engineers and investors across the San Francisco Bay Area, the firm found most humanoid robots remain in development, training and tele-operated demonstrations, with real-world deployments closer to pilots than clear return-on-investment opportunities. The report examined whether robotics is approaching its own ChatGPT moment and highlighted advances from GeneralistAI and Skild AI, along with Figure AI's demonstration involving 250,000 packages, suggesting the field may advance through a mosaic of signals rather than one defining breakthrough. Basic humanoid locomotion is effectively a solved problem while complex dexterity remains more difficult, and simple repetitive tasks are already trainable. Investment in the broader robotics infrastructure is keeping pace: Tesla Inc has been preparing production lines for Optimus, XPeng Inc has launched an automated production line for its IRON humanoid robot, and Figure AI is securing access to up to 100,000 Vera Rubin GPUs from NVIDIA Corp to train the models powering its humanoid robots. The report concluded robotic autonomy is developing along a curve rather than arriving through one decisive breakthrough, with broad generalizability remaining the longer-term goal.

Nvidia has committed at least $6.5 billion to photonics companies over the three months prior to May 2026, including $500 million to Corning for advanced optical connectivity solutions and $2 billion each to Lumentum, Coherent and Marvell, according to CNBC. The impact has been significant, with Marvell shares gaining 183.4% year to date, while Lumentum, Corning and Coherent have advanced 142.5%, 68.8% and 60.3%, respectively. Goldman Sachs projects the total addressable market for optical networking to grow from roughly $15 billion in 2026 to about $154 billion by 2028, a more than ninefold increase. Adoption of co-packaged optics is expected to accelerate starting in 2027, reaching roughly 20% to 30% penetration by 2028, while Mordor Intelligence estimates the global photonics market will reach $1.46 trillion by the end of 2026. Investors seeking diversified exposure can consider the Roundhill Photonics & Optics ETF LYTE, with $254.2 million in assets under management across 19 holdings; the KraneShares Photonic and Optical ETF LUMA, with $9.76 million in net assets; and the Tema Photonics & Optical ETF LAZR, with $94 million in net assets across 27 companies.

Intel and SK Hynix are exploring partnership opportunities to manufacture memory chips in the United States, reportedly centered on Intel's long-delayed Ohio manufacturing complex, with possibilities ranging from SK Hynix leasing capacity to a broader joint venture involving cloud companies. No formal partnership has been announced, and SK Hynix stressed that no specific plans or arrangements have been finalized. Since reports of the potential partnership surfaced Wednesday, Intel shares have surged roughly 10%, while SK Hynix has climbed about 5%; SK Hynix stock has risen 20% since its U.S. Nasdaq debut in July, while Intel is up about 4% over the same period. Intel's turnaround is gaining traction independently, with Q2 revenue jumping 25% year over year to $16.1 billion, Data Center and AI revenue surging 59% to $6.3 billion, Intel Foundry revenue up 31%, and management guiding Q3 sales to $15.8-$16.8 billion, while the Zacks Consensus calls for fiscal 2026 revenue to rise nearly 18% and FY26 EPS to soar to $1.50 from $0.42 per share last year. SK Hynix reported record Q2 performance amid robust AI demand, began mass shipments of HBM4 with production expected to ramp through the second half of the year, and shipped samples of HBM4E, while the Zacks Consensus Estimate calls for FY26 revenue to soar over 250% to $240.09 billion and EPS to jump more than 500% to $25.69. Despite its AI-memory leadership, SK Hynix trades at a forward P/E multiple of 7X, while Intel trades at more than 100X forward earnings; SK Hynix carries a Zacks Rank #2 (Buy), while Intel holds a Zacks Rank #3 (Hold).

Nvidia, Google and Emerald AI have launched the AI Energy Management Alliance, or AEMA, a coalition that dynamically manages the electricity use of data centers in response to grid conditions. Emerald AI founder and CEO Varun Sivaram said the alliance's founding members are joined by a cohort of 20 launch partners, including the AI lab Anthropic, the semiconductor firm Analog Devices, and the energy companies AES, NRG, Constellation, RWE and National Grid. Sivaram said Emerald AI, which was founded under two years ago, is building with Nvidia and Digital Realty the world's first from-the-ground-up power-flexible AI data center, a 100 megawatt facility in Manassas, Virginia, that comes online later this year. He said Google, one of the founding members, has already done a gigawatt of demand response for its data centers, while Emerald and Nvidia have completed six demonstrations around the world, in London, Phoenix and Virginia. Sivaram said the alliance is talking to the FERC commissioners, state regulators and the administration about a grand bargain in which flexible AI data centers act as good citizens to grids and communities in return for faster and larger connections to the power grid.

KLA Corporation is benefiting from accelerating artificial intelligence infrastructure spending, which is driving higher investments in leading-edge foundry and logic technologies and boosting demand for its process-control solutions. In fiscal 2026, the Semiconductor Process Control segment's revenues increased 12% year over year, and overall fiscal 2026 revenues rose 12% year over year to $13.58 billion. During the fourth quarter of fiscal 2026, management raised its calendar 2026 wafer-fabrication equipment market estimate, including advanced packaging, to the low-$150 billion range, up from its previous $140 billion-plus view and approximately $120 billion in calendar 2025. KLAC expects second-half 2026 revenues to be approximately 20% higher than the first half, aided by high-performance computing, high-bandwidth memory, increasing extreme ultraviolet adoption in DRAM and advanced packaging technologies such as hybrid bonding, and said backlog was expected to reach about $12.5 billion. The company faces intensifying competition from Applied Materials, which expects its process diagnostics and control business to grow more than 50% in calendar 2026, and from ASML, which expects advanced foundry and logic revenues to rise about 25% in 2026 and extreme ultraviolet revenues to increase roughly 45%.

Micron demonstrated a 512GB server memory module that holds as much memory as four 128GB modules, with AMD and Intel having validated it for future platforms, but volume production is not expected until the second half of 2027. The company said HBM4 has already crossed $1 billion in revenue, with the 12-high ramp running twice as fast as HBM3E, and CEO Sanjay Mehrotra told investors that DRAM and NAND industry demand continues to significantly exceed industry supply, a tightness he expects to persist beyond calendar 2027. Micron has signed 16 Strategic Customer Agreements covering roughly 25% of revenue over their term, with cumulative minimum-price RPO of approximately $100 billion. The stock trades at a trailing PE of 21x and a forward PE of 6x, with the average analyst target at $1,513.11, against consensus fiscal 2027 EPS of $156.0661 on revenue of $244.46 billion, up from a fiscal 2026 EPS estimate of $73.4388. Micron is now the only US maker in both DRAM and NAND after the SanDisk spinoff left Western Digital out of NAND, though the 2027 pricing backdrop the new module lands into cannot be forecast from here.

Intel CEO Lip-Bu Tan warned this week that memory prices have jumped 5x to 7x and now account for roughly 75% of the cost of some lower-end phones and laptops, cautioning the shortage could deepen in 2027. Tan told investors the industry faces one of the most severe supply constraints in its history across leading-edge logic, silicon wafers, memory, and subscripts, and said these shortages will persist for the foreseeable future. Intel is pivoting production toward data-center CPUs and now guides 2026 capital expenditure above $20 billion, with 2027 spending set to run significantly higher. Micron Technology echoed the outlook, saying it sees tightness continuing beyond 2027, aligned with SK hynix flagging 2027 as potentially the industry's worst supply year. Micron's $100 billion in take-or-pay remaining performance obligations and $22 billion in customer deposits and letters of credit lock in floor pricing above prior peak margins, while its fiscal third-quarter gross margin hit 84.9% on DRAM pricing gains in the low 60s percentage range and NAND pricing in the mid-80s percentage range.

Seagate Technology received a buy rating and a $899.28 price target from 24/7 Wall St., implying 14.82% upside over the next 12 months at a 90% confidence level. The call follows a blowout fiscal Q4 report on July 28, 2026, in which Seagate posted revenue of $3.63 billion, up 48.49% year over year, and non-GAAP EPS of $5.71, beating expectations by 12.11%, while full-year FY26 free cash flow hit a record $3.105 billion, up 279.58%. Management now guides to annual revenue growth of a minimum of 20% over the next few years, up from prior mid-teens targets, and CEO Dave Mosley said nearline capacity is almost fully allocated through calendar 2027. Mosaic 4 HAMR drives deliver up to 44 terabytes per drive, and Mosaic 5 is targeting 50 terabytes with qualification shipments in late 2027. The stock trades at a forward P/E of 23, roughly in line with Western Digital's 21, while Micron's forward P/E of 6 highlights how much Seagate has already re-rated; the bear case lands at $685.64 on hyperscaler concentration risk.

Apple's iPhone 18 Pro and Pro Max went on sale Friday, the company's first new iPhone launch under CEO John Ternus and its first to hit the market amid the AI-driven global memory and storage shortage. The memory crunch pushed the base iPhone 18 Pro to $1,199 and the base iPhone 18 Pro Max to $1,299, a $100 increase over the iPhone 17 Pros. The foldable iPhone Duo is not due out until Oct. 23 and will cost $1,999. Demand for the Pro models is running hot, with orders placed today facing a three-week wait for delivery. Deepwater Asset Management managing partner Gene Munster wrote Friday that Pro model pre-order lead times are about 15% longer than a year ago on launch day, and that the dual-cycle launch, which splits the base iPhone 18, iPhone 18e and iPhone Air 2 into early next year, appears to be helping early sales. The new phones ship with the improved Apple Intelligence platform and AI-powered Siri AI, which remains in beta.

Meta Platforms' new AI assistant Muse hit number one in the App Store within one week of launch, driving a 60% jump in Meta AI daily interactions before any international rollout. Wall Street carries 55 Buy or Strong Buy ratings and zero Sells on the stock, with a consensus 12-month price target of $755.28, and analysts model $305.9 billion in 2027 revenue and $33.94 in 2027 EPS. Shares trade near $676, roughly 20x forward earnings against the S&P 500's ~22x, and reaching $1,000 would require about a 48% gain and a forward multiple near 29x on 2027 consensus. More than 1 million businesses use Meta AI agents weekly, with early customer Movida reporting a 44% increase in daily bookings, while new ranking models drove an 8.3% increase in ad clicks and a 15.7% uplift in conversions on Facebook. Meta reported 27.96% year-over-year revenue growth in Q2 with ad pricing up 12%, and CEO Mark Zuckerberg said the company is fielding offers for compute at a significant premium over what it paid.

Amazon Web Services posted $42.23 billion in revenue, growing 37% year over year, the segment's fastest growth in 18 quarters, with a 39.4% operating margin and a $496 billion contracted backlog. Amazon's chips and AI businesses each eclipsed run rates of more than $25 billion in the second quarter, both growing at triple-digit percentages year over year, while 98% of Amazon's top 1,000 EC2 customers use Graviton and Anthropic and OpenAI have made multi-year, multi-gigawatt commitments to Trainium. Q2 operating income landed at $27.46 billion, up 43.2% year over year, and advertising is a $70 billion-plus trailing-twelve-month business growing 26%. Capex reached $54.21 billion in a single quarter, up 68.4% year over year, and free cash flow swung to negative $7.6 billion on a trailing-twelve-month basis, though most AI capacity is contracted for at least five-year terms. Since Amazon reported Q2 on July 30, 2026, AMZN moved from $230.08 to $251.19, while SPY went from $747.03 to $762.70 and QQQ went from $687.99 to $716.92. Andy Jassy said AWS could become a few hundred billion dollar revenue business and now believes it will be at least double that and very possibly a trillion dollar annual revenue business in time.

Micron Technology is trading at roughly 6x forward earnings against a consensus analyst price target of $1,513.11, implying about 59% upside from its current $953.49 share price. The Boise-based memory maker, the only U.S.-headquartered producer of DRAM and NAND at scale, has signed 16 multi-year take-or-pay contracts covering roughly 25% of revenue, backed by $22 billion in cash deposits and letters of credit, with remaining performance obligations near $100 billion. CEO Sanjay Mehrotra said AI has elevated the value of memory, and management expects high-bandwidth memory supply tightness to persist beyond calendar 2027, with data-center DRAM and NAND bit shipments projected to more than double from two years ago. Micron has already shipped over $1 billion in HBM4 revenue, with the 12-high volume ramp tracking twice as fast as HBM3E 12-high, and fiscal fourth-quarter guidance calls for record revenue of $50 billion plus or minus $1 billion and non-GAAP earnings per share of $31 plus or minus $1 at a gross margin near 86%. The stock is up 234.28% year-to-date and 501.33% over the past year, and beginning December 9, 2026, Micron plans to increase capital returns and eventually return 100% of excess cash to shareholders.

Nvidia CEO Jensen Huang said the company expects to sell roughly twice as many chips next year as it does this year, calling it one of his strongest signals yet that AI demand remains far from exhausted. Speaking to reporters Thursday at a summit in Scotland with King Charles III, Huang said the reason is AI's contribution to different industries and economies, and that people in almost every country Nvidia operates in want to invest in AI. The forecast adds to Nvidia's increasingly aggressive outlook: the company recently said it expects roughly 70% growth in the fiscal year ending January 2028, which would put annual revenue around $673 billion. Nvidia does not disclose total chip unit shipments, making Huang's volume forecast difficult to translate directly into revenue, though its most important products include data-center GPUs, particularly Blackwell and next-generation Rubin systems; Huang said last fall that Nvidia had shipped about 6 million Blackwell GPUs over four quarters. The comments come as investors debate whether hyperscaler spending and model-training demand can keep growing at the breakneck pace seen during the first phase of the AI buildout, with the key question being not simply whether Nvidia can double chip volumes but what that mix looks like, since higher shipments of premium Blackwell and Rubin systems could support enormous revenue growth while greater unit volume at lower average selling prices would produce a different margin outcome.

Bank of America forecasts the global semiconductor market will nearly double to $3.2 trillion by 2030 from $1.7 trillion in 2026, citing continued strength in AI infrastructure, memory and data-center demand. Within that total, BofA expects memory sales to climb from $937 billion in 2026 to $1.8 trillion by 2030, core semiconductor sales to rise from $739 billion to $1.35 trillion, and server-related sales to increase from $359 billion to $848 billion. The bank also projects wafer-fabrication equipment spending will more than double to $359.8 billion by 2030 from $155.9 billion in 2026, with 2027 largely fully booked and contracted across compute, networking and memory vendors and supply expected to stay tight in 2028. The Philadelphia Semiconductor Index gained around 3.14% on Thursday after five negative sessions, though it remains up about 91.19% over the past year and 59.14% year to date. BofA flagged a potential slowdown in hyperscalers' AI spending as the biggest risk to its outlook, even as recent calls for a more measured pace of AI development from figures including Dario Amodei, Sam Altman and Elon Musk have weighed on the broader AI trade.

ASML Holding N.V. is riding strong momentum in its Installed Base Management business, which generated 2.8 billion euros in sales in the second quarter, up 31.8% year over year and about €300 million above management's expectations. The business benefits from the large and expanding fleet of ASML systems already operating at semiconductor manufacturers, and continued expansion of the EUV installed base is increasing the company's overall service opportunity. Demand is being driven not only by fleet expansion but also by software-led upgrades that improve productivity without significant machine time or production disruption. ASML expects to ship about 65 Low-NA EUV systems in 2026, with EUV revenue projected to rise roughly 45%, and management expects the combination of upgrade and service revenue to drive more than 30% growth in Installed Base Management sales in 2026. Sustaining that pace will depend on continued semiconductor investment and customer demand. Elsewhere in the semiconductor space, Monolithic Power Systems is gaining content across AI infrastructure and has extended its capacity goal beyond $6 billion, while Applied Materials delivered a record third quarter on the rapid global build-out of AI infrastructure.

United Microelectronics Corporation shares have soared 250.1% over the past year, far outpacing the industry's 25.8% gain, as improving foundry demand and AI exposure drive optimism. The company's 22/28-nanometer nodes contributed 37% of second-quarter revenues, with 22nm sales hitting another record, and management expects third-quarter wafer shipments to rise by a high single-digit percentage sequentially with capacity utilization exceeding 90%. UMC expects AI-related revenues to approach $300 million in 2026 and exceed $1 billion within three years, and it recently began mass-production delivery of silicon photonics wafers from its 12-inch Singapore facility, with plans to open its own 12-inch silicon photonics platform to broader customers in 2027. Its collaboration with Intel on a 12nm process platform is expected to begin production in Arizona in 2027. The 2026 earnings estimate now stands at $1.23 per share, up 75.7% over the past 60 days, and the stock carries a Zacks Rank #1 (Strong Buy).

Xeal launched Laitent, which it calls the world's first edge inference compute network using idle EV charging capacity, tapping more than 200MW of permitted, installed electrical infrastructure across 1,600+ properties. Xeal, a member of NVIDIA Inception, plans to deploy over 100,000 NVIDIA GPUs alongside EV charging infrastructure, and has secured partnerships with Rafay Systems for AI infrastructure orchestration, Spectrum Business for dedicated enterprise-grade fiber, dozens of real estate and property managers, and a Tier 1 inference provider for up to 5MW of compute. The first Laitent Pod will be brought online with partner JVM Realty by the end of 2026. Each Laitent Pod is about the size of one parking space, contains up to 48 NVIDIA Hopper or Blackwell Ultra GPUs, requires no water hookup, and runs quiet at less than 65 decibels, offering sub-20ms latency in metro areas. Xeal said EV charging sites typically operate at less than 10% of permitted capacity, and it taps the remaining 90% for compute, with property owners able to add as much as $1m in property value for little-to-no upfront investment. Looking beyond the initial 200MW of installed charging capacity, Xeal plans to unlock over 1GW of existing headroom across real estate and EV charging deployments.

Nvidia CEO Jensen Huang said Thursday he expects the company to sell twice as many chips in the next year as it did this year, a signal to investors that demand has not yet peaked. Huang attributed the outlook to AI investment spreading across industries, economies and countries. Management said consumer expectations point to another doubling of growth next year, and Nvidia projected around 70% revenue growth for the fiscal year ending January 2028, bringing revenue to almost $673 billion. Management classified the outlook as supply tight, implying Nvidia's capacity to ship chips could remain one of the largest impediments to growth. Last year, Huang said the company shipped around 6 million Blackwell GPUs in four quarters, and Rubin is now coming up as the next big part of the product cycle.

GMKtec officially unveiled the EVO-X5 Pro, its flagship desktop AI supercomputer, at its global launch event at IFA 2026 in Berlin, with the system launching on September 28. Powered by AMD's Ryzen AI Max+ PRO 495 processor, the EVO-X5 Pro is described as the world's first desktop system capable of running a 300-billion-parameter large language model entirely offline. The processor features 16 Zen 5 CPU cores, 32 threads, boost speeds up to 5.2 GHz, and an XDNA 2 NPU delivering up to 55 TOPS of AI performance, while the system supports up to 192GB of LPDDR5X-8533 unified memory with 273 GB/s of memory bandwidth and up to 160GB dynamically allocated as graphics memory. The EVO-X5 Pro received two honors at the exhibition, the PCWELT Best of IFA 2026 SmartZone award and a BEST OF IFA 2026 Innovation Award. Michael Nordquist, Vice President of Client Business at AMD, joined the launch event, and GMKtec said the two companies plan to deepen cooperation across the ROCm software ecosystem, AI software optimization, and the broader Agentic PC platform.

Taiwan Semiconductor Manufacturing has moved its 2nm process into commercial production and raised its capital expenditure targets to support the rollout. The chipmaker said the higher spending is preparation for an intensifying global semiconductor arms race among leading foundries and chip designers. The accelerated 2nm commercialization keeps TSMC aligned with chip designers that need leading-edge capacity for AI and high-performance computing, and the bigger budget gives it more room to secure large multi-year supply commitments from major customers. One early indicator to watch is how quickly 2nm and related advanced nodes contribute a rising share of wafer revenue as new lines ramp up, including whether customer wins such as the MediaTek 2nm decision broaden into multiple large-volume programs that keep those advanced fabs near targeted utilization levels from 2026 onward.

The State Department approved a potential $24.3B sale of 48 Lockheed Martin F-35 Lightning II aircraft to Saudi Arabia, along with 49 Pratt & Whitney F135-PW-100 engines and related equipment, a deal that still requires congressional approval and would strengthen Saudi Arabia's air defense and interoperability with U.S. and allied forces. Separately, Bloomberg reported that the U.S. could postpone announcing new tariffs on China and other countries until after President Donald Trump meets Chinese President Xi Jinping next week; a previously planned trade report was expected to recommend a 7.5% tariff on Chinese goods, while the final rate could potentially exceed the 20% level agreed during earlier talks. Nvidia CEO Jensen Huang said the company expects to sell roughly twice as many chips next year as this year as demand for AI infrastructure expands. China's ChangXin Memory Technologies is preparing to enter the NAND flash-memory market, with Reuters reporting that CXMT plans an R&D production line at its new Beijing plant and has established a research institute in the city, broadening its focus beyond DRAM. OpenAI has hired Brian McCarthy from SpaceX as vice president of worldwide sales, Fortune reported; McCarthy joined SpaceX through its acquisition of Cursor, where he led global revenue and field operations, and previously held senior sales roles at Rubrik, ThoughtSpot, AppDynamics and Qlik.

Nvidia CEO Jensen Huang said demand for artificial intelligence computing is accelerating, describing the resulting investment cycle as a flywheel that is "really, really flying." Speaking with CNBC's Jim Cramer at Dreamforce in San Francisco, Huang estimated that a 1-gigawatt Nvidia AI factory costs about $50 billion to $60 billion to build but can generate roughly $50 billion in annual rental revenue, implying a return on invested capital of about one year. He said token generation has increased 25-fold in less than a year, with open models now accounting for nearly 70% of generated tokens, up from about 30%. Huang also said Nvidia Grace Blackwell systems now rent for about $16 per GPU hour, versus roughly $5 under some contracts signed a year ago, and he expects AI testing to become a third major category of infrastructure alongside training and inference, requiring additional data centers.

Salesforce and Nvidia unveiled Koa, a domain-specific reasoning model engineered for Salesforce's Agentforce ecosystem, at Salesforce's annual Dreamforce event. Salesforce CEO Marc Benioff praised Nvidia as "exquisite," saying the AI chip leader has the product for the moment and that the industry has never seen anything quite like it. By hosting Koa directly on Salesforce's private infrastructure, enterprise clients maintain full control over model weights and sensitive data while achieving lower token inference costs. Nvidia CEO Jensen Huang joined Benioff on stage at the Dreamforce keynote to detail the partnership, which underscores a broader move toward domain-tailored AI agents. Guggenheim analyst John Difucci said the decision to create a proprietary model is interesting at this juncture in Salesforce's AI journey, as many application vendors have postured away from the model race for multiple years, and he sees the announcement playing into more task-specific and open-source models becoming a key theme for enterprise deployments. Stifel analyst J. Parker Lane said the announcements and the participation of Jensen Huang, Dario Amodei, and Sam Altman at Dreamforce underscore the open approach the company is taking on both the model and UI layers as it leans in on its data and context layers as its core moat in the Agentic Era.