AI Server OEM & System Integration

Dell Technologies has joined a planned $15 billion AI data center initiative in Japan aimed at supporting large-scale AI workloads by pairing new data facilities with dedicated power supply capacity. The company also introduced rugged laptops with on-device AI features and added security aimed at field and mission-critical use cases. Dell's participation plugs directly into its existing hardware, software and support services business, tying its gear more tightly to power-hungry enterprise workloads and edge environments. The move fits a narrative centered on converting a US$95b AI backlog and a large pipeline into deployed systems, while the rugged laptops push AI closer to field workers in sectors such as public safety and energy. Dell still relies heavily on hardware in segments where HP and Lenovo also compete on price and features.

Dell Technologies has booked more than $130 billion of AI server orders over the past year and ended its latest quarter with a $95 billion AI backlog, with its AI customer base expanding to more than 6,500 customers. The company says some AI engagements require more than 50 unique designs as customers optimize systems around workload performance, power, cooling and data-center constraints. Dell's stock has risen more than fourfold since the end of February and trades at just 21.3 times forward earnings, with revenue expected to surge 69% this year before growing another 20% next year. Hedge fund sentiment strengthened in the second quarter, as the number of funds in Insider Monkey's database holding the stock rose from 72 to 77 and the value of those positions more than doubled from about $1.7 billion to $3.6 billion. Dell acknowledges component costs are rising, particularly for memory and storage, and that supply remains constrained across much of the AI infrastructure chain.

Super Micro has started shipping server racks built for NVIDIA's next-generation Vera Rubin AI architecture, sending its shares up 4.3% in the afternoon session. The start of shipments reinforces the server solutions provider's role in the AI infrastructure supply chain, as Vera Rubin-ready racks house and interconnect the high-density systems data centers need for advanced AI workloads. The stock was trading at $43.69, up 4.4% from the previous close. Super Micro is up 41.1% since the beginning of the year, but at $43.69 per share it remains 25.6% below its 52-week high of $58.68 from October 2025. The company's shares are extremely volatile, having logged 69 moves greater than 5% over the last year.

Hewlett Packard Enterprise received a US$1.2 billion order from Vultr for its new AMD Helios AI servers, marking the first commercial deployment of the Helios racks that combine AMD chips with HPE Networking hardware. The deal is an early revenue milestone for HPE's integrated AI data center platform as Helios moves from launch to customer rollout, and it gives the company a concrete reference customer for model training and inferencing workloads. The US$1.2 billion Vultr order is only one piece of the broader Hewlett Packard Enterprise story, and the company's Helios rollout plugs into its wider role supplying infrastructure for AI workloads at cloud providers and enterprises worldwide. The contract lines up with HPE's existing narrative, which leans on AI infrastructure, Juniper networking and higher margin recurring services as key catalysts. Investors will watch whether HPE converts this first Helios deployment into a visible pipeline of similar rack scale AI deals and repeat orders from Vultr, along with disclosed Helios contract wins and networking order intake linked to AI racks.

CoreWeave announced availability of the NVIDIA Vera Rubin NVL72, with Cognition as its first production customer, alongside new support for the NVIDIA Vera CPU. Cognition, which uses CoreWeave for training, reinforcement learning and inference, reported that the Vera Rubin NVL72 delivered up to 4.8x higher total token throughput for SWE-2 inference workloads versus a GB200 NVL72 baseline and 3.8x higher output-token throughput for reinforcement-learning workloads. CoreWeave said a Vera rack can contain 128 CPUs and 11,264 cores, theoretically supporting more than 11,000 concurrent isolated environments, and that testing showed more than three times faster agent sandbox startup times compared with an x86 CPU. The company will offer Vera on bare metal using the same operating model and economics as the rest of its infrastructure, aiming to monetize CPU-intensive infrastructure alongside accelerator hours. CoreWeave remains heavily dependent on NVIDIA's technology roadmap and faces competition from hyperscalers and specialized GPU clouds including Microsoft Azure and Nebius Group N.V., which closed four deals in the quarter averaging more than $1 billion each and plans roughly £1.7 billion in U.K. AI compute expansion expected to deliver 65 MW when fully operational in 2027.

JERA, Japan's major power producer, announced a partnership with Dell Technologies and RHAELM, a UK-based AI infrastructure developer, to develop a Hyperscale AI Data Center project in Chiba Prefecture near Tokyo, valued at approximately 15 billion dollars, with a power capacity of about 400 megawatts. It is positioned as one of the largest single-site AI infrastructure projects in Japan. Meanwhile, Apollo Global Management plans to join as RHAELM's investment and financial partner. A key point of the project is that it will be developed on land adjacent to JERA's Chiba thermal power plant, with JERA providing the site and 400 megawatts of power under a long-term agreement of 15 to 25 years, while RHAELM is responsible for development, construction, management, and financing. Dell will supply rack-scale AI infrastructure. The project is scheduled to come online in phases starting in 2028 and is expected to use the full 400 megawatts of power by 2029. The three companies also aim to create a model that expands to other JERA power plant sites during the 2030s, with the goal of developing total data center capacity at the multi-gigawatt level across Japan.

France's state-owned supercomputer giant Bull restarted its plant in Angers, western France, on the first of the month after completing an equipment upgrade. The facility has doubled its production capacity from six racks a month to twelve, responding to European demand for stronger computing infrastructure for artificial intelligence. The expansion involved an investment of 80 million euros, and the company says capacity can be raised to 24 racks a month by 2027 depending on demand. The European Union plans to invest 7 billion euros between 2021 and 2027 in building out a supercomputer network, and according to Bull, the Angers plant is the only factory in Europe dedicated to manufacturing supercomputers. Chief Executive Emmanuel Le Roux said the company booked a record number of orders this year, winning 15 of the 18 tenders held by EuroHPC, which is jointly funded by the EU and its member states.

Palantir Technologies and Armada announced a partnership to deliver sovereign AI on infrastructure manufactured in the United States and allied nations, with Palantir naming Armada its Inaugural Certified Modular Data Center Partner. The joint offering combines Palantir's Sovereign AI Operating System with Armada's infrastructure layer: Galleon modular data centers, the Armada Platform that keeps every deployment under the customer's control, and the Sovereign AI Grid that connects deployments into a resilient distributed system. Palantir will validate and integrate its Sovereign AI Operating System, built on AIP, Ontology, Foundry and Apollo, on Armada's Galleon modular data centers, extending the Palantir Sovereign AI OS Reference Architecture developed with NVIDIA to a ruggedized, modular data center. The Armada Platform distills the latest open-source models, including NVIDIA Nemotron, for fine-tuning and inference on infrastructure the customer owns and controls, without reliance on any external cloud, including any operated by Armada, and can operate fully air-gapped where the mission requires it. Enterprises and governments gain the ability to run open-weight models on compute, data, and physical infrastructure they own and control outright, deployed in months rather than years.

Advanced Micro Devices has entered into a definitive agreement to acquire World Labs, an artificial intelligence model and research company led by AI pioneer Dr. Fei-Fei Li, in an all-stock transaction valued at approximately $8.2 billion that is expected to close by the end of 2026. World Labs specializes in spatial-intelligence models that generate, reconstruct and simulate interactive 3D environments from text, images and video, and is also developing technologies for robotic learning and simulation. AMD said the deal will give it deeper visibility into how emerging AI workloads are evolving across reasoning, robotics, simulation and physical AI, helping it tailor future hardware, software and system architectures, and strengthening its position in physical AI and robotics alongside its Ryzen AI Embedded X100 processors and Kria AI Robotics Developer Platform. The acquisition fits AMD's shift from selling individual chips toward becoming a full-system AI infrastructure provider, complementing its Instinct, EPYC, Pensando and ROCm ecosystem and its recently launched Helios rack-scale AI platform, which AMD says offers up to 15% higher inference throughput at the same rack power and up to 30% more tokens per dollar than competing solutions. The deal comes as AMD's Data Center revenues surged 107% year over year to $6.7 billion in the second quarter of 2026, with management expecting Data Center segment revenues to more than double year over year in 2027 and estimating the data-center AI accelerator market will reach roughly $1.4 trillion by 2030.

CoreWeave Inc. announced it will expand its compute portfolio with the NVIDIA Vera CPU, the first CPU designed for AI agents, at its Fully Connected AI cloud conference in San Francisco. The company said rack-scale Vera at CoreWeave puts 128 CPUs and 11,264 cores in a single rack, with BlueField-4 DPUs and Spectrum-X Ethernet switching, providing room for more than 11,000 concurrent agent environments. In testing, CoreWeave achieved more than 3x faster agent sandbox startup times on NVIDIA Vera CPU compared to an x86 CPU. Chen Goldberg, executive vice president of product and engineering at CoreWeave, said general-purpose infrastructure bottlenecks agentic AI and that Vera is the first CPU explicitly designed to accelerate it. CoreWeave said NVIDIA Vera runs bare metal under the same platform, consumption models and economics as the rest of its fleet, and is natively enabled through products such as CoreWeave Sandboxes.

Lenovo announced Lenovo AI Express, a new quick-start program developed with NVIDIA that accelerates the path to production AI by cutting order-to-ship times to as few as 15 business days for eligible configurations. The program offers three validated quick-start configurations of the Lenovo Hybrid AI Factory with NVIDIA: a Small tier shipping from 15 days for inferencing for tens of users at 30+ TPS with model sizes from 7B to 70B, built on Lenovo ThinkSystem SR650a V4 with two NVIDIA RTX 6000 PRO Blackwell Server Edition GPUs; a Medium tier from 20 days for hundreds of users with model sizes from 70B to 400B, built on Lenovo ThinkSystem SR675 V3 with eight NVIDIA RTX 6000 PRO Blackwell Server Edition GPUs; and a Large tier from 25 days for thousands of users with up to a trillion parameters, built on Lenovo ThinkSystem SR680a V4 with NVIDIA HGX B300. Lenovo cited its 2026 CIO Playbook finding that organizations expect an average $2.79 return for every $1 invested in scaled AI execution, with 93% of enterprise respondents anticipating positive returns. Ashley Gorakhpurwalla, President of Lenovo's Infrastructure Solutions Group, said the program gives customers an accelerated path to deploying AI infrastructure while mitigating the risk of overbuilding or costly delays, and NVIDIA vice president of enterprise platforms Chris Marriott said the ready-to-ship solutions combine Lenovo's validated systems with NVIDIA accelerated computing, networking and AI software. The configurations support the latest AMD and Intel CPUs and can be extended with Red Hat AI Factory with NVIDIA or NVIDIA AI Enterprise software, plus Veeam Kasten for AI data resilience, and Lenovo is expanding access through its global partner ecosystem via the Lenovo 360 framework.

DeepSeek is extending its AI software to Huawei's Ascend processors, giving the Chinese AI developer another computing platform as it works to build an alternative to Nvidia's ecosystem. The company has made programming infrastructure for Ascend available as open-source software, including tools for computation and communication, along with TileLang, a programming language designed to help developers build software for AI chips. DeepSeek and Huawei have also worked together on a computing system using 128 Ascend 950 processors, with the two companies optimizing the system's processing and chip-to-chip communication, Reuters reported. The development comes after Huawei recently introduced a new generation of AI processors and related computing systems, and Huawei has said its AI infrastructure is expected to see broader use for model training next year. For Nvidia, the development adds to efforts by Chinese technology companies to develop local alternatives around both hardware and software.

BOCI has initiated coverage of Lenovo Group with a bullish outlook on the stock. BOCI noted that the momentum of Lenovo's Infrastructure Solutions Group (ISG) division, which is driving the company's growth, underscores solid corporate demand for full-stack solutions. It said the ISG division's pre-tax margin is approaching double digits and will become a key driver of profit growth. In the first quarter of fiscal 2027, covering April to June 2026, the company's AI-related revenue rose 60 percent year on year to about 9.3 billion US dollars, accounting for 35 percent of total revenue. In its partnership with Nvidia, the first batch of GB300 NVL72 rack products has already shipped, and the Rubin platform is scheduled for mass production in the second half of 2026. BOCI maintained its target price based on an estimated price-to-earnings ratio of 17 times for fiscal 2028, ending March 2028. It had previously valued the company at a P/E of 10 times as a pure hardware business, but raised its valuation as the ISG business grew. It cited rising memory prices weighing on PC and smartphone demand, server delivery delays, tariff issues and geopolitical risks as potential risk factors.

Aoni Electronics announced after market close on September 30 that its wholly owned subsidiary Aoni Intelligent Technology (Zhongshan) Co., Ltd. signed a procurement contract with Company C to purchase GPU computing card products, with a total contract value of 1.821 billion yuan including tax. This is the third major computing power procurement order disclosed by the company recently: on September 22, it signed a 1.67 billion yuan contract with Company A, and on September 24, it signed a contract with Company B worth 232 million US dollars, equivalent to about 1.556 billion yuan. The announcement said that the counterparty names and some contract terms are trade secrets and have been exempted from disclosure in accordance with regulations. Both Company B and Company C are operating normally with good credit standing and have no related-party relationship with the company. The company stated that the contract is a major contract in its daily operations and does not require review by the board of directors or shareholders' meeting. If smoothly fulfilled, it is expected to have a positive impact on the operating results for the performance year. In the first half of this year, Aoni Electronics achieved operating revenue of 907 million yuan, up 215.44 percent year on year, and net profit attributable to the parent company of 16.8146 million yuan, turning from loss to profit compared with the same period last year. Among this, revenue from high-performance computing equipment business was 508 million yuan, up 1,625.53 percent year on year.

NetApp and Supermicro announced a strategic collaboration to deliver jointly validated AI infrastructure solutions spanning large-scale AI factories, enterprise deployments, neoclouds, and sovereign AI initiatives. The joint offering combines Supermicro's AI-optimized compute and rack-scale systems with the NetApp Platform, and the companies say it will help customers accelerate deployment, improve GPU utilization, strengthen data governance, and reduce operational complexity. As an initial implementation, NetApp and Supermicro will deliver NetApp Novus Data Director software running on qualified Supermicro servers with NetApp AFF A90 systems as the high-performance ONTAP data layer, a first step in a broader effort to co-engineer integrated infrastructure for enterprise AI, neocloud, sovereign AI, and large-scale AI factory environments. The two companies are also collaborating on NetApp AIPod with Supermicro, a NetApp-validated converged infrastructure architecture based on NVIDIA Reference Architecture guidance that supports AI training, inference, and agentic workloads from dozens to hundreds of GPUs. Alvaro Celis, Senior Vice President and Chief Commercial, Partner and Ecosystem Officer at NetApp, said the combination gives customers secure, reliable AI solutions they can deploy quickly, while Supermicro Chief Revenue Officer Matthew Thauberger said it offers a validated path where infrastructure and data are designed to work together from the start.

Uniserve Communications Corporation has signed a five-year AI compute services agreement with an enterprise client for two NVIDIA B300 GPU-based servers to be deployed at its Vancouver data centre. The company said it has received an initial customer prepayment and is proceeding with equipment procurement. At the minimum rate, payments over the full 60-month service period would total approximately US$3.15 million before taxes, though actual payments could be higher if the revenue-sharing provision exceeds the monthly minimum. Chairman and Chief Executive Officer Gautam Lohia called the deal the first of many engagements of this nature as Uniserve moves to providing AI ready infrastructure. Equipment delivery, installation and service activation are anticipated to occur before year end.

Kiatnakin Phatra Securities, or KKPS, has raised its target price for Hana Microelectronics, or HANA, to 61 baht from 42 baht, while maintaining a buy recommendation, based on a P/E level about 1.5 standard deviations above the historical average, up from the previous 1 standard deviation. The research team raised its normal profit forecasts for HANA in 2026 and 2027 by 8% and 31% respectively, expecting profit growth of about 20% in 2026 and 53% in 2027 after lifting revenue and gross margin assumptions to reflect opportunities from new businesses. Normalized net profit is expected at about 999 million baht in 2026, rising to 1.533 billion baht in 2027 and 1.78 billion baht in 2028, while earnings per share, or EPS, is forecast at 1.13 baht, 1.73 baht and 2.01 baht respectively, up from 1.05 baht, 1.32 baht and 1.46 baht in the same periods. On AI business opportunities, the research team said clarity has improved for two of HANA's three AI projects, with the Phononic TEC and High-density PCBA projects entering the capacity ramp-up process and expected to begin generating significant revenue from early 2027. The two projects are estimated to generate combined revenue of about 2.5 to 3.5 billion baht in 2027, or roughly 12% to 17% of HANA's revenue in 2025. Meanwhile, the OSAT plant in Ayutthaya province has seen its utilization rate rise to 62% from 55%, and combined losses from Power Master Semiconductor and HANA Technologies, Inc. narrowed to about 116 million baht in the second quarter of 2026 from about 388 million baht in the fourth quarter of 2025. Opportunities from PMS's data center and solid state transformer businesses have not yet been included in the main forecasts, as further clarity is needed and a clearer picture is expected in the fourth quarter of 2027.

Xingyun Technology disclosed two major sales contract announcements after market close on September 29. Its wholly owned subsidiary Shenzhen Xingyun Holdings Co., Ltd. signed a sales contract for computing servers with a VF client, with a total amount of approximately 239 million yuan. Its wholly owned overseas sub-subsidiary Syncloud International Pte. Ltd. signed an international trade sales framework agreement and two sales orders with a VG client, with a combined amount of approximately 573 million yuan. The two contracts together total more than 800 million yuan. The Shenzhen Xingyun contract is worth approximately 239.2 million yuan and has been delivered and accepted. Because the contract amount exceeds 50 percent of the company's most recently audited main business revenue and the absolute amount exceeds 100 million yuan, it constitutes a major contract. The Singapore sub-subsidiary's framework agreement is valid until August 6, 2027, with no specific contract amount stipulated. The two signed sales orders total approximately 572.7 million yuan and have also been delivered and accepted. The announcement cautioned that the contracts involve large-scale equipment procurement spending, with risks of overly rapid asset expansion and insufficient funding. The relevant funds will mainly come from the company's own capital and financing from financial institutions.

Kesi Technology announced on September 29 that the company recently signed a sales contract with Shenzhen Yiweier Computer Technology Co., Ltd. The contract amount is 160 million yuan including tax, for the Kesi AI large-model high-computing-power server project. The announcement shows that the hardware part of the high-computing-power servers consists of computing products from an original equipment manufacturer, and the products Kesi Technology delivers to Yiweier are integrated computing all-in-one machines that have already completed deployment of Kesi Technology's proprietary AI large-model software and system debugging.

The Singapore Economic Development Board (EDB) said on September 28 that Singapore's manufacturing output in August surged 15.4% year on year, driven by broad-based growth across most industry clusters. On a seasonally adjusted basis, manufacturing output in August fell 0.5% month on month. The report said output rose year on year in nearly every sector except chemicals. The precision engineering cluster soared 33.9%, led mainly by the production of semiconductor machinery and equipment. The electronics cluster rose 28.5% on the strength of server and related products, semiconductors, and data storage equipment, amid strong demand related to artificial intelligence. The transport engineering cluster expanded 9.5%, led by land vehicles and aircraft. The general manufacturing industries cluster grew 1.5%, and biomedical manufacturing edged up 0.4%. Chemicals was the only cluster to decline, with output falling 12.7%, weighed down by the petroleum and petrochemical segments.

Morgan Stanley raised its price target on TD Synnex Corp to $359 from $334, reiterating an Overweight rating and naming the technology distributor its most favored enterprise hardware name. The call followed fiscal third-quarter results in which revenue and earnings per share beat consensus estimates by 13%-21%, while fourth-quarter guidance came in 12%-25% above Street expectations. Morgan Stanley lifted its fiscal 2027 and 2028 earnings estimates by 12%-15%, citing stronger growth and operating leverage. Within the company, Distribution gross billings rose 27% year-on-year, while gross billings at Hyve Solutions, TD Synnex's data-centre infrastructure unit, jumped 117%, as enterprise AI demand shifted from experimentation toward production. Operating expenses fell to 3.63% of revenue, the lowest level in more than five years, and Morgan Stanley expects free cash flow to turn positive in the fourth quarter as investments in Hyve's new customer programs ease.

Microsoft plans to expand its global data center capacity to more than 38 gigawatts by 2032, more than tripling its current roughly 12-gigawatt footprint, according to Bloomberg News reporting cited by Reuters and attributed to people familiar with the plans. Only about 2 gigawatts of Microsoft's current capacity is dedicated to AI-specific chips, a share the company expects to grow to roughly one-third of the much larger 38-gigawatt total by 2032, though Microsoft has not confirmed the target in any public filing or statement. The reported roadmap rests on strong cloud demand: Azure and other cloud services revenue rose 43% year over year in fiscal Q4, Microsoft Cloud revenue reached $59.3 billion for the quarter and $214.4 billion for fiscal 2026, and commercial remaining performance obligations climbed 84% year over year to $678 billion, with roughly 30% expected to be recognized as revenue over the next 12 months. Funding the buildout will be costly, with Microsoft expecting more than $50 billion of capital expenditures in fiscal Q1 2027 and about $175 billion during calendar 2026 under its broader capex definition, after roughly $145 billion in fiscal 2026. Power and political hurdles could slow the plan, as Texas temporarily halted approvals for new data center grid connections and New York imposed a moratorium on large new data centers, even as Microsoft signed a 20-year agreement with Constellation Energy tied to the restart of the 835-megawatt Three Mile Island Unit 1.

Fengzhushou announced on the evening of September 28 that it has signed major computing power business contracts totaling 9.677 billion yuan. Among them, the company signed a Computing Power Service Agreement with Company D, with a contract value of 5.717 billion yuan, tax inclusive, excluding cabinet service fees. Cabinet services will be settled separately based on actual usage, and the service cooperation period is 60 months. At the same time, the company signed a Computing Power Server Procurement Agreement with Company C, with a total contract value of 3.96 billion yuan, tax inclusive, under which Company C will deliver all products in batches. Fengzhushou stated that the procurement project funds will be raised through a combination of its own funds, financial leasing, and supporting working capital loans, with own funds expected to account for 5% to 20% and financial leasing financing expected to account for 80% to 95%. The company said this comprehensive computing power service project is a routine business measure based on its main business and aimed at advancing the implementation of existing operations, which will help improve its computing power service system and enhance market competitiveness.

Analysts at Kasikorn Securities remain positive on the electronics group in the fourth quarter of 2026, expecting second-half 2026 earnings to accelerate notably, led by DELTA benefiting from capacity utilization and recovering margins, HANA from the ramp-up of AI-related products, and KCE from higher PCB sales volumes and upward ASP adjustments. AI and data centers are the main structural growth drivers underpinning continued demand for power electronics, advanced PCB, PCBA, and semiconductor components, supported by high demand from hyperscalers, gradually easing component constraints, an improved product mix, and a weaker baht, which is a positive factor for exporters' gross margins. However, risks remain from high raw material costs, particularly copper, fiberglass, copper clad laminate, and various electronic components, as well as the uneven recovery of end markets unrelated to AI, especially the automotive and EV segments. Strategically, the analysts recommend gradually accumulating electronics stocks on market dips, with DELTA rated Buy at a 300 baht target price, HANA rated Buy at a 56 baht target price, and KCE rated Hold at a 60 baht target price.

Jabil shares jumped 4.8% in afternoon trading after Akamai Technologies authorized the electronics manufacturing services provider to buy approximately $1.7 billion of memory components on Akamai's behalf. Akamai disclosed the authorization in a filing, saying the purchase supports capital spending tied to its expanded agreement with Anthropic, which now totals $11.6 billion of contractual commitment over seven years. Akamai said capital spending tied to that commitment is about $5.5 billion, and that 2026 capital spending would rise by about $1.7 billion to pre-purchase components, including memory. Jabil was already Akamai's contract manufacturer under a 2019 agreement, and the filing does not say how much of the $1.7 billion Jabil keeps. After the initial pop, the shares cooled to $319.80, up 2.9% from the previous close.

Dell Technologies shares climbed 4% Friday after Morgan Stanley raised the probability attached to its $756 bull-case scenario following a meeting with Chief Operating Officer Jeff Clarke, while leaving its Neutral rating and $511 target unchanged. The shift centers on whether AI spending represents another hardware replacement cycle or a broader expansion in infrastructure demand. Clarke told the brokerage that AI agents could drive inference-token usage 87 times higher by 2030, potentially creating additional demand for computing capacity. Morgan Stanley's unchanged base-case valuation means the note did not alter its formal recommendation, but the higher probability assigned to the more optimistic scenario gave investors another data point as they assess whether Dell's AI infrastructure growth can persist beyond the current spending cycle.

Advanced Micro Devices crossed the $1 trillion market capitalization mark for the first time on Monday, with shares surging 9.9% amid a broader tech-stock rally. The milestone caps a September in which CEO Lisa Su sat at President Trump's table during the state dinner with Chinese President Xi Jinping, two seats from Trump and next to Elon Musk, and rang the opening bell at the New York Stock Exchange with first lady Melania Trump. AMD stock is up 193% year to date, far outpacing Nvidia's 22% gain. The chipmaker beat Wall Street expectations across the board for the second quarter, posting record non-GAAP earnings per share of $1.66 against consensus estimates of $1.61, with total revenue up 50% year over year to a record $11.5 billion and data center revenue more than doubling to $6.7 billion. AMD projected second-half 2026 server CPU revenue to grow 80% year over year and 70% in 2027, and expects data center revenues to more than double in 2027 with AI GPUs growing well over 100%. The company also unveiled its next-generation Instinct MI450 Series GPUs and 6th Gen EPYC Venice CPUs for its Helios rack-scale AI systems, alongside deals for Microsoft to deploy Helios across Azure AI services beginning in the second half of 2026 and for Anthropic to deploy up to 2 gigawatts of Instinct MI450 GPUs beginning in the first half of 2027, a deal that could be worth tens of billions of dollars over time, with AMD committing to invest up to $5 billion in Anthropic.

Akamai Technologies signed an $11.6 billion, seven-year contract to supply Anthropic with computing power, sending its shares up 21.83% in premarket trading. The deal is the largest in Akamai's history and builds on a $1.8 billion agreement the two companies struck earlier this year, giving the Claude developer access to central processing units. Akamai expects capital spending tied to the contract to total about $5.5 billion, more than six times its entire 2025 capex, with most of it going on servers, chips and networking gear next year. CEO Tom Leighton said revenue from the Anthropic business should reach $150 million to $300 million next year and an annual run rate of about $1.7 billion by 2028. Akamai also issued Anthropic a warrant to buy Series B shares at $111.33 each, convertible into 7.7 million common shares, with a portion equal to about 2% of Akamai's common stock vesting when the computing agreement starts in the second half of next year and the rest vesting over the contract's term. Leighton called the warrant, Akamai's first in a cloud deal, a serious step, and it lands as some investors question circular AI deals in which suppliers and customers invest in each other.

Nvidia Corp. CEO Jensen Huang said the artificial intelligence infrastructure boom will eventually produce more computing capacity than the market needs, but he does not expect that turning point for another two to three years. Speaking on The Ezra Klein Show, Huang said demand and supply will eventually invert and markets will slow into a "period of digestion" that could last six months, nine months or a year, though he added it will not happen next year or in the next couple of years. He also argued that AI labs should not release systems they cannot control. Nvidia reported second-quarter revenue of $96.2 billion, up 106% from a year earlier, while Data Center revenue surged 117% to $89 billion. TrendForce estimates the world's nine largest cloud providers will spend more than $886.7 billion this year, up about 90%, before capital expenditures climb to roughly $1.3 trillion in 2027, and it raised its 2026 AI server shipment growth forecast to nearly 31%. Former Fidelity fund manager George Noble compared the boom to "dot-com and subprime all merged in one," and Reuters reported investors are demanding higher yields on AI-linked debt as hyperscalers prepare for another surge in borrowing. A Polymarket market puts the probability of an AI-industry downturn by Dec. 31 at 10%, resolving "Yes" only if at least three specified stress events occur within 90 days, including Nvidia falling 50% from its all-time high, the SOXX semiconductor ETF falling 40%, or the bankruptcy of OpenAI or Anthropic. Nvidia shares closed Wednesday down 1.5% at $225.51 and were about 1% lower in Thursday premarket trading.

TD Synnex Corp reported record quarterly results with non-GAAP gross billings of $31.8 billion, up 40% year-over-year and above the high end of guidance. Non-GAAP operating income rose 55% to $736 million and non-GAAP EPS climbed 59% to $5.68, also above the high end of guidance, while GAAP operating income rose 68% to $643 million and GAAP EPS rose 89% to $5.18. Within the total, the Distribution segment contributed $24.8 billion in gross billings, up 27% year-over-year with double-digit growth in every region, and the Hyve segment contributed $7 billion, up 117%, with Hyve manufacturing gross billings growing in excess of 130% and supply chain services up in excess of 90%. Hyve's non-GAAP operating margin narrowed to 3.61% from 5.04% a year earlier on lower-margin large AI server programs, and free cash flow was a consumption of approximately $1 billion driven by higher inventory in Hyve's supply chain business and new customer programs. For the fourth quarter, TD Synnex guided to non-GAAP gross billings of approximately $31.9 billion, plus or minus $500 million, revenue of approximately $22.2 billion, plus or minus $400 million, and non-GAAP diluted EPS of approximately $5.90, plus or minus $0.25, up about 54% at the midpoint. The company returned $100 million in share repurchases and $38 million in dividends during the quarter and declared a quarterly dividend of $0.48 per share.