Everyone talks about expensive GPUs. But here's the part people miss — training one AI takes tens of thousands of GPUs running at the same time, and every fraction of a second they have to "talk to each other" with enormous amounts of data. If the link between them is even a little slow, the most expensive GPUs in the world just sit there waiting. This lesson walks through the layer of cables and switches that ties a whole cluster of GPUs into one machine — and the hundred-billion-dollar fight between Nvidia and Broadcom that's heating up on top of it.
AI networking demand locks in years of growth as suppliers sell out and raise targets
▲
Marvell and Nvidia deepen AI networking partnership with $2B investment Nvidia invested $2 billion in Marvell and expanded their partnership across custom chips, silicon photonics and optical interconnect. Marvell will supply NVLink Fusion-compatible networking while Nvidia provides switches and interconnects. This tightens the ecosystem for AI factory networking, boosting demand for high-speed links and custom silicon.
Shows a major new partnership that directly expands the AI networking supply chain and locks in demand.
Marvell raises long-term revenue targets on custom AI and interconnect demand Marvell raised its fiscal 2028 revenue outlook to $20 billion and set a fiscal 2031 target of $70–90 billion, citing custom AI chips and high-speed interconnect. It also lifted its custom chip target for fiscal 2029 to $12 billion. This signals multi-year growth for networking silicon and optics.
Provides concrete long-term growth targets that confirm the scale of AI networking demand.
Arista launches open Ethernet rack-scale portfolio for AI fabrics Arista unveiled an open Ethernet rack-scale portfolio with partners including AMD, Broadcom, Meta and Microsoft. It supports up to 1,024 XPUs and integrates switches, optics and liquid cooling, raising rack capacity to 6.5 Pb/s and cutting data center footprint by nearly 50%. This expands the market for open AI networking gear.
Introduces a major new product that broadens the addressable market for AI networking equipment.
Cisco, Credo and Lumentum report surging AI networking demand and sold-out capacity Cisco guided fiscal 2027 revenue to $72.2–73.4 billion after hyperscaler AI orders hit $9.3 billion, up 4.5x. Credo's retimer revenue jumped 115% year over year. Lumentum's CEO said opto-parts capacity is sold out through 2029. These results confirm demand is outstripping supply across switching, optics and retimers.
Shows broad-based demand strength and supply tightness across key AI networking suppliers.
Northland Capital downgrades Astera Labs and Semtech to Market Perform
Northland Capital Markets downgraded Astera Labs and Semtech to Market Perform on Friday, citing a declining risk-reward. Analyst Gus Richard wrote in a note to clients that he is not confident earnings will be strong enough to lift AI semiconductor companies, adding that connectivity companies are best positioned to outperform but that SMTC and ALAB are above the firm's price targets. Richard said a significant portion of Astera's revenue comes from a limited number of customers, and that the loss or a significant reduction in demand from one or a few top end customers would adversely affect operations and financial condition. For Semtech, he noted the company has built its business via acquisitions, warning that failure to successfully identify, acquire, and integrate a company at a favorable price may adversely impact its financial position and operating results, and that the Sierra Wireless acquisition could wind up being a negative. Richard also said risks continue to increase as growth becomes increasingly dependent on debt and AI models lose pricing power.
Artificial Intelligence › Switching & Networking Silicon/Systems Capital
Semiconductors › Analog, Power & Discrete ▼Capital
ALAB · Capital · Negative Northland Capital downgraded Astera Labs to Market Perform, citing declining risk-reward and that shares are above its price target.
SMTC · Capital · Negative Northland Capital downgraded Semtech to Market Perform, warning the Sierra Wireless acquisition could be a negative and growth increasingly depends on debt.
Tower Semiconductor and NewPhotonics Begin High-Volume Shipments of Optical Engine PICs for AI Interconnect
Tower Semiconductor and NewPhotonics announced high-volume shipments of laser-integrated, serviceable optical engine photonic integrated circuits developed for next-generation AI infrastructure interconnect. The products are built on Tower's PH18DA silicon photonics platform, which uses heterogeneously integrated indium phosphide components to place semiconductor optical amplifiers, modulators, lasers, and photodetectors on a single chip, and the companies say the integration reduces complexity and manufacturing variability while improving yield, reliability and time to market. Volume production today covers data rates from 800G to 1.6T, with 6.4T solutions planned for future volume manufacturing to serve both scale-out and scale-up architectures, and shipment of the 6.4T NPC505 chipsets is anticipated during the first half of 2027. NewPhotonics is the first player across the market to scale up production on Tower's PH18DA InP-on-silicon photonics platform, and Tower faces execution risks in scaling alongside it, along with competition from alternative architectures and other silicon photonics foundries. Tower's stock has rallied 99.97% in 2026, lifting its market capitalization to $25.85 billion, with an operating margin of 19.63%, a net margin of 16.94%, a trailing price-to-earnings ratio of 91.11x, a forward P/E of 38.46x and a return on equity of 9.98%.
TSEMI03.BK · Demand · Positive High-volume shipments of Tower's PH18DA silicon photonics optical engine PICs for AI interconnect mark real product adoption.
NewPhotonics · Demand · Positive NewPhotonics is first to scale volume production of its optical engine PICs on Tower's PH18DA platform, with 6.4T NPC505 chipsets shipping in H1 2027.
TSEMI23.BK · Competition · Negative Article flags execution risks in scaling and competition from alternative architectures and other silicon photonics foundries.
TSEMI80.BK · Capital · Neutral Stock rallied 99.97% in 2026 with rich valuation metrics, but no new company-specific financial event.
Cisco Product Revenue Jumps to $13.459 Billion as Services Stay Flat
Cisco Systems reported fiscal fourth-quarter product revenue of $13.459 billion, up from $10.886 billion a year earlier, while services revenue was nearly unchanged at $3.793 billion versus $3.787 billion. The August 12 results cover the quarter ended July 25. Services costs fell to $1.160 billion from $1.199 billion, lifting calculated gross margin to 69.4% from 68.3% and producing $45 million more gross profit on just $6 million of incremental sales. Product revenue supplied almost the entire $2.579 billion increase in consolidated quarterly sales, and a 10% decline in quarterly product sales would remove $1.35 billion of revenue, which services would need to grow 35.5% from their current base to offset. At the October 8 closing price of $114.89, Cisco traded at 22.6 times the $5.08 midpoint of management's fiscal 2027 adjusted EPS guidance and 28.5 times the $4.03 midpoint of GAAP guidance, while total product orders grew 35%, or 25% excluding hyperscalers.
CSCO · Capital · Positive Cisco's Q4 product revenue jumped to $13.459B from $10.886B, lifting gross margin to 69.4% and gross profit, with product orders up 35%.
Openreach Deploys Ciena's 1.6Tb/s Waveserver Platform in OSEA Rollout
Openreach is deploying Ciena's Waveserver platform, powered by WaveLogic 6 1.6Tb/s coherent technology, as part of its Optical Spectrum Extended Access portfolio. The deployment combines 400G and 800G client traffic over dedicated 1.6Tb/s wavelengths, with a compact 2RU chassis supporting up to 32 400G services, and enables Openreach to introduce national 800G services across the U.K. while providing a pathway to higher speeds. The solution will let Openreach offer wholesale customers tailored, high-capacity network services including connectivity for cable landing stations, data centers and core network nodes. Ciena faces competition from Corning, which announced a long-term partnership with NVIDIA to expand U.S.-based optical connectivity manufacturing capacity by 10 times and increase domestic fiber production capacity by more than 50%, and from Nokia, whose Network Infrastructure unit delivered double-digit growth in second-quarter 2026. Ciena shares have gained 27.3% in the past month, and the Zacks Consensus Estimate for its fiscal 2026 earnings has been revised upward over the past 60 days.
CIEN · Demand · Positive Openreach is deploying Ciena's Waveserver/WaveLogic 6 1.6Tb/s platform in its OSEA rollout, a concrete customer win.
GLW · Competition · Neutral Corning is cited only as a Ciena competitor via its NVIDIA optical-connectivity manufacturing partnership, not as a subject of the news.
NOK · Competition · Neutral Nokia is mentioned only as a competitor whose Network Infrastructure unit posted double-digit Q2 2026 growth.
Lumentum CEO Says AI Parts Sold Out Through Early 2029, Lifting Optical Stocks
Lumentum chief executive officer Michael Hurlston said the company's components are completely sold out through early 2029, sending shares of optical equipment makers higher. In an interview in Tokyo on Friday, Hurlston said Lumentum cannot meet about 70% of demand for some products through next year and will not be able to address 30% of demand for other products through 2028, according to Bloomberg. Just six months ago, Hurlston said Lumentum was on track to sell out its capacity by 2028. In premarket trading, Lumentum gained 3.76%, Applied Optoelectronics rose 4.83%, Fabrinet advanced 2.35%, Coherent climbed 3.15%, and Lightwave Logic added 1.57%. Lumentum, which received a $2B investment from Nvidia earlier this year along with rival Coherent, supplies advanced indium phosphide devices that help enable high-speed cloud computing and data transmission, and analysts on average expect its sales to more than double this year.
Artificial Intelligence › AI Networking & Interconnect ▲Supply
LITE · Demand · Positive CEO said Lumentum's components are completely sold out through early 2029 and it cannot meet 70% of demand for some products.
AAOI · Demand · Positive Rose in premarket as Lumentum's sold-out AI optical component demand signals strong sector demand for optical equipment makers like Applied Optoelectronics.
COHR · Demand · Positive Climbed as Lumentum's sold-out AI parts through early 2029 points to robust demand across optical equipment peers, including Coherent.
FN · Demand · Positive Advanced on Lumentum's sold-out AI component capacity, signaling strong demand for optical manufacturing peers like Fabrinet.
LWLG · Demand · Positive Added 1.57% as Lumentum's sold-out AI optical demand lifted optical equipment stocks including Lightwave Logic.
Guangzhi Technology Secures Order from Global Leading Customer for 6-Inch Indium Phosphide Substrates
Guangzhi Technology stated on an investor interaction platform on October 9 that its downstream customers for indium phosphide substrate business are domestic and overseas optical communication chip and device manufacturers, and batch deliveries have been achieved to date. The company said its domestic customers cover leading optical chip enterprises, and it has shipped in volume to leading optical communication firms overseas. Among these, the 6-inch indium phosphide substrate has secured an order from a certain global leading customer.
Tianfu Communication Responds to Rumors of Overseas Import and Export Policy Restrictions: Optical Device Products Not Affected
Tianfu Communication released an investor relations activity record announcement, responding to recent rumors about overseas import and export policy restrictions. The company stated that the policies issued so far do not impose constraints on optical device products and do not affect its business. The company said its main business is the research, development, production, and sales of various passive and active optical devices. Its sales revenue from the U.S. market accounts for a relatively low proportion, with North American sales revenue making up about 1.28% of total operating revenue in 2025. Based on current operations, the company's orders on hand and customer demand remain robust overall, and it will advance global capacity expansion in an orderly manner according to market demand and customer project timelines. The company remains optimistic about the long-term prospects of the optical interconnect industry, especially the continued growth opportunities in areas such as CPO and NPO. In addition, the company's FAU and ELS products supporting CPO already have a solid foundation in performance, yield, and delivery assurance, and these products have achieved stable mass production and delivery.
Artificial Intelligence › AI Networking & Interconnect ▲Regulation
300394.CS · Regulation · Positive Company states overseas import/export policy restrictions do not affect its optical device products, removing a regulatory overhang on its business.
Arista Unveils Open Ethernet Rack-Scale Portfolio for AI Infrastructure
Arista Networks has unveiled a new open Ethernet rack-scale portfolio designed to support high-density artificial intelligence infrastructure across both scale-up and scale-out environments, developed with AMD, Arm, Broadcom, d-Matrix, Meta, Microsoft and Qualcomm. The company's Etherlink SU-14 architecture extends Ethernet into scale-up environments, supporting up to 144 accelerators in a single rack and up to 1,024 XPUs through cross-rack configurations. For scale-out AI environments, the rack-scale solutions support technologies such as ESUN, NVIDIA's NVLink, NVLink Fusion and PCIe, built on Arista's Netdi foundation using Ultra Ethernet Consortium standards. The rack designs integrate liquid-cooled 7060EX7 switches with up to 102.4 Tbps of capacity per switch, and the company's next-generation optics, including XPO and open CPO/NPO, can raise networking capacity from 1.6 Pb/s to 6.5 Pb/s per rack compared with current OSFP-based switches. Arista faces competition from Cisco Systems, which is expanding its AI infrastructure business through its Secure AI Factory with NVIDIA, and from NVIDIA, which is advancing its Vera Rubin platform and developing Spectrum-6 Ethernet.
ANET · Technology · Positive Arista unveiled a new open Ethernet rack-scale portfolio (Etherlink SU-14, 7060EX7 switches, XPO/CPO optics) for high-density AI infrastructure.
Credo Technology Joins Open CPX MSA Co-Packaged Optics Consortium
Credo Technology Group Holding Ltd. has joined the Open Co-packaged Interface Multi-Source Agreement, or Open CPX MSA, a consortium working to advance interoperable near-package and co-packaged optics architectures for AI networking. As a member, Credo will contribute its optical DSPs, silicon photonics-based integrated circuits and other components to support the standard, which released its initial specification on September 17, 2026, defining interoperable co-packaged and near-package optical modules supporting connectivity of up to 7.2 Tbps per socket. The company also recently completed testing of its Toucan Gen6x16 PCI Express retimer, which has been added to the PCI-SIG Integrators List. Credo shares have returned 51.95% year to date, and as of October 7 the stock traded at a trailing price-to-earnings multiple of approximately 77.70x and a forward P/E of 30.19x. By the end of Q2 2026, 86 hedge funds held positions in the stock, up from 59 in Q1 2026, with BlackRock the largest institutional owner at 17.76 million shares, or 9.45% of outstanding shares.
Astera Labs Unveils Aries 7, Industry's First PCIe 7 Signal Conditioning Portfolio
Astera Labs announced Aries 7, which it calls the industry's first PCIe 7 signal conditioning portfolio spanning both copper and optical interconnects. The portfolio includes smart retimers, redrivers, cable modules, optical drivers, and optical transimpedance amplifiers, extending link reach from centimeters of PCB trace to 50+ meters over optical cables across multiple racks. Aries 7 supports the OCP Signal Conditioner Standard Footprint, and its fleet management and diagnostics are unified under the COSMOS software suite. AMD and Intel, which have each validated Aries Smart Retimers in production from PCIe 4 through 6, are collaborating with Astera Labs on the PCIe 7 transition. The products are expected to begin sampling in Q4 2026 to select hyperscale, OEM, and NPO customers, with broader availability anticipated as platforms complete qualification, and Astera Labs will showcase PCIe 7 copper and optical demonstrations at the 2026 Open Compute Project Global Summit, Oct. 12-15 in San Jose.
ALAB · Technology · Positive Astera Labs unveiled Aries 7, the industry's first PCIe 7 signal conditioning portfolio spanning copper and optical interconnects.
AMD · Technology · Positive AMD is collaborating with Astera Labs on the PCIe 7 transition, having validated Aries Smart Retimers in production through PCIe 6.
INTC · Technology · Positive Intel is collaborating with Astera Labs on the PCIe 7 transition, having validated Aries Smart Retimers in production through PCIe 6.
Upscale Launches Token Fabric Open AI Networking Portfolio
Upscale, Inc. introduced Upscale Token Fabric, an open, standards-based networking architecture that unifies scale-up and scale-out networking for AI factories. The portfolio combines SkyFabriX scale-up switch silicon, which delivers 115.2 Tbps with a roadmap to multi-petabits per second, and scale-out systems powered by NVIDIA Spectrum-X, all unified by the SkyOS software foundation and SkyCMD orchestration plane. Upscale said the architecture is an industry-first open design that connects any GPU or XPU over one fabric, supporting standards including OCP ESUN, UALoE, SUE-T, and evolving protocols such as UALink. The company cited Dell'Oro forecasts of nearly $1 trillion in AI back-end switch spending through 2030 and 650 Group projections that AI networking will surpass $200 billion by 2030. General availability is planned for early 2027, with early-access and joint-validation programs already underway, and customers can adopt silicon, systems, software, or the full stack.
Upscale AI, Inc. · Technology · Positive Upscale launched its Token Fabric open AI networking architecture, an industry-first design unifying scale-up and scale-out networking for AI factories.
NVDA · Demand · Positive Upscale's Token Fabric scale-out systems are powered by NVIDIA Spectrum-X, tying the new AI networking portfolio to NVIDIA's networking products.
Credo's Retimer Business Hits Record Revenue as Blue Heron Ramps
Credo Technology Group Holding Ltd. reported record revenue in its retimer business, led by its Screaming Eagle retimer at 100G per lane, with the Blue Heron retimer beginning to contribute at 200G per lane. The company's fiscal first-quarter revenue reached $479 million, up 114.7% year over year, and management guided for second-quarter revenue of $525-$535 million. Credo introduced Blue Heron, a 224G multiprotocol AI scale-up retimer supporting UALink, ESUN and Ethernet, in January 2026, and its Toucan Gen6x16 PCIe retimer achieved PCI-SIG 6.x compliance at 64 GT/s and joined the PCI-SIG Integrators List in August. Competitors are moving as well: Astera Labs expanded its Taurus portfolio in July with the industry's first OCP-standard footprint-compatible 3.2T Smart Retimers and Smart Redrivers, and management expects the new offerings to double the Taurus market opportunity to more than $4 billion by 2030. Broadcom now expects fiscal 2026 AI revenue of $58 billion, up 186% from fiscal 2025. Credo shares have surged 31% in the past month, and the company carries a Zacks Rank #3 (Hold).
Artificial Intelligence › AI Networking & Interconnect ▲Technology
CRDO · Demand · Positive Credo's retimer business hit record revenue led by Screaming Eagle and ramping Blue Heron, with Q1 revenue up 114.7% YoY and strong Q2 guidance.
CRDO23.BK · Technology · Positive Credo introduced Blue Heron, a 224G multiprotocol AI scale-up retimer, and its Toucan Gen6x16 PCIe retimer achieved PCI-SIG 6.x compliance.
ALAB · Competition · Neutral Astera Labs expanded its Taurus portfolio with 3.2T Smart Retimers/Redrivers, a competitive move in the retimer market where Credo is the subject.
CRDO80.BK · Capital · Neutral Credo shares surged 31% in the past month and carries a Zacks Rank #3 (Hold), a valuation/rating note rather than a company development.
HPE Draws Attention on US$1.2 Billion AMD Helios AI Rack Order From Vultr
Hewlett Packard Enterprise has drawn fresh attention after a US$1.2 billion AMD Helios AI Rack order from Vultr and plans to run self-driving networking at the 2027 Ryder Cup in Ireland. The company's shares have returned 38.63% over the past month and gained 198.26% year to date, with a 1-year total shareholder return of 181.39% and a 5-year total shareholder return of 460.84%. The stock closed at $72.09 against a narrative fair value of $68.08, a view held by 87 investors who see the shares as 6% overvalued. On simple P/E math, Hewlett Packard Enterprise trades at 35.8x earnings, below its peer average of 76.9x and a fair ratio of 39x, though above the broader Global Tech industry at 19.5x. Strategic acquisitions and expansion in high-growth technologies, including the integration of Juniper, launches of next-gen Gen12 servers, and AI-driven management platforms, are cited as enhancing HPE's competitive positioning in edge, networking, and AI.
Morgan Stanley Report Triggers Sharp Sell-off in CPO Concept Stocks
On October 8, CPO concept stocks in China's A-share market fell sharply. Yuanjie Technology and Changguang Huaxin hit their 20 percent daily limit down, Shijia Photonics dropped more than 15 percent, Dongshan Precision hit its 10 percent daily limit down, Dekeli fell more than 10 percent, and Changxin Bochuang dropped more than 8 percent. On the news front, Morgan Stanley recently issued a report noting that potential U.S. restrictions on optical modules from China may be implemented through the FCC controlled list mechanism rather than a full embargo. The core of the policy is the 65 percent U.S. content rule, meaning Chinese-made transceivers can still be imported but must meet the condition that 65 percent of the value in the bill of materials comes from U.S. companies. The report stressed that this is a reshaping of the supply chain and a redistribution of profits along the value chain, rather than a simple positive or negative from sanctions. The rule could be implemented as early as the 3.2T optical module stage, with mass production expected in 2028. At that point, Chinese manufacturers would be allowed to continue assembly, but procurement of core components such as DSPs and lasers would be locked in with U.S. suppliers. The report specifically noted that the combination of DSPs and lasers could easily reach the 65 percent threshold. The report believes the short-term impact is limited, because large-scale adoption of 3.2T products is not expected until 2029, and hyperscale cloud service providers have already arranged for U.S. domestic manufacturing. The policy goal is to target the production location of next-generation products, not to immediately replace the existing supply chain.
Artificial Intelligence › AI Networking & Interconnect ▼Regulation
688313.CG · Tariff · Negative Shijia Photons dropped over 15% as the Morgan Stanley report warned of U.S. content-rule restrictions reshaping the optical module supply chain.
002384.CS · Tariff · Negative Dongshan Precision hit its 10% limit down amid the CPO sell-off triggered by potential U.S. restrictions on Chinese optical modules.
300548.CS · Tariff · Negative Broadex Technologies fell more than 10% as part of the CPO concept sell-off driven by the Morgan Stanley report on U.S. optical module restrictions.
688048.CG · Tariff · Negative Named among CPO concept stocks that hit 20% limit down after Morgan Stanley flagged potential U.S. restrictions on Chinese optical modules via the FCC 65% U.S. content rule.
Everbright Photonics Responds to Morgan Stanley Report: 3.2T Ramp-Up Still Far Off, No Short-Term Impact on Domestic Optical Chips
A staff member on Everbright Photonics' investor hotline, responding to an interview with National Business Daily, said the company has taken note of the Morgan Stanley research notes mentioning that potential FCC rules could restrict Chinese optical modules in the 3.2T generation. The staff member pointed out that the Morgan Stanley report is not an official document, and the authenticity and specific impact of the information remain uncertain. Secondly, the projection targets 3.2T optical modules, but the volume ramp-up of 3.2T itself is not expected until 2028 to 2029, with the current market still dominated by 800G and 1.6T. Therefore, there is no substantive impact on the domestic optical chip industry in the short term. Everbright Photonics' share price is 231.76 yuan, with a market capitalization of 40.855 billion yuan.
Artificial Intelligence › AI Networking & Interconnect Regulation
688048.CG · Regulation · Neutral Company says Morgan Stanley's note on potential FCC rules restricting Chinese optical modules in 3.2T is unofficial and uncertain, with no short-term impact on domestic optical chips.
Marvell targets fiscal 2031 revenue of $70–90 billion
Marvell unveiled its plan at Investor Day 2026, targeting fiscal 2031 revenue of as much as $70–90 billion, highlighting custom-designed chips and high-speed interconnect systems as key drivers. The company aims for roughly $20 billion in fiscal 2028 revenue, above the $18.2 billion analysts had expected, and raised its fiscal 2029 target for its custom chip business to $12 billion from $10 billion. Its collaboration with Google spans AI compute chips as well as networking, memory, and storage chips, reflecting how AI investment is spreading across multiple parts of the data center. The revenue targets are long-term goals for Marvell and do not mean other companies in the supply chain, such as Astera Labs and Lumentum, will grow at the same rate.
MRVL · Capital · Positive Marvell raised its fiscal 2028 revenue target above analyst estimates and lifted its fiscal 2029 custom-chip target to $12B from $10B.
GOOG · Demand · Positive Marvell's collaboration with Google spans AI compute, networking, memory and storage chips, reflecting Google's expanding AI data-center demand.
Keysight Launches Velocity 4900 Series Layer 1 Automation Switches for AI and 6G Test Labs
Keysight Technologies announced the Keysight Velocity 4900 Series, a new line of cost-effective Layer 1 automation switches designed to automate physical connectivity in network test labs supporting AI, 6G, and next-generation services. The series launches with two models: the Keysight Velocity 4902, a high-speed switch supporting Ethernet speeds from 100 megabits per second to 100 gigabits per second with 72 QSFP28 ports and 288 connections with breakouts, and the Keysight Velocity 4906, a high-density switch with 288 SFP28 ports supporting speeds from 100 megabits per second to 25 gigabits per second. Both switches provide true Layer 1 forwarding through an ultra-low-latency, fully transparent switching matrix and support single-mode fiber, multimode fiber, direct-attach copper, and active optical cables, with port tapping and one-to-many traffic replication. Keysight said the platform replaces manual testbed configuration with software-controlled connectivity, letting teams provision complex topologies in seconds or minutes instead of hours or days of manual cabling, share scarce equipment remotely, and integrate physical test environments into CI/CD workflows. Anil Kollipara, Vice President and General Manager of Automated Testing and Assurance at Keysight, said engineering teams need test environments that can change as rapidly as the technologies they are creating. The Keysight Velocity 4902 and 4906 are available for order immediately, with initial deliveries planned for January 2027.
KEYS · Technology · Positive Keysight launched the Velocity 4900 Series Layer 1 automation switches for AI and 6G test labs, a new product line available for order.
ClearBridge Flags Arista Networks as Q3 Standout on Raised Guidance
ClearBridge Investments named Arista Networks as a leading contributor in its third-quarter 2026 commentary for its Large Cap Growth Strategy, citing solid execution and new customer scaling. The firm said Arista was a positive outlier among AI infrastructure names and raised full-year guidance in its latest update, easing concerns created by earlier supply constraints and uneven revenue trends. Arista Networks closed at $215.36 per share on October 6, 2026, returning 11.63% over the past month and 64.36% over the past year, with a market capitalization of $271.62 billion and a 52-week range of $114.52 to $216.07. The ClearBridge Large Cap Growth Strategy outperformed its Russell 1000 Growth Index benchmark, which gained 0.9%, while the Russell 1000 Value Index rose 2.6%. According to the firm's database, 91 hedge fund portfolios held Arista Networks at the end of the second quarter, up from 85 in the previous quarter.
ANET · Capital · Positive ClearBridge named Arista a top Q3 contributor after it raised full-year guidance, easing supply-constraint and uneven-revenue concerns.
Ciena Shares Jump 14% on Marvell's 60-70% Optical Growth Outlook
Ciena Corp. rebounded as much as 14.3 percent in intraday trading on Tuesday to $445.36 apiece, helped by peer Marvell Technology's growth outlook for the optical interconnects sector. At its recent Investor Day, Marvell said it expects the industry to grow by 60 to 70 percent to about $65 billion through 2030, with switching and storage alone reaching $85 billion, or a compounded annual growth rate of 40 percent. Those figures were markedly higher than Ciena's own 30 percent growth outlook for its business by 2029. Bernstein initiated coverage with an outperform rating and a $440 price target, acknowledging near-term supply constraints but calling the stock undervalued, while Evercore maintained an outperform rating and a $550 price target, citing optimism that Ciena can achieve more than $25 in earnings per share. Hedge fund holders in Ciena rose to 81 in the second quarter from 73 in the first, with combined holdings up 2 percent to $3.997 billion from $3.913 billion quarter-on-quarter.
CIEN · Capital · Positive Ciena shares jumped 14% after Bernstein initiated with an outperform rating and $440 target and Evercore maintained outperform with a $550 target, calling the stock undervalued.
MRVL · Demand · Positive Marvell's Investor Day projected the optical interconnects industry to grow 60-70% to about $65 billion through 2030, signaling strong end-market demand.
Marvell raises fiscal 2028 revenue outlook to $20 billion, stock jumps 8%
Marvell Technology raised its revenue guidance for fiscal year 2028 to approximately $20 billion, sending the semiconductor designer's stock up roughly 8% on Tuesday to its highest level since June. CEO Matt Murphy said at the company's investor day that the new target represents about 67% year-over-year growth, up from the $18 billion the company guided to in August and above the $18.2 billion Wall Street had anticipated, according to Bloomberg data. Marvell also said it sees a total addressable AI market reaching approximately $400 billion by 2030, driven by custom silicon programs with hyperscalers including Amazon, Google, and Microsoft. The company, which competes with Broadcom in high-speed infrastructure and custom AI chips, joined the S&P 500 in June and its shares are up more than 240% year to date.
Astera Labs Stock Up 118% YTD as Q3 Revenue Guidance Tops $540 Million
Astera Labs shares have climbed 117.8% year to date, outpacing the broader Zacks Computer & Technology sector's 24.2% gain and the Zacks Internet - Software industry's 7.1% growth, as well as peers Cisco Systems and Credo Technology, which rose 46.5% and 47.7% respectively. The company guided third-quarter 2026 revenues to between $540 million and $560 million, with the midpoint implying roughly 40% sequential growth, while management projected non-GAAP earnings of $1.16 to $1.21 per share. The Zacks Consensus Estimate for third-quarter revenues stands at $550.39 million, up 138.71% year over year, and the consensus earnings estimate of $1.19 per share implies a 142.86% year-over-year increase. In the second quarter of 2026, PCIe 6 products including the Scorpio AI Fabric Switches and Aries Retimers accounted for more than 50% of total revenues, and the Leo CXL memory controller line has won new design wins at major U.S. hyperscalers with volume shipments expected in 2027. Astera Labs faces rising competition from Marvell Technology, whose data center revenues represented 79% of total revenues in the second quarter of fiscal 2027, and its shares trade at a forward 12-month Price/Sales of 22.66X versus the Computer & Technology sector's 6.18X, with a Zacks Rank #3 (Hold).
Marvell Expands Nvidia AI Partnership as Nvidia Invests $2 Billion
Marvell Technology is expanding its partnership with Nvidia across advanced networking, silicon photonics and optical interconnect technologies, with Nvidia investing $2 billion in Marvell as part of the March 2026 agreement. The collaboration began in 2025 to give hyperscalers more flexibility in building custom AI computing infrastructure, and was expanded in 2026 to connect Marvell to Nvidia's AI factory and AI-RAN ecosystems through NVLink Fusion. Under the expanded arrangement, Marvell will supply custom XPUs and NVLink Fusion-compatible scale-up networking, while Nvidia provides Vera CPUs, ConnectX NICs, BlueField DPUs, NVLink interconnects, Spectrum-X switches and rack-scale AI compute. Marvell also worked with Microsoft and Utimaco to launch Azure Payment HSM v2, a cloud-native payment security platform built on Marvell's LiquidSecurity hardware security modules, with Utimaco contributing Atalla Payments Module software and Microsoft providing Azure. Marvell competes with Astera Labs and Broadcom in networking, and its shares have gained 239.5% year to date, trading at a price-to-sales multiple of 14.78X versus the Zacks Electronics - Semiconductors industry's 5.15X, while the Zacks Consensus Estimate for fiscal 2027 earnings implies year-over-year growth of 48.2%.
Fabric.AI Taps Kearney to Lead Commercialization of Neural I/o Optical Interconnect Platform
Fabric.AI has engaged global management consulting firm Kearney to lead and accelerate the commercialization strategy for its Neural I/o MicroLED-based optical interconnect platform. The company, together with its development partner Kopin Corporation, has executed non-disclosure agreements with several leading semiconductor and data center technology companies, with technical discussions already underway. Kearney's mandate is to convert that early engagement into a structured commercial pipeline by sharpening product positioning, identifying and prioritizing target customers and channel partners, and preparing Fabric.AI for active sales discussions expected to begin after planned hardware demonstrations in early 2027. James Altucher, an independent strategic advisor to Fabric.AI, said Neural I/o addresses the two constraints defining the next generation of AI data centers, how much data can be moved and what it costs in power to move it, and that Kearney's role is to help turn confidential discussions into commercial relationships. Fabric.AI trades on Nasdaq under the ticker FABC, while Kopin Corporation trades under KOPN.
FABC · Demand · Positive Fabric.AI engaged Kearney to commercialize Neural I/o and has NDAs plus technical discussions underway with leading semiconductor and data center firms.
KOPN · Demand · Positive Kopin is Fabric.AI's development partner on the Neural I/o platform, which has NDAs and technical talks underway with leading semiconductor and data center companies.
Nokia CEO Says Data Centers Could Be Built Twice as Fast Without Supply Bottlenecks
Nokia CEO Justin Hotard said the tech industry could build data centers twice as fast if not for supply constraints, telling CNBC that continued demand signals the AI buildout is still in its early days. "If we could build 2x faster, our customers could build 2x faster, they probably would," Hotard said, adding that he does not believe the industry is overbuilding. Nokia, one of the world's biggest telecommunications equipment makers, has pivoted into the AI infrastructure space, and its stock is up around 130% over the past year. Investments in the AI buildout, from data centers to chips and related infrastructure, will total $10.3 trillion from 2025 to 2032, an average of 3.63% of U.S. gross domestic product per year, according to research presented at the Brookings Papers on Economic Activity. Hotard said demand is not solely dependent on frontier labs like Anthropic and OpenAI releasing new models, arguing that even without another frontier model in the next three years, deploying existing technology could yield tremendous progress.
Artificial Intelligence › AI Data Center & Build-out ▲Supply
Artificial Intelligence › AI Networking & Interconnect ▲Supply
Artificial Intelligence › Build-out, Construction & Engineering ▲Supply
Artificial Intelligence › AI Compute & Accelerator Silicon ▲Supply
NOK · Demand · Positive Nokia CEO says AI data-center demand is still early and supply-constrained, signaling strong demand for Nokia's AI infrastructure equipment.
Narui Radar plans to acquire 51% stake in Xintai Communications for 162 million yuan, resuming M&A push into optical communications
Narui Radar announced on the evening of September 30 that it plans to use 162 million yuan of its own funds to acquire a 51% stake in Guangzhou Xintai Communication Technology Co., Ltd. After the transaction, Xintai Communications will become its controlling subsidiary. Founded in 2013, Xintai Communications focuses on integrated optical transmission systems and optical transmission optimization solutions. In 2025, it achieved revenue of 109 million yuan and net profit of 19.72 million yuan; in the first half of 2026, revenue was 66.43 million yuan and net profit was 13.14 million yuan. Using June 30, 2026 as the base date, the income approach valued the full equity of its shareholders at 317 million yuan, a premium of 385.58%, and this transaction adopts the income approach conclusion. The counterparty commits that Xintai Communications' non-GAAP net profit from 2026 to 2028 will be no less than 25 million yuan, 29 million yuan, and 38 million yuan respectively, representing a first-year commitment growth of about 30% compared with the 19.31 million yuan non-GAAP net profit in 2025. Earlier in March this year, Narui Radar terminated its plan, which had been in preparation for more than 14 months, to acquire a 100% stake in Tianjin Sigma Microelectronics Technology Co., Ltd. and withdrew all application documents.
Artificial Intelligence › Optical Interconnect & DCI Capital
688522.CG · Capital · Positive Naruida plans to acquire a 51% controlling stake in Xintai Communications for 162 million yuan, resuming its M&A push into optical communications.
Arista Networks Lifts 2026 Revenue Guidance to US$12.60 Billion on AI Demand
Arista Networks raised its 2026 revenue guidance to US$12.60 billion, including at least US$3.50 billion from AI-related business, after reporting second-quarter fiscal 2026 revenue of US$3.00 billion, a 37.7% year-over-year increase. The higher outlook underscores how AI-driven cloud networking has become a central revenue pillar for the company, shaping its growth mix across hyperscale and enterprise customers. Arista was also added to the S&P 100 in September 2026, a move that reflects its growing weight in large-cap U.S. equity indices just as management ties more of its outlook to AI-related revenue. The company's longer-range narrative projects US$21.4 billion in revenue and US$8.2 billion in earnings by 2029, requiring 26.6% yearly revenue growth and an earnings increase of about US$4.2 billion from US$4.0 billion today. Some of the lowest-ranked analysts had already assumed about US$18.6 billion of 2029 revenue and US$6.9 billion of earnings before this guidance, illustrating how differently Arista's AI exposure and long-term supply commitments can be read.
ANET · Demand · Positive Arista raised 2026 revenue guidance to $12.60B, including at least $3.50B from AI-related business, on AI-driven cloud networking demand.
ANET · Capital · Positive Arista was added to the S&P 100 in September 2026, reflecting its growing weight in large-cap U.S. equity indices.
Nvidia Boosts Buyback by $150B as AMD Strikes $8.2B World Labs Deal
Nvidia said its board increased its share buyback program by $150 billion, leaving $235 billion in total authorization, the largest buyback program in U.S. history according to data compiled by Bloomberg. Advanced Micro Devices said it would acquire AI model and research lab World Labs in an all-stock transaction valued at nearly $8.2 billion, a deal Citi analyst Atif Malik said gives the company deeper visibility into how AI models are evolving, while RBC analyst Srini Pajjuri said it could help position AMD to offer full-stack solutions for robotics, simulation and physical AI markets. Micron Technology reported fourth-quarter results and an outlook that beat expectations as its strategic customer agreements continued to increase, and Accenture reported fiscal quarterly results that exceeded expectations on both revenue and bookings, sending its shares up nearly 16% on Thursday. HPE closed nearly 4% higher after saying it won a $1.2 billion order of AMD's Helios AI rack from cloud infrastructure company Vultr, and raised its fiscal 2027 networking growth outlook to the high-teens to low-20s on a percentage basis, with data center networking revenue expected to grow at a low-to-high 50s percent compound annual rate through fiscal 2029. Anthropic intends to launch its initial public offering before the Thanksgiving holiday, according to Bloomberg, which also reported that Broadcom's Wall Street syndicate is starting to gather $60 billion of fresh AI chip financing to benefit Anthropic PBC and other companies, while Meta Platforms is adding a new Meta Enterprise Platform for organizations, tapping MongoDB CEO Chirantan Desai to lead it, and introduced Muse for Small Business.
AMD · Capital · Positive AMD said it would acquire AI model and research lab World Labs in an all-stock transaction valued at nearly $8.2 billion.
HPE · Demand · Positive HPE won a $1.2 billion order of AMD's Helios AI rack from cloud infrastructure company Vultr.
HPE · Capital · Positive HPE raised its fiscal 2027 networking growth outlook to high-teens to low-20s and expects data center networking revenue to grow at a low-to-high 50s percent CAGR through fiscal 2029.
NVDA · Capital · Positive Nvidia's board increased its share buyback program by $150 billion, the largest in U.S. history.
World Labs · Capital · Positive AMD agreed to acquire World Labs in an all-stock deal valued at nearly $8.2 billion.
ACN · Capital · Positive Accenture reported fiscal quarterly results that exceeded expectations on both revenue and bookings, sending shares up nearly 16%.
Meta Unveils Petal Subsea Cable Linking US and France
Meta announced Petal on Sept. 21, a roughly 4,300-mile subsea cable connecting the U.S. and France that is expected to enter service in 2029 and will be the first to deliver petabit-scale capacity across an ocean. The cable will carry 1 petabit per second, or about 125,000 gigabytes per second, double what today's most advanced transoceanic cables can handle, using two-core fiber that fits two light paths inside each strand. Petal is one piece of a much larger seafloor empire: Meta says it has invested in more than 20 subsea cable projects touching every continent except Antarctica, led by Project Waterworth, announced in February 2025, which will stretch more than 31,000 miles and connect the U.S., Brazil, South Africa, India, and other regions across five continents. Petal is being developed with Japan's NEC and Sumitomo Electric Industries, with French telecom Orange handling the landing on France's Atlantic coast, while Waterworth will be wholly owned by Meta, only its third solely owned cable according to telecom research firm TeleGeography. Meta has not disclosed what Petal or Waterworth will cost, but TechCrunch reported before Waterworth's announcement that the project could top $10 billion, a small slice of the company's expected capital expenditures of $130 billion to $145 billion this year, nearly double the $72.2 billion it spent in 2025.
HPE Raises Fiscal 2027 Networking Outlook, Lands $1.2 Billion Vultr AI Order
Hewlett Packard Enterprise raised its fiscal 2027 networking revenue growth forecast to between high-teens and low-20s percent, up from the 14% to 17% projection it gave earlier in September, and secured a $1.2 billion order from cloud infrastructure provider Vultr for AMD-based artificial intelligence server racks paired with Juniper networking switches. The company also lifted its annual cost-synergy target from the Juniper Networks acquisition to $800 million by the end of fiscal 2028, up from at least $600 million, and projected Data Center Networking revenue to grow at a low-to-high 50s percent compound annual rate through fiscal 2029. Following the networking investor day presentation, Citigroup raised its price target on the stock to $92 from $76 while keeping a Buy rating, citing confidence in durable demand and expanding AI infrastructure orders. A cooler-than-expected U.S. jobs report added secondary momentum, with the Bureau of Labor Statistics reporting that employers added 29,000 nonfarm payroll jobs in September, trailing the 84,000 consensus estimate from economists polled by Dow Jones. HPE shares jumped 7.8% in the afternoon session and are up 189% since the beginning of the year, reaching a new 52-week high of $69.77 per share.
Broadcom Guides Q4 Revenue to $34.8 Billion After Q3 Beat
Broadcom reported third-quarter fiscal 2026 non-GAAP earnings of $3.32 per share, up 96.4% year over year and 3.11% above the Zacks Consensus Estimate, with revenues surging 85.5% year over year to $29.59 billion and beating the consensus mark by 0.41%. AI semiconductor revenues jumped 221% year over year to $16.7 billion, representing 56% of total revenues, while Semiconductor Solutions revenues surged 127% year over year to a record $20.84 billion, or 70% of total revenues, and Infrastructure Software revenues increased 29% year over year to $8.75 billion, contributing the remaining 30%. For the fourth quarter of fiscal 2026, Broadcom expects revenues of approximately $34.8 billion, representing 93% year-over-year growth, including about $26.1 billion of semiconductor revenues and roughly $8.7 billion of Infrastructure Software revenues. The company raised its fiscal 2026 AI semiconductor revenue outlook to $58 billion, up 186% year over year and above its prior $56 billion view, and said it has secured supply to support approximately $115 billion of AI semiconductor revenues in fiscal 2027, with fiscal 2028 AI semiconductor revenues seen reaching $230 billion. Broadcom also generated record free cash flow of $13.67 billion, equal to 46% of revenues, paid $3.1 billion in dividends, reduced long-term debt by $5.6 billion, and its board approved a quarterly dividend of 65 cents per share.
AVGO · Capital · Positive Broadcom beat Q3 estimates with EPS up 96.4% and guided Q4 revenue to $34.8B, plus record $13.67B free cash flow and debt reduction.
AVGO · Demand · Positive AI semiconductor revenues jumped 221% to $16.7B and the company raised its fiscal 2026 AI revenue outlook to $58B with supply secured for $115B in fiscal 2027.
Celestica Inc. reported second-quarter revenue growth of 62% year over year and an 83% rise in adjusted EPS, then raised its 2026 outlook and said revenue growth should accelerate again in 2027. The company expects its 1.6T networking programs to ramp significantly in 2027, is preparing to deliver custom racks for OpenAI, and expects its AMD Helios opportunity to ramp in the first half of next year, with management describing both as multibillion-dollar opportunities. Celestica trades at about 18.5 times forward earnings while consensus calls for roughly 75% EPS growth next year, though three customers accounted for 32%, 17% and 14% of Q2 revenue and roughly $1 billion of capital expenditures is planned for 2026. Hedge fund interest weakened slightly in the second quarter, with the number of funds holding the stock falling from 67 in Q1 to 63 in Q2 while the total value of their positions rose from approximately $4.38 billion to $4.84 billion. Management says customers are already discussing requirements for 2027, 2028 and even 2029, making the upcoming October 27 investor day particularly important.
OpenAI · Demand · Positive Celestica is preparing to deliver custom racks for OpenAI, a multibillion-dollar opportunity.
AMD · Demand · Positive Celestica expects its AMD Helios opportunity to ramp in the first half of next year, described as a multibillion-dollar opportunity.
Coherent Jumps 10.5% After Bernstein Initiates With Outperform and $350 Target
Coherent shares jumped 10.5% in the afternoon session after Bernstein rated the materials and photonics company Outperform with a $350 price target, citing optical-module upside as data-center switches move to higher speeds. According to Streetinsider, Bernstein framed $350 as roughly 24% upside, valuing the stock at 22 times estimated 2028 earnings per share of $16, and argued optical-module demand should keep expanding with faster networking gear. Citi commentary tagging Coherent as an AI hardware beneficiary reinforced the bullish AI-infrastructure read-across, alongside investor focus on Coherent's PhotonLink optics platform and an AI security demo amid a broader bid in networking names. The move follows a 5.2% gain two days earlier on news that Coherent demonstrated scalable physical-layer security for artificial intelligence infrastructure developed alongside CUbIQ Technologies. Coherent is up 63.1% since the start of the year, but at $317.01 per share it still trades 25.7% below its 52-week high of $426.89 from June 2026.
Coherent, Lumentum, Ciena Surge as Bernstein Initiates With Outperform
Optical networking stocks rallied Thursday after Bernstein initiated coverage of the group with an Outperform rating, calling the companies structural AI winners. Coherent jumped 11%, Lumentum rose 7%, Ciena gained 7%, and Arista added 0.4%. Coherent led the move after launching PhotonLink, an optics platform for AI infrastructure that packages its services, photonics technologies, and components including lasers, fibers, and detectors into one platform for AI data centers. The company said it already has 20 customer engagements split between co-packaged optic and near-packaged optic assemblies, plus five engagements in emerging chip-to-chip connectivity. The gains extend a strong 2026 rally for the group, with Coherent up 73%, Lumentum up 183%, Ciena up 62%, and Arista up 56% year to date.
COHR · Capital · Positive Bernstein initiated coverage of Coherent with an Outperform rating as a structural AI winner.
COHR · Technology · Positive Coherent launched PhotonLink, an optics platform for AI infrastructure, with 20 customer engagements.
CIEN · Capital · Positive Bernstein initiated coverage of Ciena with an Outperform rating, calling it a structural AI winner.
LITE · Capital · Positive Bernstein initiated coverage of Lumentum with an Outperform rating, calling it a structural AI winner.
ANET · Capital · Positive Bernstein initiated coverage of the optical networking group with an Outperform rating, naming Arista among structural AI winners.
Credo Technology Tops Q1 Estimates With $479 Million Revenue, Guides Q2 to $525-535 Million
Credo Technology Group reported first-quarter fiscal 2027 revenues of $479 million, up 114.7% year over year and 9.6% sequentially, marking its seventh consecutive quarter of triple-digit year-over-year revenue growth. The revenue base remains concentrated among large AI infrastructure customers, with the top four customers each contributing at least 10% of revenues, led by the largest at 33%, the second-largest at 28%, the third at 13% and the fourth at 10%. Non-GAAP gross margin was 68% on non-GAAP gross profit of $325.8 million, while non-GAAP net income was $236.3 million, a non-GAAP net margin of 49.3%. For the fiscal second quarter, Credo expects revenues of $525 million to $535 million, with GAAP gross margin projected between 62.9% and 64.9% and non-GAAP gross margin between 67% and 69%; for fiscal 2027, management expects total revenues to grow more than 85% year over year. The DustPhotonics acquisition expanded Credo's optical stack, and management expects ZeroFlap optics, silicon photonics PICs and optical DSPs to each contribute more than $100 million in fiscal 2027 revenues, supporting a target of more than $600 million in optical revenues for the year.
Artificial Intelligence › AI Networking & Interconnect ▲Demand
Artificial Intelligence › AI Compute & Accelerator Silicon ▲Demand
CRDO · Capital · Positive Credo topped Q1 estimates with $479M revenue and guided Q2 to $525-535M, with FY2027 revenue growth above 85%.
CRDO · Demand · Positive Revenue concentrated among large AI infrastructure customers, with top four each contributing at least 10%, and optical products expected to each exceed $100M in FY2027.
Software-Defined Networking Market to Reach $82.59 Billion by 2030, MarketsandMarkets Says
The global Software-Defined Networking Market is projected to grow from USD 35.74 billion in 2025 to USD 82.59 billion by 2030, a CAGR of 18.2% over the forecast period, according to MarketsandMarkets. Declining hardware and connectivity costs are lowering barriers to network modernization and supporting broader SDN deployments across data centers, enterprise networks, telecom infrastructure, and cloud environments. Within the market, the hybrid SDN segment is expected to register the higher growth rate, while the cloud service providers segment is expected to grow at the highest CAGR, with hyperscalers such as AWS, Google Cloud, and Azure using SDN for multi-tenant isolation and elastic workloads. Asia Pacific is projected to be the fastest-growing region, driven by 5G buildout, digital economies, and cloud services, with China's 5G development and India's expanding digital infrastructure supporting adoption. Key players named in the report include Cisco, Huawei, VMware, Dell EMC, IBM, Ericsson, Ciena, and Hewlett Packard Enterprise.
Volantis Raises $88M Series A for Photonic AI Inference System
Volantis, a semiconductor company building a new category of AI inference system, announced an $88 million Series A co-led by Lachy Groom and Abstract Ventures, with participation from John Doerr, VXI Capital, Triatomic and Susa Ventures, plus angel investors Dwarkesh Patel, Naveen Rao and Sholto Douglas. The company's first system, A-1, is being designed to run models exceeding 20 trillion parameters at up to 10,000 tokens per second per user while reducing inference cost per token. Volantis says A-1 will increase memory capacity and bandwidth simultaneously by nearly two orders of magnitude, using a photonic interconnect that connects compute chips to memory and aggregates the bandwidth of large numbers of memory chips into a unified pool. The architecture uses custom micro-VCSELs rather than external lasers, drawing on the existing gallium arsenide VCSEL supply chain and avoiding indium phosphide supply constraints, with end-to-end links consuming less than one picojoule per bit. The financing will support development and commercialization of A-1 and its photonic memory architecture, including expanding the engineering team, and Volantis plans to deliver its first integrated inference engines to customers in 2027.
Artificial Intelligence › Custom Silicon / ASIC Capital
Volantis · Capital · Positive Volantis raised an $88M Series A co-led by Lachy Groom and Abstract Ventures to fund development and commercialization of its A-1 photonic AI inference system.
JPT replies to private placement review inquiry, plans to raise no more than 1.373 billion yuan to fill expansion funding gap
Shenzhen JPT Opto-electronics Co., Ltd. disclosed its reply to the review inquiry letter regarding its issuance of shares to specific investors, with total funds to be raised not exceeding 1,372.97 million yuan. After deducting issuance expenses, the proceeds are intended for a high-density optical interconnect product manufacturing construction project, a JPT headquarters and R&D center upgrade construction project, and supplementary working capital. Of this, the high-density optical interconnect product manufacturing construction project is planned to receive 765.60 million yuan, the headquarters and R&D center upgrade construction project is planned to use 217.37 million yuan, and supplementary working capital is planned at 390.00 million yuan. JPT stated that its overall funding gap is estimated at 1,404.6439 million yuan, larger than the 1,372.97 million yuan financing size, so this financing is necessary. During the reporting period, the company's operating revenue was 1,225.6253 million yuan, 1,453.8462 million yuan, 2,074.0243 million yuan, and 1,487.0184 million yuan respectively, with a compound annual growth rate exceeding 30% over the past three years. In 2025 and the first half of 2026, the capacity utilization rates for optical fiber device products were 109.33% and 100.78% respectively. As of July 5, 2026, the order backlog for this business was 122.8067 million yuan, and cumulative intended orders of approximately 100 million US dollars are expected to be secured in 2026.
688025.CG · Capital · Positive JPT plans a private placement of up to 1.373 billion yuan to fund optical interconnect manufacturing, HQ/R&D upgrade, and working capital, addressing its funding gap.
688025.CG · Demand · Positive Article cites >100% capacity utilization, a 122.8 million yuan order backlog, and ~$100 million of intended 2026 orders for its optical fiber device business.
Digital Realty and Blackfuel Deploy AI Inference Platform on ServiceFabric in Barcelona
Digital Realty announced that Blackfuel has selected its BCN1 data center in Barcelona as the foundation for deploying and scaling its AI inference platform. The multi-megawatt deployment is built on liquid-cooled Dell Technologies AI Factory infrastructure and AMD GPUs, and integrates Blackfuel's inference platform with ServiceFabric, Digital Realty's software-defined interconnection platform. Delivered through private connectivity, ServiceFabric provides a low-latency, low-jitter network path that can reduce Time-to-First-Token by 50 milliseconds, shortening overall task completion time by 1 to 2.5 seconds for agentic workflows requiring 20 to 50 sequential inference API calls compared with public internet connectivity. ServiceFabric extends across more than 800 Digital Realty and third-party data centers worldwide, giving Blackfuel a global connectivity fabric to scale AI services beyond Barcelona. Chris Sharp, Chief Technology Officer of Digital Realty, said the next phase of AI will be measured in useful tokens per megawatt, and Barcelona is the first proof point.
CoreWeave to Add NVIDIA Vera CPU for AI Agent Workloads
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.
Artificial Intelligence › AI Networking & Interconnect ▲Technology
CRWV · Technology · Positive CoreWeave expands its compute portfolio with NVIDIA Vera CPU, claiming 3x faster agent sandbox startup for AI agent workloads.
NVDA · Technology · Positive CoreWeave adopts NVIDIA's Vera CPU, the first CPU designed for AI agents, as part of its AI cloud fleet.
CScale Exits Stealth With $145 Million Series C for AI Optical Interconnect
CScale exited stealth, announcing $145 million in Series C funding to accelerate development and commercialization of optical interconnect for AI scale-up, bringing the company's total funding to $188 million. The round was co-led by Atreides Management, Valor Equity Partners, and Premji Invest, with Sutter Hill Ventures continuing its support alongside existing investor Maverick Silicon. NVIDIA and Intel Capital joined as CScale's first strategic investors. The Palo Alto company is architecting its interconnect to contain optical failures without interrupting compute, targeting the gigawatt-class AI data centers expected by the end of the decade, where a single scale-up domain will span thousands of tightly coupled accelerators across dozens of racks. CEO Martin Lund said lasers will fail but compute should not, and Atreides Management Managing Partner and Chief Investment Officer Gavin Baker said every optical failure is a compute failure at that scale.
