Goldman Sachs significantly raises forecast for optical transceiver market: size may reach $148.5 billion by 2028
① High-speed products have become the core growth driver, with the market size for 800G and above optical transceivers expected to increase from $45.2 billion in 2026 to $129.5 billion in 2028; ② The expansion of AI rack-scale systems, increased penetration of ASICs, and higher optical transceiver usage per AI chip are jointly accelerating the ramp-up of demand for 1.6T and 3.2T modules; ③ The penetration rate of silicon photonics in high-speed optical transceivers continues to rise, while Co-Packaged Optics (CPO) remains primarily in the early adoption phase in the short term.
Express News | Serenity: Amid surging demand for AI computing power, laser manufacturers may have greater room for expansion; the Lumentum case is worth watching.
On September 7, Serenity published a note stating that its previous revenue estimates, based on Sivers' (SIVE) annual production target of 100 million CW DFB lasers, were merely illustrative models. More importantly, Serenity is bullish on laser companies because they are not limited to single laser products and can further expand into other segments of data center optical communications. Citing Lumentum (LITE) as an example, Serenity noted that after acquiring Cloud Light (which primarily focuses on pluggable optical modules), Lumentum was able to expand its potential serviceable market within data centers to more than five times its previous size. Additionally, Lumentum's OFC presentation materials indicated that its ELS business, based on UHP laser chips, could offer approximately a two-fold expansion opportunity in total addressable market (TAM). Serenity also pointed out that the laser business of just one UHP laser wafer fab is expected to reach an annual revenue scale of approximately $5 billion once capacity ramp-up is complete. Given the roughly 55%-65% gross margin for CPO lasers mentioned by AAOI, lasers themselves already demonstrate strong profitability. Serenity believes that even with continued industry capacity expansion, there may still be a supply-demand gap in the laser market. Furthermore, as multiple technology cycles overlap—including NPO, CPO, and 1.6T pluggable optical modules—laser manufacturers are poised to continuously enter new product markets. Therefore, compared to companies that remain long-term at a single tier of the supply chain, Serity argues that the market should assign a higher valuation premium to laser companies.
Major Wall Street banks interpret this year's SEMICON highlights: Advanced packaging is evolving toward 3D system integration, while CPO manufacturing bottlenecks lie in testing.
SEMICON 2026 sends new signals for the semiconductor industry: advanced packaging is upgrading toward 3D system integration, CoWoS capacity continues to expand and spill over to OSATs, CPO is accelerating mass production, while testing and equipment lead times have become new bottlenecks, and the AI ASIC ecosystem is further maturing. UBS Group is bullish on Taiwan Semiconductor's 3DFabric capacity expansion, Citi focuses on CPO manufacturing, and Nomura warns of capacity constraints in testing and equipment segments, as AI-driven capacity expansion accelerates into the backend processes.
OpenAI Makes Initial Breakthrough in Automated Research as the “AI Training AI” Paradigm Accelerates; Is Computing Power Still a Major Bottleneck for the Future?
① OpenAI published the article "Accelerating Research: An Inside View," publicly disclosing progress on RSI in the form of internal data for the first time; ② OpenAI expects to achieve a fully "automated AI researcher" by March 2028; ③ Team member Kevin Liu pointed out on social media: "RSI is likely to be the core driver of leaps in AI capabilities in the coming years."
U.S. Large-scale Data Centers Could Triple by 2030: Kobeissi
The Trade Desk to Leave S&P 500 as Bloom Energy, Everpure, Illumina Join Index
Communication Services Sector Rises in August; Paramount Leads Gains
Express News | According to reports, Anthropic is expected to file its IPO prospectus in late September and could begin its IPO roadshow as early as mid-October.
Non-farm payrolls data rattled the broader U.S. stock market, yet AI hardware stocks surged against the trend. The semiconductor index rose more than 3%, with SanDisk jumping 12%.
The prospect of Federal Reserve rate hikes has weighed on overall valuations and certain software growth stocks, while the industrial logic underpinning demand for AI computing power, storage, and data center interconnectivity remains intact despite interest rate expectations. SK Hynix, a memory chip manufacturer, closed up more than 8%, and Coherent, an optical communications company, rose 6.6%. Market participants noted that the strong performance of chip stocks on Friday had "nothing to do" with the non-farm payrolls data, but was instead driven primarily by the ongoing sector rotation between semiconductors and software.
Catalyst Watch: Apple's High-stakes iPhone Event, Oracle, CPI, and the NFL Returns
AI Infrastructure Capex Race: These Tech Stocks Are Stepping up Spending
Cost per million tokens drops below $1! Goldman Sachs: “Volume surge, price collapse” shakes the logic of AI capital expenditure.
Goldman Sachs’ latest report shows that the SDLLMTK index, which measures AI inference pricing, plummeted by 29% in August alone, falling below the $1 per million tokens threshold for the first time and marking a cumulative 50% decline from its peak. Price cuts, user migration to low-cost open-source models, intensifying competition, and local inference are structurally eroding the per-token pricing model. The widening divergence between rising volume and falling prices is dismantling the core narrative that “more tokens equal more revenue.”
Here's How Much $100 Invested In Lumentum Holdings 10 Years Ago Would Be Worth Today
Lumentum Holdings (NASDAQ:LITE) has outperformed the market over the past 10 years by 23.2% on an annualized basis producing an average annual return of 36.64%. Currently, Lumentum Holdings has a
"Welcome to the AGI Era"! OpenAI Releases GPT-6 Astra, Trained on 100,000 GPUs, with Cybersecurity Capabilities Reaching "Critical" Level for the First Time
Astra API is priced at $10 per million input tokens and $50 per million output tokens, approximately 2.5 times the cost of GPT-5.6 Sol. OpenAI stated that Astra has met the company’s internal threshold for “critical” cybersecurity capabilities for the first time. The model can identify previously unknown software vulnerabilities and develop corresponding exploits with minimal human intervention, although its most advanced features are not yet fully available.
Jensen Huang Says AI Is a 'Five-Layer Cake.' Nvidia Is Just One Slice
When Nvidia Corp (NASDAQ:NVDA) CEO Jensen Huang addressed world leaders at the G20, he didn't spend much time talking about GPUs or the company's latest chips. Instead, he offered a simple framework
Cost savings from memory cuts are being redirected to networking! Institutions have crunched the numbers on NVIDIA’s downgraded Rubin Ultra configuration.
① NVIDIA has lowered the HBM memory specifications for Rubin Ultra: from HBM4E 12-Hi to HBM4 8-Hi; ② SemiAnalysis points out that, after adopting this downgraded configuration, HBM costs have been reduced by more than 50%; ③ The focus of value within NVIDIA's internal AI hardware is shifting: no longer solely pursuing maximum HBM memory capacity per card, but instead relying on large-scale cluster interconnects.
Overseas Research Daily 0902 | ING: Hawkish Fed expectations may push the U.S. Dollar Index back to 100
1. Goldman Sachs: South Korea's exports rebounded strongly in August, with chips accounting for the entire month-on-month increase; 2. Goldman Sachs: AI demand remains robust, providing multiple catalysts for semiconductor equipment, memory, and EDA sectors; 3. UBS Group: The structural deficit in the global copper market is widening, with copper prices projected to reach $15,500 per ton by 2027; 4. Deutsche Bank: The next wave of AI investment is shifting toward high-speed connectivity, with a positive outlook on Lumentum and Coherent.
Will the next wave of winners emerge amid the AI boom? Deutsche Bank is bullish on hardware stocks, especially the “duo of laser giants”!
① Deutsche Bank points out that hardware companies in the optical network business are poised to be the biggest winners in the AI era, listing Coherent and Lumentum as its highest-conviction investment picks; ② The bank assigns Buy ratings to five hardware stocks, including Lumentum and Coherent, and Hold ratings to three others, including Dell Technologies.
Complex tasks see cost reductions of up to 45%, with significant enhancements in coding and research capabilities; how Fable 5.1 underpins Anthropic's $2 trillion valuation
On Tuesday, Anthropic released a new version of its flagship AI model, Fable 5.1, enhancing programming and research capabilities while significantly reducing certain usage costs. This move directly responds to pricing pressure from corporate clients and reflects the increasingly fierce market competition among U.S. AI model providers. The most core change in the new version lies in the adjustment of pricing strategy: Anthropic has reduced fees by 75% for the model's re-processing of already handled information, a step the company states will lead to a "significant decrease" in overall operating costs. Dianne Penn, head of product management at the company, pointed out that this segment accounts for part of
OpenAI’s supply cutoff to Cursor triggers reshuffle in AI coding sector: rivalry between tech giants serves as catalyst, while Anthropic ramps up compute capacity to capture market share
OpenAI has announced the termination of its AI model services for Cursor, a coding tool under SpaceX, citing concerns over SpaceX's compliance with service terms and referencing prior breaches following Elon Musk's acquisition of Twitter. The contract will officially terminate on November 12. Currently, OpenAI models account for only about 5% of Cursor's traffic, and Anthropic has indicated it will increase computing capacity to fill the gap.