D
+10
From optical interconnect veteran to all-in 'picks-and-shovels' supplier: Why LITE?
The stock price surged nearly 10x within a year, pushing market cap above $70bn. It drew NVDA to invest $2bn for an equity stake in Mar 2026.
+10
Deep Research
D
+20
IBKR: The Quiet Master of Ultra-Efficient Trading
In this cyclical, market‑driven brokerage industry, Interactive Brokers ($Interactive Brokers(IBKR.US)) is an outlier: it barely advertises, and its terminal interface long deterred retail novices. Brand awareness was so limited that it once failed to crack the top 10 in investment app download rankings.
Compared with the meteoric rise of HOOD over the past five years and the deep brand equity of SCHW, IBKR appears understated. Yet it has consistently delivered 70%+ OPM and approx. $2 mn revenue per employee per year. Dolphin Research believes the edge lies in 'licenses and technology'...
+20
D![Google AI]()
![Google's Network Evolution Summary Generation ICl domain-chips (Scale-Up) Cross-]()
![海豚投研 DolphinResearch TPU v7 海豚投研 olphinPesearch Dolphin Ironwood semianay Tray y]()
![L T 5 2 X 1 1 投研 Res Res or 1 1 se M A W Source： Google]()
+27
Can GOOGL's 'Optical Net' take on NVDA's dominance?
In the previous piece by Dolphin Research, we compared two third-party GPU vendors (3P GPUs) — $NVIDIA(NVDA.US) and $AMD(AMD.US) — on data center networking. NVDA's GB300 was a clear overmatch.
AMD managed to scale up from 8 to 72 GPUs using Broadcom switches, delivering Helios. However, by the time it shipped, the competitor had already moved to Vera Rubin.AMD achieved parity in GPU interconnect bandwidth, but the cost economics were unfavorable and the engineering design lagged. AMD only just cleared the passing line...
D![a 2 1 三 c )SanDisk PROFESSIONAL]()
![Sandisk revenue Unit： $mn Flash Revenue Y/Y 25，000 arC 200% 20228 20，000 Dolphin]()
![Sandisk Shipment Volume Unit：EB SNDK estimated(EB) 3) Y/Y 140 120 DolphinResearc]()
![Sandisk ASP Unit： ($/GB) SNDK Flash $/EB assumptions ($/GB) V Y/Y 0 0.20 hinRese]()
+34
If NAND is 'born prolific,' how does SanDisk defend an 80% GPM?
NAND is in a tight balance, with the call squarely on demand. Any let-up in supply discipline could cause the market to unravel.
D![High Bandwidth Memory HBM]()
![20 USD/GB 18 16 14 12 10 8 Fig. 5： DDR5 spot vs contract price 6 Spot price 4 2]()
![1，000 1，100 Stock 900 800 700 600 DolphinResear 500 400 SK Hynix 300 200 100 Sam]()
![Parameters of Qwen Flagship Models DRAM Needed to Store Model Param. 3，000 Dolph]()
+17
Storage Takes the Baton from Compute: Why HBM Stands Out?
From today's vantage point, H2'25 was a pivotal inflection for the AI hardware stack. Until then, 'compute', led by Nvidia GPUs, was center stage. After that, 'storage' — DRAM, NAND and HDD — became the biggest bottleneck across the chain.
As the chart shows, DDR5 DRAM prices surged roughly 7–9x within six months after Sep-25, with NAND up 4–5x over the same period. Correspondingly, ...
D![AI]()
![Metric/Protocols NVLink 5.0 NVLink 6.0 UALink1.0 UALoE Bandwidth per lane/ link]()
![Figure 1： Total AI/ML Networking Market Other (PCIe， UALink， 一sco兰云凹u3cu>u& Othe]()
![AMD MI300X Intra-tray Connectivity AMDMI300X INFINITY PLATFORM" 8x AMD Instinct"]()
+32
AMD vs. NVDA: Can Helios Measure Up?
In the market for commercially sold GPUs, competition is essentially a two-horse race between $NVIDIA(NVDA.US) and $AMD(AMD.US). In the MI300 era, AMD captured little of the AI compute upside due to the lack of a system-level solution, with interconnects at its core.
Now Helios is here. That raises two questions: 1) does AMD truly have the capability to go head-to-head with NVIDIA, and 2) as interconnect schemes compete and iterate, how will supply-chain dynamics shift, who benefits, and who loses? We will examine these issues through the lens of interconnect.
D![SanDisk a ISK]()
![AI compute boom HDD supply gap Enterprise eSSD surge Shifts to eSSD Direct hit·e]()
![Sandisk stake entering 2026 7，513，019 shares Debt-for-equity exchange Feb17-19，2]()
![SanDisk Shareholding structure(2025/12/31) FRMLLC Vanguard Group BlackRock Weste]()
+11
AI inferencing boom: Can beleaguered WDC rise from the ashes?
This AI wave has turned NAND flash from a cyclical, oversupplied commodity into a scarce strategic resource. SanDisk, spun off and freed from the drag of the legacy HDD business, is making a pure-play bet on NAND and data-center SSDs, catching the updraft.
AI servers, when running inference, processing complex tokens and prompts, handling multimodal inputs (video, images), and generating synthetic data, need to read and write historical data at high frequency and in tight loops. Legacy cloud storage architectures can no longer keep up with these workloads...
D![TM]()
![40% 35% 30% 25% DolphinResearch 20% SBUX Mobile order&pay Share in US 15% 10% 5%]()
![Starbucks store count estimation 25，000 DolphinResearch DolphinResear Research s]()
![Starbucks store density by region hinResearch DolphinResear DolphinResearch 店/10]()
+7
SBUX: Retreat from China—Can the Coffee King Rebound on Home Turf?
U.S. comps have turned negative in recent years, China has been 'lost', and growth has stalled.
How much is this cash cow still worth?
D![图片 1，共 14 张]()
![Connectivity in an AI datacenter Frontend network/WAN 1 DC-to-DC connectivity/ S]()
![Fat-Tree Topology L2 L2 L2 L2 SW SW SW SW L1 L1 L1 L1 L1 L1 L1 L1 SW SW SW SW SW]()
![Network Scaling Challenges & Medium arch bsarcn arc Tier Physical Scale Latency]()
+10
AI-era DCI: Beyond the chip, compete as a network. Is there a China play?
Under the Scaling Law, model parameters are ballooning and multi-turn dialogue requires ever more state. Single-card compute and on-card memory can no longer support SOTA model inference and training, so AI workloads must run on ultrascale clusters built from massive numbers of xPUs.
This is the AI interconnect fabric. It links single cards into clusters and 'stitches' tens of thousands of heterogeneous chips, within and across clusters, to operate as one machine.Recent roadmaps from GPGPU and AI ASIC vendors make this clear. The industry's focus has shifted from chasing per-card peak compute to the engineering of large-scale clusters...D![SHEIN]()
![Shein net revenues by type Product revenues Service revenues Service revenues(%)]()
![Shein total revenues Total net revenues yoy% 450 DolphinResearch 41% Research Do]()
![Shein revenue growth drivers Active customers Total orders Annulized ASP($) 12 e]()
+15
SHEIN: Temu's assault and tariff headwinds—can the $100bn unicorn reinvent itself?
This piece tackles three questions—1) By dissecting the prospectus, what is SHEIN really, where does its revenue come from, and is it profitable?
2) Everyone talks about the small-batch, fast-turn model—how does it actually work, where are the moats, and why can't Zara, Temu, or Amazon replicate it?3) With an IPO imminent, what valuation looks rich, and what price sits in the 'strike zone'?