Oracle reported earnings today. I opened its chart on SoDEX — the stock was down nearly 12% in pre-market trading. sodex.com/join/JELLYZ This is not an isolated case. Over the past week, three key companies across the AI infrastructure chain reported earnings, and the market reaction was surprisingly consistent: - June 3: Broadcom reported FY2026 Q2. Results beat — the stock sold off sharply. - June 4: Ciena reported FY2026 Q2. Results beat — the stock also fell sharply. - June 10, after the close: Oracle reported FY2026 Q4. Results beat — the stock dropped anyway. Three earnings reports. None of them were bad. Three stock reactions. None of them went up. So what went wrong? First, Oracle: strong demand, but heavy financing pressure. The numbers were solid: Q4 revenue of $19.18B, up 21% YoY; adjusted EPS of $2.11; OCI cloud infrastructure revenue up 93% YoY. The problem is not demand. The problem is investment. In FY2026, Oracle's CapEx reached $55.66B while free cash flow was -$23.69B. FY2027 CapEx could reach as high as $95B, funded by continued debt and equity financing for AI data center construction. Oracle's central tension is clear: AI cloud orders are strong, but fulfilling them requires massive upfront spending on data centers, GPUs, networking, power, and land. The market isn't worried about whether demand exists — it's worried about whether these orders can generate a high enough return on capital, and when free cash flow will turn positive again. Then Broadcom: the business is strong, but expectations are too high. Revenue was $22.19B, up 48% YoY; AI semiconductor revenue was $10.8B, up 143% YoY. Very strong — yet the stock still fell. Expectations for core AI suppliers have become extreme. Broadcom guided Q3 AI chip revenue to around $16B — strong, but not enough for the market's more aggressive hopes. Broadcom didn't fall because AI ASIC and networking demand is weak. It fell because the stock had already priced in too much of the future. Finally, Ciena: revenue beat, but the trade was too crowded. Revenue was $1.57B, up 40% YoY; adjusted EPS was $1.64, up 290%; full-year revenue guidance was raised to around $6.3B. Again, not a bad report. Ciena's problem: the market had long been trading it as a core beneficiary of AI optical networking — AI data centers need higher-bandwidth, lower-latency optical connections, and Ciena sits directly in that part of the chain. But after a sharp year-to-date rally, the bar for another positive surprise was simply too high. Three companies, one signal. Oracle provides cloud and databases. Broadcom provides ASICs, AI networking chips, and infrastructure software. Ciena provides optical networking and data center interconnects. Placed inside the AI infrastructure chain, they are links in the same chain: AI data center construction → AI ASICs / networking chips → optical networking / interconnects → cloud revenue and compute monetization Look at all three together and the signal is clear: AI infrastructure trading is moving from phase one to phase two. In phase one, the market bought the narrative: who has AI orders, who is in the chain, who benefits from data center expansion. In phase two, the market buys verification: can orders turn into revenue, revenue into profit, profit into free cash flow? Can CapEx generate enough ROIC? Is valuation already stretched? Can guidance keep moving higher? The selloffs across all three companies show that the market will keep trading AI — but it will no longer blindly reward every AI infrastructure company. Going forward, the real upside may belong to two types of companies: the surest recipients of AI CapEx dollars — GPU, ASIC, HBM, networking, and power chain companies — and the operators that can prove CapEx returns by turning orders into revenue, profit, and free cash flow. If you also follow U.S. AI stocks, you can view and trade related U.S. equity contracts on SoDEX, including $ORCL, $MU, and other AI infrastructure names.

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use. Its capabilities exceed those of any model we’ve ever made generally available.

Why should AI stock investors pay attention to AI CapEx? AI capital expenditure refers to the fixed-asset investments cloud giants like Amazon, Microsoft, Google and Meta — plus some vertical AI players — make in GPU clusters, data centers, networking, storage and power infrastructure. In essence, it's the starting point of the entire AI supply chain. Once hyperscalers raise CapEx, capital flows down the chain: GPU / ASIC → HBM → AI servers → networking equipment → data centers → power infrastructure → cloud revenue That's why the pace of AI CapEx directly reflects compute demand, supply-chain orders, application innovation and the ability of AI products to scale. Looking at the data, the CapEx cycle of the four major cloud giants (Amazon, Microsoft, Google, Meta) splits into four stages: 2011–2023: Traditional cloud expansion. CapEx was driven by enterprise cloud migration, SaaS, video, advertising, e-commerce and storage. 2024: An extraordinary acceleration begins. After being down 2.5% YoY in 2023, combined CapEx jumped 54.8% to $228.4 billion. Post-ChatGPT and GPT-4, AI infrastructure had firmly entered the tech giants' budgets. 2025: The arms race confirmed. CapEx grew a further 64.6% to $376 billion — proof that 2024 wasn't a one-off rebound, but the start of sustained expansion in AI compute demand. 2026: Explosive growth. On current guidance, combined CapEx could reach $710 billion, up nearly 89% YoY. This is no longer an extension of the cloud cycle — it's a massive buildout as tech giants race to secure the next generation of compute. And the expansion is far from over. As free cash flow gets consumed by CapEx, the giants are leaning more on external financing: Alphabet recently moved forward with an equity raise of around $80–85 billion, and Meta is exploring more options to fund its data center buildout. So where is the money going? 1. AI chips and accelerators — Nvidia and AMD GPUs, Google TPUs, Amazon Trainium, Microsoft Maia. The most visible part of the spend. 2. HBM, DRAM and enterprise SSDs — model parameters, training data and inference cache all need high-speed memory; the stronger the GPU, the greater the demand. 3. AI servers and rack-scale systems — hyperscalers buy full servers and, increasingly, rack-scale systems like GB200 and GB300, not individual GPUs. 4. Networking and optical modules — training spans thousands of GPUs, so switches, NICs, optical modules and interconnects become critical. 5. Data centers and power — land, buildings, liquid cooling, transformers, grid connections and long-term power agreements, all built for high power density. So AI CapEx isn't just about buying GPUs — it's about building an entire "compute factory." That's why the AI trade has widened from Nvidia to HBM, memory, servers, optical modules, data centers, power equipment and liquid cooling. For investors, the real question isn't how much the giants spend — it's whether that spending converts into large, sustainable AI revenue. Short term, CapEx means supply-chain orders; medium term, cloud compute capacity; long term, the winner won't be whoever spends the most, but whoever turns each dollar of CapEx into the most revenue and profit. This cycle may look like a model race on the surface. Underneath, it's a race for compute, power, memory and data center capacity. Therefore, the Big Four cloud giants — along with the core suppliers capturing the largest share of AI CapEx — are the key players in this AI infrastructure cycle. If you want to invest in this theme, you can trade them on @sodex_official such as $GOOGL, $MSFT, $MU and $SNDK. sodex.com/join/JELLYZ #SoDEX #SoSoValue #AI

Memory stocks like SK Hynix and Samsung have recently pulled back more than 15%. What is the market worried about? 1. Fear of weaker DRAM demand from customers like Nvidia. Nvidia's Rubin NVL72 is reportedly set to cut total DRAM capacity per rack from an originally planned 55TB to 28TB, while memory per cabinet may drop from 192GB to 96GB. The market has read this as a sign of weaker memory demand, triggering fears that the expected "volume and price upcycle" in memory may fail to materialize. But in our view, the market has overreacted. The reason for this adjustment is not weak demand — it's supply bottlenecks. First, due to tight allocation and extremely low packaging yields, SK Hynix and Samsung cannot supply enough 192GB memory modules to meet demand. To ensure every Rubin GPU has sufficient DRAM, Nvidia is proactively managing supply-chain risk by lowering memory per cabinet. Second, Nvidia's original reference specification was already very high, and the new architecture is highly flexible. Even with a reduced factory-shipped configuration, end users can still provision additional memory capacity through cloud providers if needed. So this adjustment does not materially change the reality of tight supply or the growth trajectory of memory demand. 2. Macro pressure: Fed / BoJ rate concerns and a heavy data calendar. Concerns are rising that the Fed may stay restrictive for longer and that the Bank of Japan may keep hiking. The remarks and policy stance of the new Fed Chair, Kevin Warsh, are worth watching closely. The upcoming U.S. CPI release on June 10 and the FOMC meeting on June 16–17 could further fuel risk-off sentiment. 3. A big prior rally, high expectations, and a market waiting for fresh catalysts. Memory stocks delivered a blistering rally in April and May. The market now needs fresh industry-specific catalysts to support further upside. The sharp sell-offs in Broadcom and Ciena after strong earnings reports are a case in point — which makes Micron's earnings on June 24 worth watching closely. In summary, June is a macro-driven month, but the fundamentals of SK Hynix and Samsung remain solid. Unless clear reversal signals emerge, investors can keep buying the dip in $SKHXUSD and $SAMSUNGUSD on @sodex_official and wait for macro risks to ease. sodex.com/join/JELLYZ #SoDEX #SoSoValue