For me, the neocloud model becomes credible again when it proves cash returns, not just revenue growth and huge backlogs. I would look for high GPU utilisation, positive free cash flow after capex and interest, and contracts that remain profitable as GPUs age and rental prices fall. Customer diversification is also important because dependence on a few hyperscalers increases risk. The strongest evidence would be neoclouds funding more of their next GPU cycle from operating cash rather than repeatedly adding debt. My credibility test: utilisation → cash flow → lower leverage → diversified customers. If these improve while AI compute demand remains strong, the model deserves another look. If revenue rises but debt and capex grow even faster, the core concern remains unresolved.

I would focus on HBM rather than chase a broad memory rebound. SK Hynix still has the strongest technology and market position in HBM, while Micron may offer the more interesting catch-up story if it continues gaining HBM share and securing long-term AI orders. Micron has also said demand should exceed supply through 2027 and 2028, which suggests CXMT’s expansion may not immediately create an industry-wide glut. CXMT is a genuine longer-term threat, especially in conventional server DRAM, but advanced HBM remains much harder to replicate at scale. SanDisk and WDC could recover with strong data-centre storage demand, but that is a somewhat different AI infrastructure thesis and I would rank it behind HBM for now. My preference: SK Hynix technology lead > Micron HBM share gains > SanDi

TSMC’s results make me reluctant to cut semiconductor exposure after the sell-off. Record profit and strong guidance suggest the underlying AI chip demand story remains intact, even if valuations and higher rates are creating much more volatility. Among the choices, I would favour Nvidia first. TSMC’s strong leading-edge demand supports the view that AI accelerator spending has not collapsed. AMD is worth watching, but I would want clearer evidence that its order growth and market-share gains are translating into sustained earnings before becoming more aggressive. The bigger caution is SOXL. A 13% one-day fall shows how quickly 3x leverage can punish investors when semiconductor sentiment turns. Even if the long-term AI thesis remains intact, leveraged ETFs add another layer of timing risk

I would hold QQQ, but avoid adding aggressively at current levels. September CPI at 3.7% shows inflation is still sticky, and higher-for-longer rates are particularly painful for expensive, long-duration tech and AI names. The fact that QQQ fell nearly 2% while defensive sectors held up better shows that rotation risk is real. That said, I would not abandon tech either. AI-related earnings growth remains strong, and selling QQQ after a rate-driven correction risks missing the rebound if inflation cools again. My preference: keep the core QQQ position, but direct some new money towards healthcare, financials and other reasonably valued sectors. This reduces concentration risk without making a binary bet against AI. For me: Hold QQQ + diversify new capital > sell QQQ completely. The next

I picked the S&P 500, a market I already follow. CME offers E-mini and Micro E-mini S&P 500 futures contracts. Buying an S&P 500 ETF such as CSPX gives me an investment whose value tracks the underlying shares, and I can hold it indefinitely. A futures contract is different: it gives leveraged exposure to the index through a contract with a specified expiry, requires margin, and gains or losses are settled as the market moves. That leverage is the biggest distinction for me. Futures let you control much larger exposure with less upfront capital, but losses can therefore build much faster as well. The market I would most like to understand better is the S&P 500, particularly how Micro E-mini futures are priced, rolled between expiries and used for hedging rather than specul

I practised Fibonacci retracement on VWRA, using a recent swing low to swing high. I would mark the 38.2%, 50% and 61.8% retracement levels and watch whether price finds support around them rather than treating Fibonacci as a prediction by itself. For a hypothetical long setup, my first target would be a retest of the previous swing high. My second target would be the 1.618 Fibonacci extension if momentum continues. The stop-loss would sit below the swing low, where the original bullish setup is clearly invalidated. The product I would like a deeper walkthrough on is VWRA. I invest in it regularly, so I would be interested in how Fibonacci, support/resistance and trend indicators can be used without turning a long-term DCA strategy into short-term market timing.

My one-line trend rule: I consider a market in an uptrend when it consistently forms higher highs and higher lows, with price holding above its medium-term moving average. Before entering any trade, I would write down the price level that invalidates my thesis. If that level is reached, I should reassess or exit rather than move the goalposts simply because I do not want to take a loss. I am more of an investor than a trader. I prefer diversified ETFs, regular DCA and a long time horizon rather than trying to predict short-term price movements. I may pay attention to earnings, valuations and market cycles, but they mainly influence how cautiously or aggressively I deploy capital, rather than changing the long-term plan.

I picked the CME Micro E-mini S&P 500 (MES). Contract size: US$5 × S&P 500 Index Minimum tick: 0.25 index point Tick value: US$1.25 Expiry: Quarterly, on the third Friday of Mar, Jun, Sep and Dec. At an index level around 7,765, one contract has about US$38,825 notional exposure. • 1 tick = US$1.25 • 1% move = 77.65 points × US$5 = about US$388.25 gain or loss Margin varies over time and by broker, so I would check the current requirement before trading rather than treat it as fixed. I would be more comfortable with MES than the full-size E-mini. The smaller multiplier gives much better control over position sizing and losses while still providing S&P 500 exposure. Leverage still means even a Micro contract needs strict risk management.

The data still looks closer to “spending as usual” than broad cutbacks. JPMorgan’s debit and credit card spending rose 10% YoY, while Citi reported US card spending up 11%. Credit quality is holding up too, with JPMorgan’s card charge-off rate falling to 2.91%. But I would not interpret that as consumers feeling comfortable. Persistent inflation and high borrowing costs are squeezing affordability, and banks are preparing for risks beneath the surface. For me, the warning signal would be spending slowing at the same time as delinquencies and charge-offs start rising. Until then, the US consumer looks resilient, but increasingly dependent on employment and wage growth staying strong.

A bank selling off more than 5% after an earnings beat tells me the market is trading the next few quarters, not the quarter just reported. Citi’s headline numbers were strong, but investors may be focusing on what comes next: credit costs, loan growth, funding margins and the impact of higher long-term yields. Jamie Dimon’s warning that forces are “shifting under the surface” adds to that caution. To me, the sell-off does not necessarily mean the results were bad. It suggests expectations were already high and investors are questioning whether current earnings are sustainable. If banks keep falling despite beats, that becomes more meaningful. It could signal that the market is pricing slower growth or rising credit risk before those pressures become obvious in headline earnings.

For the rotation out of memory into compute to last, I would want three confirmations. First, TSMC and the hyperscalers need to maintain or raise AI capex guidance. That would support Nvidia, AMD and other logic names even if conventional memory pricing becomes less certain. Second, compute earnings need to show that demand is translating into margins and cash flow, not merely huge AI spending. That matters increasingly as investors question the returns being generated from AI infrastructure. Third, memory needs a genuinely weaker fundamental signal. China's DRAM progress alone may create a valuation scare, but it does not necessarily mean near-term oversupply. If HBM remains tight and DRAM contract prices continue rising, money could rotate back quickly. So my test is simple: strong AI ca

If ASML’s order book had disappointed, I think the damage would have spread well beyond semis. ASML sits near the front of the AI capex chain. Weak bookings would raise questions not only about chipmakers, but also data-centre equipment, power, cooling and other infrastructure names whose valuations assume years of sustained AI investment. The second-order hit could reach the mega-cap hyperscalers if investors start questioning whether their enormous capex plans will generate adequate returns. But ASML actually delivered the opposite signal. Customers are raising equipment forecasts, order momentum is increasing, and ASML says planned supply for 2027-28 may not meet demand. Some customers are even discussing 2029-30 orders. That makes TSMC the next confirmation test. If TSMC also maintains

The 3.6% headline worries me more because another acceleration from 3.4% would strengthen the “sticky inflation” narrative and keep pressure on long yields. But the bigger market-moving risk is a core CPI miss. If core comes in at 0.2% MoM as expected, that would still show some underlying cooling. A surprise back to 0.3% or higher would be much harder for the bond and equity markets to dismiss, especially with valuations already sensitive to higher-for-longer rates. So: headline 3.6% is the bigger concern, but core above 0.2% is the bigger potential shock.

I’d vote A: stronger-than-expected corporate earnings. At this stage of the bull market, valuations already reflect plenty of optimism around AI and future growth. To push the S&P 500 sustainably higher into Q4, I think companies need to prove that earnings can catch up with those expectations. AI monetisation remains important, but it ultimately needs to translate into revenue, margins and free cash flow. Lower inflation and rates would help valuations, but easing driven by economic weakness could be a mixed blessing. So Q3 earnings are the key test for me. Strong earnings plus resilient guidance would justify higher prices. If earnings disappoint, high valuations leave much less room for error, regardless of the AI narrative.

I see Chinese competition as the greater long-term risk. AI funding constraints are serious, but they are ultimately cyclical. If returns on AI infrastructure improve or financing conditions ease, spending can accelerate again. Chinese memory capacity is different. Once CXMT achieves competitive yields and scale, that supply does not simply disappear when the cycle turns. It could create persistent pricing pressure, particularly in conventional DRAM, and reduce the scarcity value enjoyed by incumbent producers. HBM and leading-edge products provide some protection because technology, yields and customer qualification remain substantial barriers. But over time, Chinese competitors can climb that ladder too. So I would frame it this way: AI funding determines how strong the next demand cycl

My conviction is lower, but not broken. CXMT changes the long-term risk premium for memory. If it can turn 4F² into competitive, high-yield mass production, Chinese supply could eventually pressure conventional DRAM pricing and challenge the industry's current structure. But announcing an architecture is very different from producing server-grade DRAM economically at scale. Yield, reliability, customer qualification and volume are the real tests. Until CXMT demonstrates those, I see the sell-off as the market pricing in a credible new risk rather than proof that the memory cycle has fundamentally changed. Hock Tan's warning matters too. If capital becomes the constraint on AI build-outs, demand expectations could cool even before new Chinese supply arrives. So I would still hold quality me

I’m watching provisions more closely than NIM. Higher rates can support net interest income, but that benefit is fairly visible and can be partly offset by higher deposit costs. Provisions tell us something more important: whether banks themselves are seeing deterioration beneath the surface. If NIM stays healthy while provisions remain contained, that is a strong combination for banks. But if provisions jump sharply, especially across both consumer and commercial books, I would treat that as an early warning that higher rates are finally feeding through to credit stress. So for me: NIM drives earnings, but provisions drive the risk outlook. In this environment, the second matters more.

I think the divergence can hold through CPI, but probably not indefinitely. The semiconductor sell-off is being driven by a sector-specific issue: fears that CXMT's DRAM advances could eventually weaken memory pricing power. CPI, meanwhile, is a macro catalyst affecting rates and valuations across the whole market. Consensus expects September headline CPI at 3.6% YoY and core CPI at 2.5%. A softer core print could pull yields down and give semis a strong relief bounce, especially after the sharp sell-off. But a hot print would be more dangerous for chips because they would face two pressures at once: higher discount rates plus renewed concerns about Chinese memory competition. So my base case is that CPI determines whether this divergence narrows or widens. Soft CPI = semis likely catch up

My vote: Citigroup (C). Citi looks like the most interesting upside candidate this earnings round. Expectations already point to strong YoY profit growth, but the bigger question is whether its restructuring can finally deliver better operating efficiency and sustainable returns. JPMorgan remains the quality leader, but Citi may have more room to surprise if revenue growth, cost discipline and credit quality all hold up. With bank valuations pressured by higher Treasury yields, a convincing earnings beat plus solid guidance could produce a stronger market reaction. I’m watching NII, expenses, credit provisions and management’s outlook most closely.

I would not call the DRAM floor from the share prices alone. The signal I want is pricing plus inventory. Interestingly, the underlying DRAM market still looks tighter than the sell-off suggests. TrendForce expects conventional DRAM contract prices to rise another 10–15% QoQ in Q4, while supplier inventories remain very low as capacity continues shifting towards server DRAM and HBM. So my floor signal would be when spot prices stop weakening, contract-price expectations stabilise, and PC/mobile inventories begin falling without suppliers increasing conventional DRAM capacity aggressively. That would suggest supply discipline, rather than simply another temporary bounce. The biggest risk is that HBM profitability encourages too much new capacity that eventually spills back into conventional