1The $1trn tech drawdown is a fight over whether this cycle's chip earnings are durable
The week closed with the Nasdaq down 4% and roughly $1trn of market value gone in two sessions, and the chips took the worst of it — , and among the hardest hit. The sharpest tell was Micron: it beat, crossed $1trn in market value, traded up 20% after hours and carried the Nasdaq 2% higher overnight — and the whole move was gone by the opening bell. In Korea, the KOSPI, which had roughly doubled this year on the memory names, hit a 10% limit-down as a leveraged position unwound. That is forced selling, not a change of view.
Joseph Wang, on Fed Guy, reads the fade as an "earnings bubble" starting to deflate, a thesis he credits to Peter Berezan at BCA. The claim is that semiconductor earnings are transitory rather than a new level — a V-shaped top of the kind the pandemic beneficiaries printed — with new fab capacity and hyperscalers' in-house silicon compressing margins from here. The sequencing is the load-bearing part: the price falls first and the downgrades arrive afterwards, so a name still carrying buy ratings on the way down is what the thesis predicts, not a refutation of it. He adds a demand leg to it. Chinese open-weight models, GLM 5.2 among them, are now good enough to threaten the revenue of the closed US labs carrying trillion-dollar valuations, and efficiency gains are showing up in GPU-leasing indexes where compute demand is no longer surging.
Felix, on Felix & Friends, watched the same week and landed on the opposite conclusion. He describes a "philosophical war" between one camp that thinks big tech is overspending on AI with unproven returns — and is selling the capex names, among them — and another that thinks it is early and the infrastructure builders win. His own read is that the pricing is internally inconsistent: , and trade as though AI were peaking, on valuations he calls historically low against businesses still growing, while small speculative AI is priced as though the boom runs to 2030. Where the next breakthrough happens is contested too — he flags two top-ranked Google researchers leaving for OpenAI and Anthropic, calling too bureaucratic, on the same day CNBC framed Google's position as burnishing its best weapon in the AI-supremacy fight.
Two caveats travel with this. The video evidence here is two channels, both Saturday uploads; four of the six channels covered here published nothing in the window, so this is two readings rather than a chorus. And Felix shows no fundamentals work behind the structural read, in a video that doubles as a funnel to a free session and a paid report, with "your portfolio is broken" framing — the argument can be judged on its merits, but the urgency is sales copy.
What separates the two readings is margins, not the tape. If new capacity and in-house silicon compress them, a beat that round-trips by the opening bell is the pattern rather than an oddity; if they hold, that same fade is the discount Felix is describing. Nothing in the week's price action distinguishes the two.
2The memory price behind the quarter's best earnings is now a cost the device makers are passing on
RAM prices have risen roughly four-fold in two years. On one side of the trade that is the earnings story — Micron's beat sits on it. On the other it is an input cost, and it has now reached the shelf: Apple raised product prices citing RAM-cost inflation, with down about 5% on the day in Wang's account, and Xbox and Switch hardware in the same position. fell in the same week on memory-chip-sector weakness, per Google News.
The tape here is contested. Yahoo's session summary carries the identical RAM headwind but has Apple up 3.14%, against Wang's −5% for the price-hike reaction, and nothing in the day's material reconciles the two. The mechanism is the part they agree on, and it runs one way: the same input price that lifts the producers' earnings is a cost to the device makers, and the device makers are now handing it to the customer. That is where a demand drag would begin, and it has not shown up in volumes yet.
3The week's largest one-day loss came from a company that now mostly sells AI data-centre capacity
SpaceX lost $400bn of market value in a single session — the second-largest one-day wipeout on record. Its IPO popped about 48% and then gave all of it back, and it issued a bond immediately afterwards, which raised dilution fear against a cash-burning business trading at roughly 100 times revenue. The mechanism has little to do with rockets: on Felix's account the majority of revenue now comes from selling AI data-centre capacity to Anthropic and Google — a "Neocloud" — so the equity moves with the AI narrative, and he calls the most rate-exposed stock in the market.
Two structural facts sit ahead of it, pulling opposite ways. Lockup expirations are still pending, which is supply. NASDAQ-100 inclusion means index trackers, among them, have to buy it regardless of any of the above, which is demand that arrives on a schedule rather than on a view.
The operating business, meanwhile, kept selling: rose 1.71% on a Starlink in-flight Wi-Fi deal. And the financing worry is not confined to one name — had its worst week since the 2001 dot-com bust as AI-financing concerns escalated, per CNBC. What is being repriced across both is the cost of funding the build-out, not the demand for what it produces.
4Rate-hike talk has replaced cut expectations, and the debasement trade broke as the dollar ran
With inflation running near 4%, Felix's read is that cuts are off the table and a hike is the growing expectation — "kryptonite for tech", in his phrase, which is the transmission line from rates into the drawdown described above. Wang supplies the institutional detail: the new Fed chair is deliberately opaque, and treasury-connected voices are floating a hike outright.
The currency is moving with it. USD/JPY broke 160, and the dollar is at its highest since late last year. The debasement trade unwound in step — gold below its 200-day moving average, silver and bitcoin falling hard. Flows inside equities point the same way, with healthcare names running to record highs on CNBC's account, which is what a defensive rotation looks like.
One divergence in that picture was already breaking as the videos published. Wang notes oil having fallen back roughly to pre-Iran-war levels while yields stayed high, and treats the pair as a puzzle. The same morning's news feed carries a tanker struck in the Strait of Hormuz, renewed US military strikes against Iran, and Iran saying it struck US-linked targets — the worst escalation since the peace deal. Both Saturday videos describe oil as calm; that is the first thing in the day's material to go out of date, and it means the energy and geopolitical read carries live headline risk into the next session.
What the sources recommended
One buy lean and five names flagged to sell or avoid came out of two videos and the news feed, and the buy is not phrased as a buy by the person making it.
The buy lean. , and , from Felix, framed as real-cash-flow AI infrastructure unfairly beaten down — valuations he calls historically low against businesses still growing — with the line that "skilled investors start writing checks when there's blood in the streets." He never says buy outright, so it is a buy-on-weakness lean rather than an explicit call. It also collides head-on with Wang's earnings-bubble case on the same names, which is the day's central unresolved disagreement.
The sells. from Felix, on the bond raise, the pending lockups, the cash burn at roughly 100 times revenue and the rate exposure. from CNBC, on its worst week since the 2001 dot-com bust as AI-financing concerns escalated. from Wang, down about 20% on earnings as AI compresses professional-services margins, alongside Stripe data he reads as a solo-entrepreneur wave. from Wang, on the RAM-cost price hikes. And semiconductors as a group under the earnings-bubble thesis — a sector-level call from one source, not a set of single-name exits.
Monday opens into quarter-end selling with Hormuz live
The next session is Monday 29 June, and the quarter ends the following day. Wang flags the rebalancing that goes with it — out of equities, into bonds — which is a mechanical headwind that has nothing to do with either side of the chip argument and will be indistinguishable from it in the tape. reports Tuesday after the close, with consensus EPS at 3.27 on the Finnhub calendar. faces what Yahoo calls a vital Agentforce test, and carries an analyst endorsement into its aerospace spinoff debut, per CNBC. If the Hormuz strikes continue, the energy and defence read is where it lands first, and the calm-oil framing from Saturday's videos is already the weakest claim on the page.
