Key Takeaways
- The S&P 500 fell 1.43% to 7,674.37, breaking its 50-day average but holding above its 200-day average.
- The Nasdaq 100 fell 2.45% to 29,309, finishing on its 50-day average as investors moved away from the basket.
- The semiconductor ETF SOXX fell 5.52% to $520.04, now 21% below its June high.
- Amazon's free cash flow fell to a $7.6B outflow after AI-related property and equipment spending rose $66.1B.
- Nvidia's earnings guidance will test the assumption that capital can stay cheap for the AI buildout.
This week's equities report examines a market reshuffle. Money is moving from crowded AI names into the supporting infrastructure of cloud, data, and power. The major indexes declined, breaking short-term momentum.

The equity market is telling one story with many voices. Money is moving out of the AI names everyone already owns and into the parts of the trade that are harder to see. Investors are reshuffling their hand, folding away from crowded semiconductors and doubling down on self-funded cloud, data, storage, power, and security infrastructure. Sometimes, the specific bets look nothing alike. While some are leaning into utilities and energy, others are bidding defense equipment or storage. Underneath all those positions we find portfolios structured to answer the same question: who owns the cash, power, storage, and distribution needed to fund AI?



SPX — The S&P 500 fell 1.43% to 7,674.37 and ended its first losing week since late July. The S&P 500 bet is one that can continue paying off while hiding the fragility sitting underneath, since its market weights concentrate exposure in the companies spending most heavily on AI. The setup can make money while long yields remain contained. However, this might as well obscure the trap of getting a favorable result for the wrong reasons, given that passive demand can lift the index even when breadth and household demand weaken. This week’s decline was large enough to break short-term momentum, but not the broader uptrend: SPX closed below its 50-day average near 7,731 and above its 200-day average near 7,601. The driver was higher long yields reducing the value of distant AI returns, while oil and weak household demand raised the risk that inflation and slower consumption arrive together.

NDX — The Nasdaq 100 fell 2.45% from 30,046 to 29,309 and finished almost exactly on its 50-day average of 29,309. The classic “concentration builds wealth; diversification preserves it” explains why many portfolios are opting for staying neutral on the basket while doubling down on selected constituents. The Nasdaq 100 bundles software margins, cloud cash flow, chip scarcity, and long-duration rate expectations into a single index; buying the whole thing, therefore, dilutes the exact exposures worth holding. QQQ lost about 2.4%, with most of the damage arriving on Tuesday as semiconductors, memory and other AI suppliers unwound. A move back above 30,000 after Nvidia reports would repair the structure; a weekly close below 29,000 would establish a lower high and shift attention toward the late-July base.

SOXX — SOXX fell 5.52% to $520.04. SOXX now sits about 21% below its June 22 high near $656, which places the sector in a correction even though its net asset value remained up roughly 74% for the year through Thursday. Semiconductors sit at different points in the cycle. That’s why it makes sense to concentrate bets and move exposure around as events unfold. The index treats many heterogeneous bets the same: compute (NVDA), custom chips and networking (AVGO), second-source compute (AMD), cyclical memory (MU), or capacity machines (AMAT).

Mag7 — The equal-weight Mag7 proxy MAGS fell 1.44% to $67.28, but it outperformed QQQ by about one percentage point and SOXX by more than four. The basket was not the center of the selloff. Suppliers and memory names were. Still, justifying a basket position is getting harder to defend, as the seven companies no longer share a thesis. Alphabet, Microsoft, Meta, and Amazon fund AI from large existing businesses. Nvidia sells the equipment and now supports customer credit. Apple owns distribution while avoiding the largest infrastructure bill. Tesla prices autonomy far above current vehicle economics. For all those reasons, 44% of 6,371 filers cut Mag7 exposure while 42% added; meanwhile semiconductors still had more buyers than sellers.

IWM — IWM fell 1.64% to $300.10 after reaching a 52-week high of $305.18 on August 14. Small caps remain a trade on domestic credit, refinancing, and operating leverage. Lower front-end rate expectations help immediately, but the 10-year near 4.74% and the 30-year above 5.2% still set borrowing costs for the businesses inside the index. That is why IWM did not hold its high even as investors continued to price eventual Fed relief. Small caps need credit, not only lower policy expectations. The Russell 2000 is a bet on domestic credit, refinancing, bank lending, and operating leverage. It can also be seen as a tactical rates trade. Small companies feel short rates, bank lending, wage costs, and domestic demand faster than large multinationals. Lower front-end yields reduce refinancing pressure immediately, while revenue and interest coverage improve much later. Short covering after weak data is exactly the type of scenario that produces a right outcome for the wrong reason.

AMZN — AMZN fell about 1.5% with no fresh company event. This was a common name on most Q2 13-F filings. Few companies can turn an existing customer base into several businesses at once. Amazon is the leading one doing this at scale. They have distribution, logistics, advertising, subscription, and cloud. Retail and ads create customer data and cash; AWS supplies the highest–margin growth engine; the entire group reinvests heavily on the AI buildout. Morgan Stanley presented a path to $500 by end-2027. That thesis rests on requiring AWS’s current run rate to become the opening leg of a move toward roughly $1T of annual revenue over 8-10 years, which implies around 20–25% annual growth for most of a decade. Amazon’s Q2 is a good start: AWS grew 37% to $42.2B, a $169B annualized pace, and produced $16.6B of operating income. On the flip side, the bill is equally large. Trailing operating cash flow rose 33% to $161.4B, but free cash flow fell to a $7.6B outflow after AI-related property and equipment spending increased $66.1B year over year. The bull case needs utilization and high-margin consumption to outrun depreciation, power, and memory costs.

AAPL — AAPL gained about 1.1% and was one of the few Mag7 names to finish higher. Like Amazon, Apple also owns strategic assets that compound a strategic advantage beneath the surface. Apple owns devices, operating systems, identity, payments, app distribution, custom silicon, and user permissions. That control lets Apple adopt outside models, move work between devices and private cloud, and wait for inference costs to fall. Investors own it for recurring services, customer retention, buybacks, and the option to make AI useful without owning the largest model.

NVDA — NVDA traded ~4.6% lower on the week prior to earnings. Jensen is a man on a mission: for some he is the hero securing necessary resources for the age of superintelligence; others see him as a villain that only recently got an X account to stir the market with propaganda.

Regardless, Nvidia is the focal point to stress-test the assumption that capital can stay cheap and available for longer. Morgan Stanley estimates about $2.9T of global data-center investment through 2028. With roughly $1.4T coming from hyperscaler cash flow, that leaves a $1.5T funding gap spread across private credit, corporate debt, securitization, and other capital. Since Nvidia helps finance its own buyers, its commitment blurs the line between real demand and supplier-supported demand. Its guidance after Aug 26’s earnings will be one of the keys to assess how much longer the AI credit wheel can keep turning.

GOOG — GOOG finished roughly flat. Alphabet is one of the preferred bets for investors looking to participate in the AI mania without relying on outside credit or one model customer. It’s easy to forget about how large its vertical integration actually is. Search and YouTube generate substantial cash to fund cloud, models, TPUs, autonomous driving, and cybersecurity. This diversification provides the company with a lot of optionality, which can make them quite resilient during market downturns.

MSFT — MSFT lost about 2.5% this week. Similar to Google, Microsoft benefits from a self-funded balance sheet that can outlast leveraged ones. By selling productivity software, cloud infrastructure, cybersecurity, databases, operating systems, and developer tools through long-standing enterprise relationships they can generate recurring cash. Its risk lies in depreciation, customer concentration, and whether booked cloud obligations become high-margin revenue.

SPCX — SPCX is one of those tickers where the business and the stock send opposite signals. This week SpaceX completed two launches within roughly 38 minutes on August 17, taking its 2026 mission count above 100. The stock then tested its $150 debut level as about 319M shares became eligible for sale after the lockup expired on August 20. Launch cadence and Starlink scale keep improving, while new supply tests how much investors already paid for that growth. Starlink funds the launch system and AI build. The risk is that AI spending and unlocked shares absorb cash and demand faster than operating growth can replace them.

META — META lost about 6.8%, the worst weekly move among the major self-funded platforms, without a comparable new operating shock. The user graph and customer data is the moat; the high capex is the headwind. Q2 revenue rose 28% to $60.8B, but costs rose 55%, capital spending reached $31.1B, and free cash flow fell to $784M. With management expecting $130B to $145B of 2026 capex, wishful thinking invites the idea that Meta can fund that bill from advertising, which supposedly would make it safer than a leveraged cloud provider.

PLTR — Palantir is AI exposure without owning power plants, chips, or data centers. It also offers extra short squeeze optionality. This week PLTR gained about 3.4% while semiconductor and capacity names sold off.

AVGO — AVGO fell about 6.2% with the chip selloff. Broadcom sells custom AI chips and networking to hyperscalers seeking alternatives to Nvidia. It can earn a position in a portfolio as a cash-generating supplier, beyond generic semiconductor proxy. Q2 revenue reached about $22.2B, AI semiconductor revenue $10.8B, and free cash flow exceeded $10B. There are current discussions opening up the possibility that they are financing $60B of AI debt, with a possible junior tranche taking the structure toward $100B, after participating in a separate $35B compute package in June. Custom chips still offer strong cash flow and customer-specific demand. The new tradeoff is that growth depends more heavily on special-purpose debt and a small number of model companies.

AMD — AMD fell about 8%, a larger loss than Nvidia and Broadcom. NVDA is the AI platform (general-purpose GPUs + full systems); AVGO is the custom-silicon and networking supplier (custom AI ASIC + networking); and AMD is the merchant-compute challenger (GPUs + CPUs + full systems), with a buyer proposition to get competitive without depending entirely on NVDA. Overall, the market treats second-source compute as higher beta when capital becomes more expensive.

MU — MU finished close to flat after a violent round trip. Memory shares rallied on Monday as investors paid for HBM and storage scarcity, then Micron fell about 7% on Tuesday during the chip unwind. Nothing in the underlying memory cycle changed that quickly. Often presented as a tactical cycle long, it’s worth keeping an eye on price action, especially measuring if, or how much, it shrugs off good or bad news. Micron is the go-to for memory exposure, selling into an HBM (High Bandwidth Memory) shortage and capturing a premium on that scarcity. It earns commodity-like margins while capturing a memory premium on AI demand that requires more specialization than conventional DRAM and NAND offerings. A broader DRAM/NAND/NOR mix, with a growing HBM share flowing into AI data centers, is what should smooth out the old boom-bust earnings swings.

SKHY — On August 19, SK Hynix approved the purchase and cancellation of shares worth about 40T won, roughly $28.6B, and said more than half of free cash flow from 2025 through 2027 would support shareholder returns. After the KRW 40T repurchase was announced, Seoul gained 12.7% on Thursday and another 2.3% on Friday. At Friday’s ECB exchange rates, that Korean slice was worth about $124.98, leaving SKHY at a 30.8% premium, only modestly below Tuesday’s 32.0%. The operating case remains strong, with second-quarter gross and operating margins of 83.2% and 76.3%, but the capital claims are stacking: KRW40T of repurchases plus KRW54.32T of fab approvals equal 128.5% of first-half free cash flow. November 19, when the buyback ends, can be the time stop.

INTC — INTC lost about 12% on the week. Intel raised about $20B through a common-stock offering priced at $95, selling 210.5M shares. This move de-risks the balance-sheet in the short-term, giving more room for Intel to outcompete. The cost is dilution for existing holders, which now are forced to trust Intel’s ability to attract new customers and capital.

ORCL — Oracle raised about $43B of debt and $5B of equity during fiscal 2026 and expects another large debt-and-equity program in fiscal 2027, including a planned $20B at-the-market share facility. Customer prepayments lower the initial burden, but creditors and shareholders still fund capacity years before all of it produces revenue. Oracle uses cash from its database business to fund an AI cloud whose leases start before much of the contracted revenue arrives. That naturally involves a timing risk. This week ORCL closed at $146.47, down 2.7% for the week despite a 3.1% Friday rebound. Oracle has $260B of leases that have not started and $13.3B of purchase obligations, or $273.3B against $42.5B of equity. The contracted backlog of $638B covers those commitments 2.33 times, but only 12% should convert within twelve months while the leases begin between fiscal 2027 and fiscal 2029.

SNDK — SNDK closed at $1,596.08, down 2.7% for the week, after gaining 8.9% on Monday and losing 9.0% on Tuesday. Monday extended the post-Investor Day bid: management targets mid-to-high-teens revenue growth and roughly 80% non-GAAP gross margins from fiscal 2028 through 2030, while U.S. pressure against Chinese memory sourcing helped the wider complex. Tuesday then repriced the same earnings as bond yields rose and investors questioned the financing behind AI capex. At Friday’s $250.6B market value, SanDisk trades at 21.9 times fiscal-2026 net income and 23.6 times after removing the $807M Nanya mark. The contracts are $59.8B of remaining obligations plus $31.3B signed after year-end, but $2.742B of reported cash is customer advances and refundable deposits. The second-order risk appeared on Friday: YMTC’s parent filed to raise $4.9B for production upgrades and research. Peak margins finance the supply that later attacks them. In other words, a margin-duration test.

CRWV — Given the risks, this may be a position not worth chasing once it’s already slipped away. Q2 revenue reached $2.58B and backlog landed at $104.2B, with more than $25B of additional commitments signed early in Q3. Roughly half the backlog is active; the other half still needs chips, power, buildings, and customer acceptance. Management raised 2026 capex guidance to $35B–$39B after spending $9.4B during the quarter, showing how far back equity holders sit behind interest, leases, depreciation, and construction risk.

PANW — Palo Alto Networks aggregates security products inside a shared platform. It sells network, cloud, identity, security-operations, observability, and AI-security products. The reason why this bet is worth flagging is because AI agents expand the number and speed of attacks, which raises the value of automated detection and repair. PANW closed at $357.87, down 6.9% for the week despite a 2.4% Friday rebound and two AI-security releases. September 1 is the next dated catalyst, for which bulls will want to see positive operating income, organic growth in next-generation security annual recurring revenue, and evidence that acquired revenue is not masking slower internal growth. The last visible quarter had $3.00B of revenue and a $183M operating loss.

UBER — Many fundamental investors are starting to like UBER at current levels. Uber owns rider demand as well as the operating layer, while autonomous-vehicle suppliers compete for access to its network. Q2 Gross Bookings gained 24% to $58.0B, adjusted EBITDA increased 33% to $2.8B, and trailing free cash flow exceeded $10B. This gives Uber plenty of room to fund autonomy without choosing one driving stack. Uber Autonomous Solutions packages dispatch, mapping, regulation, fleet financing, insurance, remote assistance, and field support around third-party systems. The Zipline partnership applies the same model to Eats, with first deployments later in 2026 and a target of one million drone deliveries per day by end-2029. Merchants and autonomy suppliers pay to reach Uber’s demand. The risk is take-rate compression arriving before autonomous volume lowers unit costs.

BABA — BABA fell after Q2 exposed the cost of Chinese AI growth. Alibaba is also one of those players that can fund cloud with commerce. The business combines Chinese e-commerce, logistics, international commerce, cloud, proprietary chips, foundation models, and consumer AI. Commerce supplies users and cash; Cloud and Qwen supply the higher-growth thesis. Revenue rose 8.6% to RMB268.95B, while cloud-and-compute revenue grew 45% to RMB48.44B. However, since capital expenditure climbed 75% to RMB67.68B, adjusted net profit fell 38% to RMB20.72B; the U.S. shares fell about 4% before the open the next day. These results validate Chinese AI while exposing its cost. Alibaba has a stronger funding base than a leveraged cloud provider, yet shareholders still pay through lower profit and cash conversion.

FN — FN fell about 19% on the week; the stock had priced a faster next leg than management guided. Fabrinet is the factory behind many of the optical parts that connect AI data centers. The company manufactures optical and electronic products for the enterprises connecting AI clusters (communications design, industrial, automotive, and data-center products). That allows FN to capture manufacturing revenue without funding expensive buildouts or taking direct GPU utilization risk. This is becoming an increasingly common bet for owning the downstream effects of AI. Investors like it because this allows them to go long on the AI thesis without having to guess which chip designer will win. The drawback, however, is that Fabrinet has less pricing power than a company that owns the design, while carrying the cost of factories, workers, inventory and customer ramps. The company already delivered 39% revenue growth in Q3 and guided Q4 revenue toward roughly 40% annual growth. It then beat Q4 sales and earnings expectations, yet the stock fell about 19% after next-quarter profit guidance came in below what investors expected. That reaction can be a trade idea itself: the market already priced in even faster growth, AI demand is still strong, and now the bet becomes a function of cash conversion.

KEYS — KEYS sold off after Q3 revenue rose. Management also guided the next quarter above prior expectations. Keysight sells the equipment engineers use to test whether chips, networks, satellites, weapons and electronic products actually work before customers deploy them. Indeed, KEYS benefits from herd behavior and price reflexivity given the current AI and defense arms-race dynamics. This is the picks-and-shovels bet on the intersection of AI and defense spending. They make the testing tools engineers can’t build advanced electronics without. As the world spends more on AI, faster chips, and more wireless, Keysight sells more gear and software to nearly everyone racing to seize the opportunity. This is how they benefit from FOMO-driven capex where no one can afford to fall behind. Q3 exposed revenue rising 37% to $1.85B and adjusted earnings 78% to $3.07 a share. Management then guided the next quarter to about $1.94B of revenue and $3.37 of adjusted earnings, both above analysts’ expectations.

BHP — Copper has replaced iron ore as BHP’s earnings engine. Underlying attributable profit gained 30% to $13.2B; copper and its gold and uranium byproducts generated $18.19B of operating earnings against $14.53B for iron ore. Net debt fell to $8.69B, and the annual dividend reached $1.72 per share, supporting BHP as a liquid expression of the power, grid, and data-center buildout. The quality check, though, is that copper output remained near 2Mt for a second year while realized prices rose about 35%, and Jansen—BHP’s massive potash mine project—took a $2.3B write-down after Stage 2’s estimated cost increased from $4.9B to $6.9B.

WDC — WDC joined the memory round trip. The stock rallied about 6% on Monday, then reversed with the sector on Tuesday. Storage offers AI-data exposure without DRAM or HBM pricing. Western Digital now focuses on hard drives after separating its flash business; the thesis depends on cloud procurement, capacity discipline, and cash returns.

VST — VST is the preferred choice for many looking for exposure to power. Vistra owns merchant and contracted power generation plus a retail electricity business. VST closed at $136.21, down 8.0% for the week, with losses in four of five sessions. No new company disclosure caused the move; we are looking at a power-factor unwind: Talen lost 13.3%, NRG lost 10.4% and XLU lost 3.5%. The remaining risk is duration: Meta’s twenty-year purchases cover more than 2.6GW and start in late 2026, while the uncontracted fleet still prices gas, weather, and wholesale power. VST is, therefore, not a simple bargain, but more of a price-versus-fundamentals divergence.

NBIS — Nebius upsized its debt issued as convertible senior notes from $4.5B to $5.0B: $3.0B of 0.50% notes due 2030 and $2.0B of 4.50% notes due 2034, with buyer options taking gross proceeds to $5.75B. Nebius now has to pay back 110% to 125% of what it borrowed, and it also swapped some old debt for new shares, which shrank every existing shareholder’s slice by about 5.5%. Total convertible debt is now around $12B. Last quarter Nebius made $582M, lost $190M, and spent $5.7B building capacity. So the whole thing is a bet. Nebius is spending huge money now to build AI infrastructure, hoping customers show up and pay fast enough to cover the debt before it comes due. Lenders pay for the servers today; customers have to pay for them later. If the customer money arrives faster than the debt grows, Nebius wins. If it’s slower, it’s in trouble.

BTDR — BTDR announced a Malaysia AI-cloud buildout targeting 350MW by Q1 2028. A five-year contract worth about $400M covers roughly half of the first 9.5MW facility, with revenue expected to begin in Q1 2027. Two days earlier, Bitdeer filed an at-the-market program for up to $1B of equity. Bitdeer combines Bitcoin mining, proprietary mining chips, hardware manufacturing, power development, AI cloud, and data-center leasing. This degree of vertical integration allows for capturing margins at several layers; although that also creates several ways in which capital can be misallocated.

HD — HD remains a defensive way to own household spending while mortgage lock-in keeps owners repairing homes they cannot afford to replace. Home Depot sells repair and remodeling products into the household balance sheet. The company filed $47.9B of Q2 sales, up 5.7%, with comparable sales up 1.7% and adjusted EPS up 5.1% to $4.92. The average ticket rose 2.8% to $92.50, and smaller projects carried demand. Cash quality remained high, with $11.4B of first-half operating cash flow against $1.7B of capex.

LOW — Lowe’s carries the same housing exposure risk as Home Depot, but provides more explicit growth from Pro customers, online sales, and home services. It’s also the more rate-sensitive version of HD. Both sell into the same housing market, but Home Depot gets more business from professional contractors and repair work that often cannot be postponed, while Lowe’s depends more on homeowners choosing to renovate.

This quarter Lowe’s total sales increased 8.3%, helped by acquisitions, but sales at stores it already owned were almost flat at +0.2%, meaning the core business barely grew. Lowe’s also kept less profit from each dollar of sales as costs rose, then cut its full-year sales outlook, expecting ~$92B in full-year sales, flat comparable sales, and adjusted EPS about $12.25. The catch-up thesis now requires trust in Lowe managing to close some of the gap with Home Depot by winning more contractors and online customers while a housing recovery eventually brings homeowners back to larger projects.

TOL — If Home Depot tells us what existing homeowners are doing, Toll Brothers tells us whether wealthy households are still willing to buy a new home. Toll Brothers builds higher-priced homes for wealthier buyers and active adults. Its customers often have more equity and cash than entry-level buyers, which reduces mortgage sensitivity. The biggest factor that explains why investors like to own it is because it offers exposure to scarce new-home supply, affluent demand, and land value. Last quarter results showed home deliveries falling 10% and profit dropping, although the average selling price approached $1M and the value of newly signed contracts rose to $2.52B. Buyers are still there; fewer homes are reaching the closing table. TOL therefore offers exposure to affluent housing rather than a broad housing recovery.

KLAR — KLAR fell sharply after Q2 showed better unit economics and weaker end demand at the same time. Klarna operates a payments network and consumer lender. Merchants pay for conversion and payment services; consumers use deferred-payment and credit products; Klarna funds the receivables and bears part of the loss risk. Investors own KLAR because embedded checkout distribution can grow faster than conventional cards, while data and underwriting may improve loss selection. That creates more growth than a normal card network can offer, but Klarna also carries credit and funding risk that Visa or Mastercard largely avoid. Q2’s release exposed both sides of the coin: the amount consumers spent through Klarna rose 18% to $36.6B, revenue grew 27% to $1.04B, and the money left after processing costs, bad loans, and funding costs rose 42% to $446M. Yet Klarna cut its full-year spending and revenue outlook because Germany weakened, and the shares fell sharply. That’s why KLAR is such a useful consumer-credit signal: the business is getting better at earning money from each transaction while the customer is becoming less dependable. Upside from here requires U.S. growth to stay strong and Europe to stabilize before missed payments rise.

TGT — Target provides a more detailed read on discretionary spending than Walmart. The reason is because Walmart can still grow in spite of cautious households since its groceries and basic goods keep them coming back. Target, on the contrary, needs shoppers to purchase home goods, clothes, beauty products and other items they can postpone.

The company posted Q2 results, with sales at existing stores and online channels advancing 3.8%, customer visits climbing 3.6% and digital sales rising 8.7%. The profit figure, however, was inflated because of a roughly $1B of tariff refund. But still, more people are choosing Target again after the company cut prices on thousands of products. Now investors will demand shoppers keep showing up without the company having to lower prices. That’s what would turn a traffic recovery into an earnings recovery.

WMT — Walmart is typically the defensive compounder that can gain share when households trade down. Still, it fell sharply ~9% (largest daily drop since 2022) after reporting earnings. Indeed, the market expected defensive perfection; the company reported its slowest pace in six years. Walmart sells groceries and household essentials through its stores, then adds higher-margin revenue through advertising, memberships and e-commerce. Quarterly revenue rose 5.9% to roughly $188B and adjusted EPS beat at $0.81, but U.S. same-store sales increased only 2.6% against 3.8% expected; traffic growth slowed to 1.5%, and shares fell as the stock had priced near-perfect execution. Management has cut prices on 11,000 items while using $2.9B of tariff refunds to absorb cost pressure; another traffic miss in Q3 would show that the multiple outran the consumer. In short, stretched consumers, higher gas prices, and the weakest US sales growth in six years all point to a strained spending environment.

MRNA — MRNA surged as much as 160% on August 19. It was the winner of the week by price returns. Moderna’s Phase 3 melanoma trial finally worked. 1,137 patients who’d had stage IIB–IV melanoma surgically removed were tracked, and at a planned early check the treatment hit both main goals: patients went longer without the cancer returning and longer without it spreading. This is the first Phase 3 win for a therapy custom-built to each patient’s tumor, and the first for an mRNA cancer treatment, so Moderna’s stock jumped about 177% on August 19, with short sellers scrambling to cover their bets pushing it even higher. The earlier mid-stage data was strong (49% lower risk of the cancer returning or death, 59% lower risk of it spreading or death), but whether that huge stock pop becomes a lasting reset depends on details the Phase 3 numbers haven’t fully shown yet: how big the benefit really is, whether safety holds up, how fast Moderna can manufacture a personalized shot for each patient, and when it can file for approval.

RDDT — Reddit entered the S&P 500 on Monday, creating a known forced buyer effect: estimated index funds would require about 16.7M shares, close to three times Reddit’s normal daily volume. The price surge, since it was independent of fundamentals, now demands closer scrutiny. Reddit isn’t competing against Google or Meta, yet they can still play a role in the age of AI by selling targeted attention and access to “authentic” human conversations. While labs like Anthropic push watermarking and AI-detection to discern what’s human vs what’s AI, Reddit can certify decades of people arguing and package that into an appreciating asset. Advertising is the core revenue engine (high intent, community-targeted ads); data licensing allows for multi-year deals selling training access to its archive; and AI-powered search and answers on-platform can keep users inside. Unlike GOOG or META, Reddit isn’t spending the capex or shopping the models; it’s the data landlord renting the raw material to everyone who is. The risk is at the other side of the same coin, since Reddit relies on Google search referrals for traffic and its AI can answer bypassing the click through.

HOOD — HOOD gained 13.7% on Friday as Bitcoin rose 6.4% and the White House revived crypto-market-structure hopes. Robinhood is the entry door for retail. It owns the consumer interface to options, crypto, cash, credit, prediction markets, etc. With a moat that capitalizes on low-friction production distribution, the end-goal is to build the financial super-app for earning across trading, lending, payments, subscriptions, etc.

COIN — Coinbase regained operating leverage as Bitcoin posted its best week in more than two years. The shares rose 8.2% on Friday as trading activity, crypto prices, and regulatory expectations moved together. One of the keys going forward is that Coinbase is broader than a spot exchange. It owns custody, derivatives, stablecoin economics, and institutional distribution, with bets like Base and prediction markets offering more optionality. So far, the stock still behaves like a volume option when crypto wakes up. When sentiment cools down, reflexivity can cut the other way as spot demand and derivatives trend lower.

CRCL — CRCL rose with crypto, but further upside has a bottleneck to overcome: its USDC reserves should grow faster than falling reserve yields to support its valuation; otherwise distribution costs can dilute each dollar of float. Circle’s Q2 results delivered $73.3B of USDC in circulation, up 19% year over year, and $14.8T of onchain volume, up 151%, yet revenue and reserve income rose only 7% to $701M as the reserve return fell 66bp as distribution, transaction, and other costs reached $412M. The bull case is that CPN, already at a $14.7B annualized payment volume, and Arc’s September 16 mainnet let Circle own more of the settlement rail. Today, on the contrary, it’s mostly used as a trading asset, instead of payments.

BULL — Q2 was Webull’s best quarter: revenue grew 51% to $198.8M, trading-related revenue increased 66% to $147.7M and adjusted operating profit reached $62.6M, a 31.5% margin. Customer assets rose 79% to $28.5B and funded accounts reached 5.13M. Overall, Webull is less a Robinhood clone than a leveraged bet on the active retail trader. Funded accounts barely changed during the quarter while options volume jumped 34% sequentially and 68% year over year. The inflection came on June 4, when new FINRA rules removed the old $25,000 Pattern Day Trader barrier. Under the old framework, a margin customer with less than $25,000 could generally make only three day trades in five business days before triggering PDT restrictions. FINRA’s new intraday-margin regime removed that fixed trade-count barrier and the $25,000 PDT minimum. This benefited Webull significantly, since its filings describe customers as 37 years old on average, with 68% reporting prior investing experience; the product itself is built around active, self-directed trading rather than passive wealth management. In June alone, equity DARTs rose 57% from May and options DARTs 37%, while options volume reached 89.4M contracts, up 112% year over year. This was not only Webull: U.S. options volume rose 45% in June and Robinhood also posted record equity and options activity. BULL, however, can now be seen as a more direct retail-volume trade than HOOD, which is diversifying into banking, cards, and prediction markets. Importantly, from May to June, Webull funded accounts did not increase at all: they stayed around 5.1M. Yet equity DARTs jumped 57%, options DARTs 37%, equity notional volume 21%, and options volume 43%. June options volume was 112% above June 2025.

GLXY — Galaxy carries two capital cycles in one name: crypto and AI. Its crypto arm benefits from trading, advisory, staking, and higher crypto prices, while Helios and its other Texas sites require years of construction and financing. The happy path is perfect timing, such that rising crypto prices buy time for the AI buildout. This crypto rally helps current earnings; data centers determine the longer rerating. The biggest risk, though, is hidden correlation, since a single credit shock can tumble down the domino pieces, hurting both crypto prices and the funding value of the data-center pipeline.

PURR — Hyperliquid Strategies, HYPE’s treasury vehicle, is a straightforward directional bet on HYPE: issue or deploy capital, acquire HYPE, stake it, and try to increase HYPE per diluted share. Thus, it becomes increasingly appealing when the wrapper trades at a premium. That makes it possible to raise capital above NAV and turn the premium into accretive token purchases. And that’s exactly what happened this week as mNAV jumped back to 1 from a 0.82x low. The vehicle still uses a 0.9x-1.1x range to guide capital allocation; sustained momentum can boost both HYPE and revive the public-equity flywheel.

MSTR — MSTR gained 6% on Friday while Bitcoin rose 6.4%, then lagged BTC by a wide margin over the full week, which is a weak result for a leveraged treasury wrapper during a 22% underlying rally.

Conclusion
The equity uptrend holds but carries new risk. Higher long yields reduce the value of distant AI returns. The risk is that passive demand lifts indexes even as underlying breadth weakens. Watch for a Nasdaq 100 close below 29,000 or a recovery above 30,000 after Nvidia reports. Also watch small caps for signs that lower policy rates translate into easier credit.


