🎥 Watch the latest 7-minute briefing · Subscribe on YouTube

The memory supercycle takes its first real guidance-driven hit while Atlassian's 30% blowout hands the "AI kills SaaS" bears their cleanest counterexample yet — and Alibaba quietly starts charging for "open" weights.

🔥 Today's Top Stories

🔥 Memory: SanDisk and Western Digital Beat — and the Whole AI Memory Trade Breaks Anyway

Both companies cleared consensus on the quarter, and both guided above Street revenue at the top end — and the sector still got destroyed. The problem is the bar: SanDisk came into the print up 469% year-to-date and Western Digital up 202%, a setup where a midpoint-below-consensus guide is a sell signal. This is the first time the memory shortage narrative has been broken by the companies' own numbers rather than by a macro shock, and it landed on a sector that has been the single most crowded AI long of 2026.

• SanDisk guided FQ1 revenue to $10.3B–$10.8B (midpoint $10.55B) vs. LSEG consensus $10.47B; adjusted EPS $44–$46. Western Digital guided $4.1B ±$100M and $4.00 ±$0.15 EPS vs. $4.04B / $3.81 consensus — both technically beats, both sold.
• Aug 6 US session: Western Digital −13% to −16%, SanDisk −6% to −11%, Micron −6%, Seagate ~−4%.
• Asia followed hard: the Kospi fell 4.6% on Aug 6 with SK Hynix down ≈10% intraday. Today (Aug 7) the Kospi closed −0.60% at 6,258.71, SK Hynix −4.88% at ₩1,422,000, Samsung Electronics +0.22% at ₩231,000 — the damage narrowing to the pure memory names rather than broadening.
• The read-through: data-center memory demand was never questioned. What broke was the assumption that consumer/PC weakness could be ignored indefinitely while contract prices only went up.

  📰 Source: Stocks making the biggest moves after hours: Sandisk, Western Digital, E.l.f. Beauty, AppLovin & more — CNBC (Tier 1) 📰 Source: Western Digital Sinks 16%, SanDisk Falls 11%, Micron Drops 6% as Memory Selloff Hits Storage Stocks — 24/7 Wall St. (Tier 2) 📰 Source: Kospi falls 4.6% as Samsung Electronics, SK hynix lead tech selloff on AI spending fears — Korea JoongAng Daily (Tier 2)

🔥 Atlassian: Cloud Reaccelerates to 31% and the Stock Rips 30%+ — the "AI Eats SaaS" Thesis Gets Its Cleanest Rebuttal

Atlassian has spent a year as the market's favorite short expression of "AI will kill seat-based software." Then it printed a re-acceleration: cloud revenue growth went up to 31% from a business that had been decelerating all fiscal 2026 — the company was guiding to 23% cloud growth as recently as February. Management attributed the reversal to AI adoption driving seat expansion and higher net revenue retention, which is the exact opposite of the bear case. For investors, this is the first large-cap datapoint arguing that AI is a seat tailwind for collaboration software rather than a substitution risk.

• Q4 FY26 revenue $1,766M, +28% y/y, vs. ≈$1,660M consensus. EPS $1.87 vs. $1.50 expected.
• Cloud revenue $1,213M, +31% y/y — the acceleration is the whole story. Subscription ARR $6,606M, +23% y/y.
• GAAP profitability returned with a 12% Q4 operating margin. Full-year revenue $6.57B.
• Shares jumped more than 30% after hours; CEO and co-founder Mike Cannon-Brookes simultaneously disclosed a plan to buy up to $250M of Class A stock on the open market.

  📰 Source: Atlassian Shares Surge as Revenue Jump Douses AI Fears — Bloomberg (Tier 1) 📰 Source: Atlassian stock skyrockets more than 30% as cloud growth roars back — SiliconANGLE (Tier 2) 📰 Source: Atlassian Announces Fourth Quarter and Fiscal Year 2026 Results — Atlassian IR (Tier 2)

🔥 Alibaba: Plans to Charge Big Commercial Users of Its Next Qwen Model — "Open Weights" Starts Collecting Rent

Alibaba intends to attach revenue-sharing terms to its next open-weight Qwen release, requiring large companies that resell the model as a service to sign a commercial agreement and hand over a cut of revenue. Researchers and small developers keep free access to the weights. This is the moment the Chinese open-weights strategy — which has functioned as a price weapon against OpenAI and Anthropic for two years — starts converting distribution into a P&L line. If it holds, the "free Chinese frontier model" input cost that half the inference-startup universe is built on has a price now.

• The plan could be introduced as early as next week; the revenue-share percentage is reportedly not finalized and negotiations are ongoing.
• It mirrors Moonshot AI's Kimi K3 license, which targets Model-as-a-Service providers above $20M annual revenue and reportedly charges up to a 30% revenue share.
• Alibaba already monetized Qwen through its own cloud; this extends monetization to third-party deployments outside Alibaba Cloud — the part that was previously free.
• Investor angle: bullish for Alibaba's cloud monetization narrative, bearish for the margin structure of Western inference resellers who standardized on free Qwen weights.

  📰 Source: Exclusive: Alibaba plans to charge big users of its next open-source AI model, sources say — Reuters (Tier 1) 📰 Source: Alibaba Reportedly Plans Revenue-Sharing Terms for Next Qwen Model — TechNode (Tier 2)

🔥 Meta: New Mexico Judge Adds $567M and a Five-Year Consent Decree — Product Design Is Now Court-Supervised

Judge Bryan Biedscheid in Santa Fe found Meta liable for creating a public nuisance in New Mexico and ordered $567M into an abatement fund — on top of the $375M imposed in March for the same case, bringing the total to $942M. The money is the smaller half of the story. The order also dictates how Facebook and Instagram must function for minors in the state for five years, including AI chatbot safeguards. That is a state court setting product parameters, and 40-odd other AGs are watching the template.

• $420M of the $567M is earmarked for youth treatment services; the remainder funds awareness, prevention and screening over five years.
• Required changes include monthly usage limits for teens, notification restrictions, tighter controls on adult contact with minors, safeguards on AI chatbots, and enhanced review of child sexual abuse reports.
• Brought by New Mexico AG Raúl Torrez, who alleged Meta designed its products to addict young users and failed to protect them from sexual exploitation.
• Investor angle: immaterial to Meta's cash position, material to the compliance-cost trajectory — and the AI chatbot safeguard provision is the first time a court has written companion-AI rules into a social media remedy.

  📰 Source: Meta ordered to pay into $567 million fund after child harms case in New Mexico — CNBC (Tier 1) 📰 Source: Judge orders Meta to pay $567 million in New Mexico child safety case — The Washington Post (Tier 1)

🔥 OpenAI: The Jony Ive Device Is a $300–$400 Doughnut-Shaped Speaker With a Camera and No Screen

Bloomberg and The Information both put a price on OpenAI's first hardware product: $300 to $400 for a hockey-puck-sized, screen-free, battery-powered speaker with moving parts intended to convey "personality," plus microphones, lights and a camera system for ambient awareness. That price sits two to four times above Amazon's and Google's smart speakers, which tells you OpenAI is not competing for the smart-home slot — it is trying to establish a premium ambient-computing category before Apple or Amazon ship a credible LLM-native device.

• Price band $300–$400; ring/doughnut form factor roughly hockey-puck sized; no display, voice-first via ChatGPT.
• Battery-powered and portable rather than tethered to one room; camera and additional sensors give the assistant physical-environment context.
• Product of OpenAI's acquisition of io, Jony Ive's hardware startup. The screen-free device family is targeted at 2027.
• Investor angle: a hardware BOM and a retail channel are a materially different cost structure than an API business — watch for this to show up as a gross-margin drag well before it shows up as revenue.

  📰 Source: What Is OpenAI's Device? A Doughnut-Shaped Speaker That Costs Over $300 — Bloomberg (Tier 1) 📰 Source: OpenAI's new AI smart speaker will reportedly sell for between $300 and $400 — TechCrunch (Tier 1) 📰 Source: OpenAI's Smart Speaker Will Cost Over $300 — The Information (Tier 1)

👤 Key People Updates

📌 Mike Cannon-Brookes (Atlassian CEO & Co-Founder): Commits Up to $250M of His Own Money the Same Day the Stock Gaps 30%

Cannon-Brookes filed an 8-K disclosing a trading plan to make open-market purchases of up to $250M in Atlassian Class A stock. The timing is the signal: buying into a 30% up-gap is the opposite of the usual founder-sells-into-strength pattern, and it is an explicit bet that the AI-disruption discount on the stock is wrong rather than merely early.

• Purchases may begin no earlier than the close of Aug 7 and remain subject to a cooling-off period. $250M is a maximum planned value, not a completed purchase — treat it as intent, not execution.
• Disclosed alongside Q4 FY26 results showing $6.57B full-year revenue and a return to GAAP operating profitability.

  📰 Source: Atlassian (NASDAQ: TEAM) posts $6.57B revenue, CEO plans $250M stock buys — Form 8-K — SEC filing via StockTitan (Tier 2)

📌 Raúl Torrez (New Mexico Attorney General): Converts a State Consumer Protection Statute Into a Product-Design Injunction Against Meta

Torrez's office won on a public nuisance theory — not a federal privacy or Section 230 fight — and got a five-year decree governing how Meta's products behave for minors in New Mexico. For every other state AG with a pending social media or companion-AI case, this is now a proven playbook, and it routes around the federal legislative deadlock entirely.

• Total exposure from the case now $942M ($375M in March plus $567M this week).
• The decree explicitly covers AI chatbot safeguards — extending a social media remedy into generative AI product surfaces.

  📰 Source: Social media giant Meta ordered to pay $567 million in New Mexico child safety ruling — France 24 (Tier 2)

🏢 Big Tech Updates

📌 Agent Plugins 1.0: OpenAI, Microsoft, Amazon, Cursor and Vercel Agree on One Package Format — With Google as a Core Maintainer

Five direct competitors shipped a vendor-neutral spec that bundles MCP servers and Agent Skills into a single portable plugin directory, with Google joining as a core maintainer. It is the second major agent-layer standard to achieve cross-vendor buy-in after MCP itself, and it does the thing that actually matters commercially: it makes agent extensions portable, which weakens every vendor's ability to lock developers in via a proprietary plugin store. Sourcing note: this is currently covered mainly by the participating companies' own engineering blogs and the spec repo — no Tier 1 outlet has independently reported it, so it is capped at 📌 despite the impact.

• Spec is minimal: a directory with plugin.json, a skills/ folder, and an optional mcp.json. Version 1.0.0, still labeled a Working Draft — no plugin store and no security spec yet.
• Launch clients: ChatGPT, Codex, Cursor, GitHub Copilot, VS Code, and Amazon's Kiro. Google plans adoption in Agents CLI and Data Agent Kit.
• Vercel initiated the proposal; the Technical Steering Committee is Amazon, Cursor (Anysphere), Microsoft, OpenAI and Vercel.
• Investor angle: standardization compresses the moat of any single agent platform and shifts value toward whoever owns the model and the distribution surface — a structural positive for OpenAI/Microsoft, a negative for smaller agent-IDE startups betting on a captive extension ecosystem.

  📰 Source: Introducing Agent Plugins — Vercel Engineering Blog (Tier 2) 📰 Source: AWS Supports Agent Plugins: An Open Standard for Portable Agent Extensions — AWS Open Source Blog (Tier 2) 📰 Source: OpenAI and four rivals just agreed on one standard for AI agents — The Next Web (Tier 2)

📌 OpenAI: Unlimited Text Chats for Free Users as ChatGPT Crosses 1 Billion Weekly Users

OpenAI removed message caps on text chat for every tier, made GPT-5.6 Luna the default for Free and Go users, and added a "Think" button for harder questions. Crossing a billion weekly users while simultaneously removing the rate limit that protected inference costs is a deliberate land-grab — it only pencils if Luna's cost per token has fallen far enough to make free-tier marginal cost negligible, which is itself the most informative datapoint here.

• GPT-5.6 Luna replaces GPT-5.5 as the Free/Go default. Plus and Pro users get an updated GPT-5.6 Sol with a slider controlling reasoning effort.
• Separate limits remain for files, images, voice and image generation — text is the only surface going unlimited.
• Follows OpenAI's Jul 30 API price cut of up to 80%, with Luna at $0.20 per million input tokens.

  📰 Source: ChatGPT brings unlimited text chats to free users — TechCrunch (Tier 1)

📌 ChainDrop: A Self-Propagating Worm Backdoors 400+ npm Packages With Two Billion Monthly Installs — and Fights Back When You Revoke the Token

Microsoft Threat Intelligence and Elastic Security Labs documented a Shai-Hulud-family worm that trojanized the keyv monorepo and then used the maintainer's stolen npm credentials to auto-backdoor every package they could publish to. It crossed organizations every two to seven minutes and finished its publishing burst in about half an hour. The detail that should worry every security team: the payload installs itself as a LaunchAgent or systemd service and calls eval on a remote-supplied handler the moment the stolen GitHub token stops working — which is exactly what happens when you revoke it.

• 400+ packages, 2B+ monthly installs. keyv alone had 600M downloads last month; flat-cache ≈580M, cacheable-request 137M+.
• Harvests credentials against 300+ patterns explicitly targeting AI tooling, cloud providers, GitHub and npm tokens. Runs on Linux, macOS and Windows via preinstall hooks.
• Uses Ethereum smart contracts for command-and-control resolution, and embeds itself in AI coding-tool config files to enable reinfection.
• Investor angle: this is the concrete version of the supply-chain risk that has been driving agent-security M&A all year — and the AI-tooling credential targeting means the blast radius now includes model API keys, not just cloud accounts.

  📰 Source: ChainDrop supply chain compromise: Anatomy of a self-propagating worm — Microsoft Security Blog (Tier 2) 📰 Source: Shai-Hulud strikes again: CHAINDROP worm hits 400+ npm packages — Elastic Security Labs (Tier 2) 📰 Source: ChainDrop Worm Hits 400 npm Packages with Two Billion Monthly Installs — Infosecurity Magazine (Tier 2)

📊 Market & Investment Signals

💬 The Tape Keeps Splitting: Dow Sets Another Record at 54,349 While the Nasdaq Drops 0.8% on Memory

Same shape as yesterday, different driver. The Dow closed at a record 54,349.12 (+0.5%, +263 points) while the S&P 500 slipped 0.2% to 7,723.55 and the Nasdaq Composite fell 0.8% to 26,363.44. This time the damage was concentrated almost entirely in storage and memory rather than in the GPU complex — a narrower, more specific rotation than the broad AI de-rating of late July. Rising Treasury yields and firmer oil added pressure ahead of jobs data.

📰 Source: Stock market today: Dow slips from record, S&P 500 and Nasdaq drop as oil gains, Treasury yields rise — Yahoo Finance (Tier 2) 📰 Source: Stock Market Today (Aug. 6, 2026): S&P 500 declines as investors weigh jobs data, earnings ahead — TheStreet (Tier 2)

💬 Hut 8: The Bitcoin-Miner-to-AI-Landlord Trade Now Carries $26.6B of Contracted Revenue — and a $177M Quarterly Loss

Hut 8 reported 949 MW of contracted IT capacity representing $26.6B of expected aggregate base-term contract value and $1.75B+ in expected average annual NOI, all leased to or backstopped by investment-grade counterparties. Q2 revenue was $74.9M against a $177.1M net loss driven by $138.6M of unrealized digital asset marks. The gap between the contracted-value headline and the current P&L is the entire investment question for this cohort: the revenue is real but arrives in 2027, and the balance sheet still swings on bitcoin in the meantime.

• $7.5B of investment-grade project financing closed in the quarter. 1,330 MW under construction across River Bend (Q2 2027 delivery) and Beacon Point (Q3 2027).
• Beacon Point's 1 GW campus is now fully commercialized at $19.6B of campus-level contract value after a second 352 MW, 15-year lease to the same investment-grade tenant.

  📰 Source: Hut 8 Reports Second Quarter 2026 Results — Hut 8 / PR Newswire (Tier 2)

💬 Suno Starts Watermarking Its Songs — the First Major Generative Music Platform to Concede on Provenance

Under mounting litigation pressure from the music industry, Suno will add audio watermarking and fingerprinting, signed a deal with Musixmatch to use its Sentinel copyright-detection system, and tightened download policy to limit mass distribution to streaming platforms. Community guidelines now explicitly ban deceptive audio and unauthorized voice likenesses. Read it as the licensing-negotiation posture of a company that expects to settle.

📰 Source: Amid legal battles, Suno says it will start watermarking songs — TechCrunch (Tier 1)

🎥 Prefer to watch? This brief becomes a daily video on The Daily Alpha — AI × markets in ~7 minutes, every weekday. ▶ Watch the latest briefing 📺 Subscribe to the channel

Generated by Alan · The Daily Alpha. Editorial, not financial advice.