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

The AI trade split in two overnight: Microsoft added ≈$260B on cheaper capex and a 45% Azure guide while Meta shed ≈8% on a spending hike and $784M of free cash flow — and Samsung then told Seoul the memory shortage runs through 2028, snapping the Kospi back 5%.

🔥 Today's Top Stories

🔥 Microsoft & Meta: The Capex Verdict Lands — MSFT +8% on Lower Spending and 45% Azure, META −8% on Higher Spending and $784M Free Cash Flow

The two-earnings night the whole month was positioned around resolved into the sharpest intra-Big-Tech divergence of the AI era, and the split was not about growth — both beat on revenue. It was about whether the spending buys a visible return. Microsoft showed accelerating cloud demand on a reduced capex line; Meta showed decelerating profit on a raised one. For investors, this is the cleanest read yet on how the market will price AI capex from here: capex is no longer punished or rewarded in the abstract, only against the revenue it is visibly converting into.

• Microsoft: revenue $90.01B, +18% y/y (est. $87.62B); adjusted EPS $4.74 vs $4.24 consensus. Azure crossed $100B in annual revenue for the first time, up 41%.
• CFO Amy Hood guided Q1 FY27 Azure to 45% growth in constant currency against a ≈41.4% StreetAccount consensus — the single number that moved the stock. Contracted revenue backlog reached $678B; Copilot paid seats hit ≈30M.
• Microsoft's calendar-2026 capex guide came down to roughly $175B from ≈$190B in April — though part of that reduction is an accounting-classification change effective FY27, not a genuine spending cut. Q4 capex alone was $41B. Stock +8.9% after hours, ~$260B of market value added in one evening.
• Meta: revenue $60.8B, +28% y/y (est. $60.22B); ad revenue $59.3B. But EPS came in at $6.18 against $7.14 consensus — a hard miss.
• Meta raised the floor of its 2026 capex range to $135–145B (from $125–145B), spent $31.1B in Q2, and free cash flow collapsed to $784M. Q3 revenue guide of $61–64B straddles the $62.68B consensus. Shares −8% after hours, −8.5% in Thursday premarket.

  📰 Source: Microsoft (MSFT) Q4 earnings report 2026 — CNBC (Tier 1) 📰 Source: Meta's stock plunges on light revenue guidance, dwindling free cash flow — CNBC (Tier 1) 📰 Source: Microsoft Q4 2026 earnings: Azure tops $100 billion — Quartz (Tier 2)

🔥 Samsung: "The Shortage Deepens Through 2028" — One Earnings Call Reverses a Two-Day Circuit-Breaker Crash, Kospi +5%

After the worst two sessions in Kospi history — a 10.84% collapse Tuesday and a second consecutive circuit breaker Wednesday, both driven by the China commodity-DRAM scare — Samsung used its Q2 call to attack the bear case directly: supply cannot physically arrive fast enough to break the cycle. The index had been down 2.05% on the open and was gaining more than 5% by the time the call ended. This is the first genuinely new information in the Korea chip rout, and it reframes the CXMT/China-lithography panic as a 2029+ problem rather than a 2027 one.

• Samsung told investors the 2027 shortage will deepen versus 2026, and continue into 2028, arguing it takes more than three years to build fab capacity and ship wafers — so meaningful supply relief before 2028 is arithmetically unlikely.
• HBM4 revenue is guided to "more than triple" quarter-over-quarter in Q3, and to exceed 60% of total HBM revenue in H2 2026 — the highest-margin part of the mix inflecting exactly as commodity DRAM fears peak.
• Samsung Electronics closed up 7.91% at ₩225,000. SK Hynix, down as much as 8% intraday, reversed to gains of 1–2% once the call began. The Kospi rebounded from 5,547.41 to 5,976.82.
• Context for the base: Tuesday's close was 6,023.66 after a −10.84% session (fourth-largest daily drop on record); Wednesday closed 5,663.24, −5.98%.

  📰 Source: Samsung Sees Deeper Chip Shortage in 2027, KOSPI Rebounds 5% — Seoul Economic Daily (Tier 2) 📰 Source: AI memory, chip stocks rise after SK hynix, Samsung rebound — Seeking Alpha (Tier 2)

🔥 The Fed Holds 9–3 and the Bond Market Overrules It: 30-Year Yield Hits 5.23%, Highest Since 2007 — With AI Debt Issuance Named as a Cause

Kevin Warsh's first real test as Fed chair produced a divided hold — and then the long end sold off anyway. The 30-year jumped 11bp after the decision to its highest level since before the financial crisis, and markets are now pricing two rate hikes in 2026, not cuts. The AI-specific transmission matters here: the same hyperscaler capex being debated on earnings calls is being financed in the corporate bond market, and that issuance deluge is now explicitly cited as a driver of long-dated yields. Rising discount rates plus rising financing costs is the one macro configuration that hits long-duration AI infrastructure equity twice.

• FOMC voted 9–3 to hold at 3.50%–3.75%, the fifth consecutive hold. Dissenters were all regional presidents: Hammack (Cleveland), Kashkari (Minneapolis), Logan (Dallas).
• The statement described activity as expanding at a solid pace with strong productivity growth and capital investment, while flagging Middle East conflict as an uncertainty. Officials' year-end rate projections span 3.6%–4.1%.
• The 30-year Treasury reached ≈5.23–5.24%, highest since 2007, after +11bp post-decision. It has now traded above 5% for 30 straight days — the longest stretch since 2007 — on fiscal deterioration plus a flood of AI-infrastructure corporate issuance.
• Despite the yield move, futures recovered Thursday: Dow +225 (+0.4%), S&P 500 +0.6%, Nasdaq 100 +1.4% on the Microsoft read-through. June PCE lands today.

  📰 Source: Fed rate decision July 2026: Divided Fed holds interest rates steady — CNBC (Tier 1) 📰 Source: FOMC statement, July 29, 2026 — Federal Reserve (primary) 📰 Source: Stock market today: Dow, S&P 500, Nasdaq futures rebound ahead of inflation data — Yahoo Finance (Tier 2)

👤 Key People Updates

📌 Amy Hood (Microsoft CFO): Guides Azure to 45% While Cutting the Capex Number — the Two-Line Move That Added $260B

Hood delivered the exact combination the Street had decided was impossible this quarter: acceleration on the top line, deceleration on the spending line. Worth noting for anyone modelling it — the capex reduction is partly definitional, not purely operational, and Microsoft still spent $41B in a single quarter.

• Q1 FY27 Azure guide of 45% constant-currency growth vs ≈41.4% consensus; Q1 revenue guided to $89.85B–$90.95B (midpoint above the ≈$89.7B expected).
• Calendar-2026 capex guide lowered to ≈$175B from ≈$190B — with an FY27 accounting-classification change contributing to the optics.

  📰 Source: Microsoft (MSFT) Q4 earnings report 2026 — CNBC (Tier 1)

🏢 Big Tech Updates

📌 Anthropic: Claude Mythos Autonomously Breaks New Cryptanalytic Ground — Cuts a NIST Post-Quantum Candidate's Key Strength in Half for ≈$100K of Compute

Anthropic's unreleased Mythos Preview model produced two genuinely novel cryptanalytic results that human experts had missed for years — including inventing its own attack technique. Neither threatens deployed systems, and Anthropic is careful about that. The investor-relevant signal is not the cryptography: it is a priced, reproducible demonstration that frontier models can now generate original results in a field where verification is objective and expert labour is scarce. That is the clearest evidence to date for the "AI as research capex" thesis that every lab's valuation now rests on.

• On HAWK, a NIST post-quantum signature candidate, Mythos found a previously unexploited lattice symmetry (a nontrivial automorphism / τ-cocycle structure) that drops smallest-key recovery from 2^64 to 2^38 operations — effectively halving key strength, so key sizes must double. Disclosed to HAWK's authors and coordinated through NIST.
• On AES, it invented a technique it named the "Möbius Bridge", eliminating a 256-way guessing step in meet-in-the-middle attacks on a 7-round test version and running 200–800× faster than the record cryptographers set in 2013. It requires ≈2^105 chosen plaintexts and does not touch production 10-round AES-128.
• Cost and effort: ≈60 hours in a multi-agent scaffold, roughly 1 billion output tokens, ≈$100,000 in API spend — versus two researchers taking nearly a month to verify the result was correct.
• ⚠️ Sourcing note: widely covered by security trade press (CyberScoop, CSO Online, The Hacker News) but no Tier 1 general-business outlet has picked it up yet, which is why this is capped at 📌 rather than led as a top story.

  📰 Source: Anthropic's Claude Mythos finds weaknesses in encryption algorithms — CyberScoop (Tier 2) 📰 Source: Discovering cryptographic weaknesses with Claude — Anthropic research (primary) 📰 Source: Claude AI Just Cracked a Post-Quantum Test Scheme and Found a Faster 7-Round AES Attack — The Hacker News (Tier 2)

📌 Apple & Amazon: The Second Half of the Gauntlet Lands After Today's Close — AWS Reacceleration vs. Apple's Memory-Cost Squeeze

Both report after the bell, and after last night they report into a market that has just demonstrated exactly what it rewards. Amazon's setup is the Microsoft trade in mirror image — AWS momentum against a capex number likely to be raised again. Apple's setup is the inverse of the whole AI complex: it is the company that pays for the memory shortage rather than profiting from it, which is precisely why it has been bid as the non-data-center hedge.

• Amazon: consensus EPS $1.82 on revenue ≈$196.3–196.9B. AWS is modelled at ≈$40.5B, +31% y/y, after Q1's +28% — the division's fastest growth in 15 quarters. Tracked 2026 capex consensus has crept to ~$207B and is widely expected to be raised again.
• Apple: consensus EPS $1.89 on revenue $108.65B. The focus is gross margin against memory chip price hikes — the direct cost of the same shortage Samsung just extended to 2028.

  📰 Source: Apple, Microsoft And 3 Stocks To Watch Heading Into Thursday — Benzinga (Tier 2) 📰 Source: Amazon's AWS may deliver a July 30 earnings beat — TheStreet (Tier 2)

📌 OpenAI: Opens Free Frontier-Model Access to 100,000 Academic Researchers — a Distribution Land-Grab Dressed as Philanthropy

OpenAI will give academic researchers free access to its most advanced models, starting with 10,000 and scaling toward 100,000 by 2027. Read commercially, this is a standard-setting move against Anthropic's enterprise strength and Google's institutional reach: researchers who build their workflows on ChatGPT and Codex become the labs, papers, and hiring pipelines that define default tooling for a decade. The cost is compute Anthropic and Google will now be pressured to match.

• Program includes privacy protections, training, and research support; scales 10,000 → 100,000 by 2027.
• Shipped alongside ChatGPT updates extending Voice to Work and bringing Codex into the desktop app.

  📰 Source: OpenAI Newsroom — recent announcements — OpenAI (primary) 📰 Source: AI News Today July 29 2026 — BuildFast (Tier 2, aggregator)

📊 Market & Investment Signals

💬 Cyera: Buys Oasis Security for ≈$1B to Control What AI Agents Can Log Into — Agent-Security M&A Has Tripled This Year

Roughly $700M cash plus stock, weeks after Cyera raised $600M at a $12B valuation. Oasis secures non-human identities — service accounts, tokens, keys, and AI agents — which is the exact attack surface the OpenAI/Hugging Face agent breach exposed. Non-human identity management is becoming a discrete, fast-consolidating security category, and the buyers are late-stage private companies paying unicorn prices for it.

• Oasis was founded in 2022 and raised ≈$195M from Accel, Craft Ventures, and Cyberstarts. Deal expected to close later this year.

  📰 Source: Cyera agrees to acquire Oasis Security for $1B to safeguard proliferating AI agents — TechCrunch (Tier 1)

💬 xAI: Sues Minnesota's Attorney General on First Amendment Grounds Days Before the Nudification Ban Takes Effect

xAI filed in federal court to block HF 1606, arguing the law is an "overbroad, content-based ban on free speech and the tools of visual expression" and imposes strict liability on providers regardless of mitigations. Whatever the merits, this is the first serious constitutional test of a state AI-content statute, and the liability numbers are the reason it matters commercially: penalties up to $500,000 per violation plus a private right of action with treble and punitive damages would make state-by-state compliance the binding constraint on consumer image generation.

• Filed July 27 in the U.S. District Court for the District of Minnesota against AG Keith Ellison; the law was set to take effect August 1.

  📰 Source: Elon Musk's xAI sues Minnesota over law to ban 'nudify' apps — CNBC (Tier 1)

💬 Anthropic: The Most Valuable Startup in the World Is Also the Most Isolated — Open-Weights Fight Escalates Into the IPO Window

Amodei's denial that Anthropic ever backed an open-weight ban is being read less as clarification than as repositioning: commentary this week frames it as joining a broader tech-leader pattern of redirecting White House pressure toward Nvidia and China policy rather than absorbing it. With an October IPO ahead, the reputational cost of being the lab Silicon Valley organises against is now a valuation input, not just a culture story.

• Amodei rejects an open-weight ban but wants mandatory pre-release safety testing for open and closed models, plus targeted measures against distillation.
• Included here as a footnote only — the underlying Amodei post was covered in this report on July 29 and there is no new event today.

  📰 Source: Anthropic Has Just Turned Up the Heat on Nvidia — Bloomberg Opinion (Tier 1) 📰 Source: Anthropic is the world's most valuable startup — and its most isolated AI leader — Axios (Tier 2)

Dedup note: the pacing-mechanism letter (1,171 signatories), the MCP stateless rewrite, and the Hugging Face breach forensics all led earlier reports this week and produced no new developments today — deliberately omitted rather than restated.

🎥 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.