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

The AI-capex trade got its verdict overnight: Amazon proved cloud demand is accelerating and raised spending to $220B anyway, Apple proved a beat is not enough, and a violent tech rebound closed out July with the Nasdaq up 2.8%.

🔥 Today's Top Stories

🔥 Amazon: AWS Grows 37% — Fastest Since 2021 — and Jassy Raises 2026 Capex to $220B, Blaming Memory Prices

Amazon closed the Big Tech earnings gauntlet with the single most bullish datapoint of the season: AWS revenue of $42.23B grew 37% year over year against a 31% consensus, the unit's fastest expansion since 2021 and its fifth straight quarter of acceleration. Total company revenue crossed $200B in a quarter for the first time. But the number that matters for the whole semiconductor complex is the capex raise — from ≈$200B to ≈$220B — and the reason Jassy gave for it: the higher cost of memory. That is the first time a hyperscaler has explicitly attributed a $20B spending increase to DRAM inflation, which reframes the memory shortage from a supplier windfall into a buyer's cost line. For investors, this is the cleanest read yet that AI demand is real and that the unit economics of serving it are getting worse at the margin.

• AWS revenue $42.23B, +37% y/y (est. $40.54B / +31%); AWS operating income $16.6B vs $10.2B a year ago, 39.4% margin
• Total net sales $200.6B, +20% y/y; operating income +43% to $27.5B; EPS $5.75
• 2026 cash capex guide raised to ≈$220B from ≈$200B, attributed to memory costs
• AI services and Amazon's homegrown silicon (Trainium/Inferentia) each cleared a $25B annual revenue run rate
• Stock rose 9–10% after hours and traded up ≈13% premarket

  📰 Source: Amazon Q2 earnings report 2026 — AWS grows 37%, capex hiked to $220 billion — CNBC (Tier 1) 📰 Source: Amazon Q2 2026 earnings: AWS grows 37%, revenue tops $200B — Yahoo Finance (Tier 2)

🔥 Apple: Beats on Every Line — EPS $2.02, iPhone +22% — and Falls 7% Anyway on a Soft Guide and "Supply Constraints"

Apple delivered a record June quarter and the market sold it hard. EPS of $2.02 beat the $1.89 consensus (+29% y/y) on revenue of $109.4B, with iPhone up 22% and Mac up 29%. The stock still dropped ≈7% because management guided fiscal Q4 revenue growth to 9–11% against a ≈12.1% Street expectation and flagged supply constraints — the polite phrase for the same memory and component squeeze that just cost Amazon $20B. Services growth decelerated and China came in light. The investor takeaway is uncomfortable for the "Apple is the safe AI-adjacent trade" thesis that carried it to the world's-most-valuable-company title four days ago: not building data centers protects you from capex, not from the bill of materials.

• EPS $2.02 vs $1.89 est.; revenue $109.42B vs $108.86B est.; total sales +16% y/y
• iPhone revenue $54.3B (+22%); Mac $10.4B (+29%); Services $30.7B, a June-quarter record but decelerating
• FQ4 guide of 9–11% growth vs ≈12.1% consensus; "supply constraints" cited
• Shares fell 6.65% after hours to ≈$311.25 from a $333.43 close, and were down ≈7.2% premarket

  📰 Source: Apple (AAPL) Q3 2026 earnings report: Live updates — CNBC (Tier 1) 📰 Source: Earnings call transcript: Apple beats Q3 2026 estimates, shares fall after hours — Investing.com (Tier 2)

🔥 OpenAI: Cuts GPT-5.6 API Prices by Up to 80% — Luna Drops to $0.20 per Million Input Tokens

OpenAI slashed API pricing across the GPT-5.6 line, with the small Luna model falling 80% and mid-range Terra falling 20%, attributing the cut to inference efficiency gains it is passing through rather than to competitive pressure. The timing tells a different story: this lands one week after Kimi K3's 2.8T open weights went free and days after Chinese labs triggered a global chip selloff. Read alongside Amazon's memory-driven capex raise, it sets up the sharpest tension in the sector right now — the cost of serving a token is being cut faster than the cost of the hardware that serves it is rising. For anyone modeling AI software margins, the deflation is now measurable in quarters, not years.

• GPT-5.6 Luna: $0.20 per million input tokens / $1.20 output (−80%); Terra: $2 / $12 (−20%)
• Flagship Sol pricing unchanged, but gains a Fast Mode running up to 2.5× faster at double the price
• ChatGPT and Codex subscription prices unchanged — Terra and Luna simply consume fewer credits
• Auto-review in the ChatGPT app and Codex CLI upgraded to GPT-5.6 Luna

  📰 Source: OpenAI cuts prices for two of its GPT-5.6 AI models as companies grow sensitive to costs — CNBC (Tier 1) 📰 Source: OpenAI Cuts GPT-5.6 Pricing Up To 80%, As AI Costs Come Under Scrutiny — Forbes (Tier 2)

🔥 Google DeepMind: Gemini Robotics 2 Controls an Entire Humanoid — Legs, Torso and Fingers — Under One Policy

DeepMind shipped the first Gemini model that drives a full humanoid body from a single learned policy, extending last year's upper-body-only system down through the torso and legs. The demos are mundane by design — walk to a shelf, crouch, place a watering can, unscrew a light bulb at a 92% success rate — which is precisely the point: mundane multi-step physical chores are the commercially addressable market. Named deployment partners include Apptronik (Apollo 2), Franka and Agile Robots, with Boston Dynamics in the technical credits. Investors should treat this as Google converting its model lead into a second platform beachhead: the embodied-reasoning variant is already callable by any developer with a Gemini API key, which is how Android-style ecosystem lock-in starts.

• First DeepMind policy to jointly control legs, torso, arms and multi-finger hands; supports multi-robot collaboration on hundred-step tasks
• 92% success rate on light-bulb unscrewing, a benchmark requiring coordinated whole-body control
• Gemini Robotics ER 2 live via Gemini API and Google AI Studio; private preview on Gemini Enterprise Agent Platform
• VLA and On-Device 2 variants limited to 100+ trusted testers and early partners

  📰 Source: Gemini Robotics 2 Expands Google's AI Capabilities for Humanoid Robots — Bloomberg (Tier 1) 📰 Source: Google DeepMind debuts Gemini Robotics 2 model series for humanoid robots — SiliconANGLE (Tier 2)

🔥 Markets: Nasdaq Rips 2.8% to Close Out July — Earnings Rescue the Tape After a Week of Chip Carnage

The AI trade got its confidence back in a single session. The Nasdaq Composite closed +2.8% at 25,122.18 and the S&P 500 +1.66% at 7,437.63, capping a month that included a Kospi circuit breaker, a Chinese-memory panic, a 1,100-point Dow rout and a Fed that held with three dissents. The rebound was led by exactly the names that broke hardest: Micron and SanDisk each rose over 9%. JPMorgan's read is that the forced deleveraging is largely done — hedge funds and large investors have likely finished dumping tech, chip and memory exposure to cut leverage, which removes the mechanical seller that made last week's declines so violent.

• S&P 500 +1.66% to 7,437.63; Nasdaq Composite +2.8% to 25,122.18
• Micron and SanDisk both +9%-plus as chip stocks caught a bid after Thursday's selloff
• JPMorgan: leverage-driven selling in tech/chip/memory names is likely exhausted
• Prediction markets had priced a 94% chance of a higher S&P open Friday on the back of Amazon's print

  📰 Source: Stock Market Today: Nasdaq futures rise as Amazon earnings lift tech stocks (July 31, 2026) — TheStreet (Tier 2) 📰 Source: How major US stock indexes fared Thursday 7/30/2026 — Washington Post (Tier 1)

👤 Key People Updates

📌 Andy Jassy (Amazon CEO): "AWS Is Booming" — and Two Businesses Nobody Modeled Separately Just Cleared $25B Run Rates

Jassy's call did two things at once: he defended a $220B capex number by disclosing the revenue lines it is funding, and he quietly repriced Amazon's silicon story. Both the AI services business and the homegrown chips unit — Trainium and Inferentia — passed a $25B annualized run rate, numbers Amazon has never broken out this cleanly. That is the strongest public evidence to date that a hyperscaler's internal accelerator program is a real revenue business rather than a bargaining chip against Nvidia.

• AI services and custom silicon each above a $25B annual run rate
• Capex raised to ≈$220B for 2026, explicitly attributed to memory pricing rather than to more compute
• AWS acceleration is now five consecutive quarters — the pattern Microsoft's 45% Azure guide set up two days earlier

  📰 Source: Amazon.com, Inc. (AMZN) Q2 FY2026 earnings call transcript — Yahoo Finance (Tier 2)

🏢 Big Tech Updates

📌 NVIDIA: Trades Below $190 as the Rotation Punishes the Second Tier Harder — AMD and Marvell Each −22% on the Week

The chip selloff has been indiscriminate in direction but not in magnitude: NVDA fell about 10% on the week while AMD and Marvell each shed roughly 22%. That spread is the market re-underwriting Nvidia's full-stack position rather than abandoning it. Consensus target sits at $303 (≈59% upside from $190), with Baird's Tristan Gerra carrying a Street-high $500. The next real catalyst is distant — Q2 FY2027 results land August 26 — which leaves the stock hostage to hyperscaler capex headlines until then, and today's Amazon raise is a favorable one.

• NVDA ≈$190 vs $303 consensus target; Baird Street-high $500
• AMD −22% and MRVL −22% on the week vs NVDA −10%
• Q2 FY2027 earnings call scheduled for August 26 (quarter ended July 26, 2026)
• Announced a joint agentic-AI research lab with KAIST in Seoul, plus the previously reported SSI partnership

  📰 Source: Nvidia is Tanking Below $190: One Wall Street Pro Sees 165% Gains From Here — 24/7 Wall St. (Tier 2)

💬 Anthropic: Two Network Failures in 24 Hours Take Down Claude, the API and Claude Code — Government Tier Unaffected

Anthropic confirmed two separate incidents inside a 24-hour window that produced elevated errors and degraded availability across claude.ai, the API, Claude Code and Claude Cowork, caused by multiple network failures cutting serving capacity. Full recovery came late on July 30. Claude for Government stayed up throughout on isolated infrastructure — a distinction worth noting for enterprise buyers evaluating tiered reliability guarantees ahead of the company's October IPO.

• Two incidents in 24 hours; elevated errors and reduced availability across consumer and developer surfaces
• Requests failed while traffic was rerouted around the affected capacity
• Claude for Government unaffected — isolated infrastructure

  📰 Source: Claude Outage July 29-30 2026 — What Broke, Fixed — explainX (Tier 2, sourcing thin)

📊 Market & Investment Signals

💬 Core PCE Cools to 3.3% — but the Market Still Prices a 63% Chance of the Fed's First Hike in Three Years in September

The Fed's preferred gauge eased in June: headline PCE fell 0.1% on the month to 3.7% y/y from 4.1%, and core rose just 0.1% to 3.3%. Both remain far above the 2% target, which is why a cooling print did nothing to remove the hike from the table — CME FedWatch still assigns a 63% probability to a 25bp increase in September after Wednesday's 3.5–3.75% hold drew three dissents in favor of hiking now. Real consumer spending rose 0.4%, meaning the economy is not doing the Fed's work for it. For AI infrastructure names financing buildouts with debt, the September meeting is now the single biggest non-earnings risk on the calendar.

• Headline PCE −0.1% m/m, 3.7% y/y (from 4.1%); core +0.1% m/m, 3.3% y/y
• Fed held at 3.5–3.75% on Wednesday with three dissents favoring a quarter-point hike
• CME FedWatch: 63% odds of a September hike vs 37% hold
• Real personal consumption +0.4% in June

  📰 Source: The Fed's preferred inflation gauge cooled in June. It might not last — CNN Business (Tier 2) 📰 Source: Core PCE Inflation at 3.3% in June, Edging Down from May — Advisor Perspectives (Tier 2)

💬 Memory Inflation Becomes a Hyperscaler Line Item: Amazon's $20B Capex Raise Is a DRAM Pass-Through

Until last night, the memory shortage was a story about supplier margins — Micron past a $1 trillion market cap, SK Hynix profit up 557%, Samsung guiding shortage conditions through 2028. Amazon's disclosure flips the frame: the buyers are now absorbing it, and disclosing it. With DRAM spot prices up sharply over the past year and 2026 contract pricing forecast to rise triple digits, every hyperscaler capex guide from here contains an embedded memory-price bet. The trade implication is that memory suppliers and cloud buyers are no longer the same trade — they are now opposite sides of the same input cost.

• Amazon: ≈$220B 2026 capex, up from ≈$200B, explicitly on memory cost
• Apple cited "supply constraints" in its FQ4 guide the same evening
• Mizuho's Vijay Rakesh has modeled 2026 DRAM contract pricing up 355% and NAND up 510%
• Micron and SanDisk both +9% today as investors re-bought the supply side of the trade

  📰 Source: AI memory is sold out, causing an unprecedented surge in prices — CNBC (Tier 1, background)

💬 Fireworks AI: $1.5B Series D at a $17.5B Valuation — 4.4× Markup in Nine Months on Inference Specialization

Older context that today's OpenAI price cut makes newly relevant. Fireworks raised $1.505B at $17.5B, up from $4B nine months earlier, on the thesis that enterprises want small models specialized on proprietary data rather than frontier general intelligence — over 95% of the 40 trillion tokens it serves daily come from such specialized models. With OpenAI now selling Luna at $0.20 per million input tokens, the pricing floor under that entire specialized-inference category just dropped, and Nvidia sits on both sides as an investor.

• $1.505B Series D led by Atreides Management, Index Ventures and TCV; Nvidia, Lightspeed, Bessemer and Menlo participating
• $17.5B valuation vs $4B in October — roughly 4.4× in nine months
• Surpassed $1B annualized revenue run rate; serves 40T+ tokens daily

  📰 Source: Nvidia-backed Fireworks hits $17.5 billion valuation as companies pursue cheaper AI models — CNBC (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.