Google Prints, Tesla Slips, Oil Won't Quit — The AI Capex Arms Race Is The Only Trade That Matters

After the close on July 22, Alphabet dropped a monster print. Revenue $119.8B, up 24%. Google Cloud $24.8B, up 82% year-over-year — that is the number that matters. And the tell for the next twelve months: FY26 capex raised to $92B, nearly doubling year-over-year.

Tesla printed the other side of that coin. Revenue $28.2B was a beat and up 26% YoY, but EPS missed, GAAP margin came in at 1.4%, free cash flow went negative, and management pushed volume production of the Cybercab, the Tesla Semi, and Megapack 3 out again. FY26 capex was pushed to $25B — Tesla is spending like a hyperscaler without hyperscaler unit economics yet.

Wrap it all in Brent still above $94, WTI near $85, US strikes on Iran now on their twelfth consecutive night, and Trump on the wires threatening more. That is the tape. Here is what happened, what to expect next, and where I think this is going.

Alphabet (GOOGL) — BEAT. Revenue $119.8B (+24% YoY) vs ~$117B est. EPS $2.85 adj vs $2.80 est. Google Cloud $24.8B, +82% YoY — the blowout line. Search a hair light on the whisper. FY26 capex guide raised to ~$92B (from ~$85B), near-doubling YoY. Stock fell in after-hours on the capex hike, not the print.

Tesla (TSLA) — MIXED, tilted miss. Revenue $28.2B (+26% YoY) beat. EPS missed. GAAP operating margin 1.4%. Free cash flow turned negative. FY26 capex lifted toward $25B. Cybercab, Tesla Semi, and Megapack 3 volume-production timelines all slipped again. Stock down ~4% after-hours.

What the market is trading off next: (1) Microsoft, Meta, and Amazon prints — do they confirm the Alphabet Cloud + capex-raise combo, or does one soft cloud number light up the 'peak margin' bear case; (2) hyperscaler read-through into AVGO / NVDA / MRVL, memory (MU, SNDK), and power/cooling (VRT, ETN, GEV); (3) the Fed reaction function to $94 Brent — a hawkish July hold with real hike-risk in the language reprices the front end again; (4) SPY 740 as the master pivot into month-end. Above 740 with Msft confirming, the tape tags 750. Below 740 on a soft print plus firm crude, the low 720s open up.

Headline: $119.8B revenue, +24% YoY, +23% in constant currency. Twelfth straight quarter of double-digit growth. EPS $2.85 vs $2.80 est. Search grew but slightly missed the whisper — that is the only nit anyone will pull on.

The line that moves the stock: Google Cloud $24.8B, up 82% year-over-year, more than doubling from the same quarter last year. That is a real number, at real scale, and it validates the entire hyperscaler AI capex thesis. Cloud services backlog swelled again, which means the revenue you are seeing is a lagging read on already-booked demand.

Capex is where the story gets loud. Alphabet raised FY26 capex guidance to roughly $92B, up from the prior ~$85B and near-doubling year-over-year. Q2 alone was in the mid-twenties, driven by servers, TPUs, and data-center construction. Management said the bill goes higher into 2027. Read that as: the arms race is not slowing, it is accelerating.

Why the stock sold off in after-hours anyway: the Street had been sniffing at 'peak capex' as a bear case and Alphabet just told them capex is going higher, not lower. That is a short-term margin worry. It is not a business worry. Every quarter that Cloud grows 60-80%, the capex investment gets validated in the revenue line 12-18 months later.

Revenue $28.2B, up 26% YoY, was a clean beat. Trailing-twelve revenue crossed $100B for the first time. That is the good half of the print.

The bad half: GAAP operating margin came in at 1.4%. EPS missed. Free cash flow flipped negative. And Cybercab, Tesla Semi, and Megapack 3 — the three products management had been pointing at as the next revenue leg — all had their 'volume production' timelines slipped again.

Capex guidance was raised toward $25B for the full year, driven by Optimus buildout, Dojo/data center compute, robotaxi infrastructure, and the AI5 chip ramp. Tesla is now spending like Meta or Oracle on AI infrastructure — but the revenue backing that spend still lives in a car business with compressed margins and a robotics/robotaxi business that has not scaled yet.

The tape's honest read: the auto business is fine, deliveries grew 25% last quarter, but the story that gets Tesla out of the low-teens P/S multiple lives entirely in FSD, robotaxi, and Optimus — and none of the three took a step forward in this print. The stock traded down 4% after-hours. That is the market pricing 'more capex, later revenue' correctly.

Zoom out. Add up what the tape has told us across the last two weeks: Alphabet $92B FY26. Meta already at the $70B+ zone. Microsoft on track for $80B+. Amazon spending materially more. Oracle sold out on capacity. Tesla raising to $25B. Even the second-tier hyperscalers are guiding higher. That is a $350-400B annual industrial buildout, running through a handful of vendors.

The read-through map:

One — silicon. NVDA gets every incremental TPU/GPU dollar it can source, but the tell now is AVGO and MRVL benefiting from the hyperscaler custom-silicon push. Every time Alphabet says 'TPU,' AVGO's XPU line gets bigger.

Two — memory. This is why the MU and SNDK trade is not over even though I'm flat both today. HBM at the leading edge is sold out through 2027. Enterprise NAND demand is finally getting a hyperscaler bid. Every capex raise from a hyperscaler is a delayed order for memory.

Three — power and cooling. VRT, ETN, GEV, PWR. You cannot spin up $92B of data centers without transformers, switchgear, chillers, and grid interconnects. This is the sub-industry the tape is still under-owning relative to the capex number.

Four — networking and optics. ANET, CIEN, COHR. The interconnect layer scales with the compute layer or the compute layer strands. Read the capex, buy the plumbing.

Five — the counter-trade. Everything that has to fund itself on the discount rate — long-duration unprofitable growth, low-quality small caps, junk beta — gets punished as capex crowds out capital and rates stay firm. The AI capex bid is a bull trade in a narrow list and a bear trade in a very wide one.

Brent above $94, WTI near $85. Twelfth straight night of US strikes on Iran. Trump is on the wires threatening more. Houthis targeting Saudi tankers in Bab el-Mandeb. Rubio saying Iran is 'not serious' about a deal. This is not a story that is fading in a session.

But the base case has not changed from yesterday: unless the Strait actually breaks, this is a headline bid on a barrel that is still moving. WTI running into the 100-day SMA, Brent testing the June 8 pivot, RSI stretched. A clean intraday tag of $86-87 WTI that closes back under $84 is still the unwind signal I am stalking. A close above $88 opens $92-95 and changes the regime.

The new wrinkle is the Fed. $94 Brent into a services CPI print that has been stuck above 3% is exactly what pushes the July FOMC toward hawkish hold with a real hike-risk tail. Fed funds futures are pricing that now. The 10-year at 4.64% is the honest reflection. Higher oil + firm yields + capex-heavy Big Tech = multiple compression pressure that only gets offset by an actual earnings acceleration. Which we just got from Alphabet Cloud and not from Tesla.

Side-by-side, this is what the hyperscalers and adjacent spenders have put on the tape for FY26 capex and what it implies for AI infrastructure demand into the next quarter (Q3 2026 prints). Sources: company press releases and earnings commentary; deltas are vs FY25 actuals.

Alphabet (GOOGL): FY26 capex raised to ~$92B (from ~$85B prior guide, vs ~$52B FY25). Delta: +$40B YoY, +77%. Signal: TPU + data-center + servers. Read-through: direct bid for AVGO custom silicon, HBM memory (MU/SK Hynix), power (VRT/ETN/GEV), interconnect (ANET). Management flagged capex goes higher into 2027 — no peak in sight.

Microsoft (MSFT): consensus and management commentary point to ~$80–85B FY26 (vs ~$56B FY25). Delta: +~$25–30B YoY, +45–55%. Signal: Azure capacity still capacity-constrained; every quarter has been 'demand > supply.' Read-through: NVDA GPU allocation, liquid cooling, grid power. July print is the confirm/deny for the Alphabet-led capex re-acceleration.

Meta (META): last guide bracketed $70–75B FY26 (vs ~$40B FY25). Delta: +~$32B YoY, +80%. Signal: Llama training clusters, custom MTIA silicon ramp. Read-through: NVDA, AVGO on MTIA, memory, and a large power/land footprint. Watch the July print for whether the top of the range moves up again.

Amazon (AMZN): AWS-led capex tracking $105–115B FY26 (vs ~$83B FY25). Delta: +~$25–30B YoY, +30–35%. Signal: Trainium/Inferentia + third-party (NVDA) GPU capacity. Read-through: mixed silicon spend, heavy real-estate/power buildout, memory pull-through.

Oracle (ORCL): OCI capex ~$35B FY26 (vs ~$21B FY25). Delta: +$14B YoY, +67%. Signal: multi-year sold-out backlog (~$450B RPO after the OpenAI/xAI deals). Read-through: pure GPU cluster spend — NVDA-heavy, high power intensity per dollar.

Tesla (TSLA): FY26 capex lifted to ~$25B (vs ~$11B FY25). Delta: +$14B YoY, +127%. Signal: Dojo/AI5, Optimus, robotaxi compute. Read-through: NVDA (still the swing GPU vendor), custom silicon foundry (TSM), and edge-compute enablers. Different customer, same demand vector.

Roll-up: the top six spenders alone are on track for ~$360–380B of FY26 capex, up from ~$260B FY25 — roughly +$100–120B YoY of net-new infrastructure demand. Even haircutting for construction lead-time and pull-forward, that translates into a durable double-digit YoY beat cadence at the vendors for at least the next 2–4 quarters.

What it implies for Q3 2026 (next quarter): (1) Silicon — NVDA data-center revenue prints another sequential record; AVGO custom-silicon (XPU/MTIA/TPU) grows faster than the Street models. (2) Memory — HBM sold out, enterprise NAND demand finally showing hyperscaler order flow, MU and SNDK guides get walked up. (3) Power/cooling — VRT, ETN, GEV, PWR see backlog extensions past 2028; this is still the most under-owned expression of the capex number. (4) Networking/optics — ANET, CIEN, COHR track compute one-for-one; any softness is a buy, not a sell. (5) Real-estate/grid — the physical bottleneck moves from GPUs to megawatts; watch DLR, EQIX, and utility interconnect queues.

The clear risk to the box: one soft cloud growth number against a rising capex guide is what lights up the 'peak margin, capex ahead of revenue' bear case. Alphabet passed that test today. Microsoft is the next tape check.

This week: Meta, Microsoft, and Amazon are the next set of hyperscaler prints (Microsoft/Meta late July, Amazon end of month). Every one of them is going to be judged on the same three lines — cloud growth, capex guide, and how much of the capex is showing up in the revenue line. Alphabet just raised the bar. If Microsoft matches with Azure growth and a lifted capex number, the AI trade re-accelerates into August. If any of the three prints soft cloud growth against rising capex, the 'peak margin' bear case gets its first real day in court.

Semis: NVDA, AVGO, MRVL are the direct beneficiaries of Alphabet's capex raise. AVGO in particular — every TPU dollar is a AVGO custom-silicon dollar. Expect the sell-side to walk numbers higher into the next print.

Memory: MU and SNDK sit exactly where I said they would after the last leg — flat, waiting for a real pullback or a new hyperscaler capex data point. Alphabet just gave us the data point. If the tape hands me a 5-8 session range compression that shakes out fast money, that is the reload.

Tesla: the print does not change the range. It stays a headline stock trading on FSD/robotaxi milestones, not on the auto P&L. The near-term bear case is margin, the near-term bull case is an unscripted Optimus or robotaxi catalyst. I would rather trade the enablers (NVDA, ARM, semis into edge compute) than the ticker.

Oil: watch the Strait, watch WTI $86-87 as the exhaustion zone, watch a $84 close as the unwind tell. Watch OPEC+ for a July-end signal. Nothing has actually stopped flowing yet.

Fed: the July meeting is a hold, but the language is what trades. Any acknowledgment of oil-driven goods inflation and the front end reprices harder. That is the risk to the 'higher SPY through summer' base case.

SPY: 740 is still the master pivot. 745 is the mid-range. A clean tag of 750 needs Microsoft/Meta to confirm the Alphabet capex story. A break of 740 on a soft Msft print plus a firm crude close is the fast tape down to the low 720s.

Alphabet: I am not chasing the after-hours dip. If the tape gives me a full session flush that reclaims by mid-day, that is the tell to fade the capex fear. If it doesn't reclaim, the peak-margin narrative gets a week to run and the entry gets better.

Tesla: no interest as a directional long here. The trade is either the puts if 300 breaks or a long-only headline scalp on a robotaxi catalyst. Not a swing name for me until margins bottom.

Semis / AI plumbing: this is where the arms race trades. AVGO on any pullback, VRT and ETN as the industrial expression, ANET as the interconnect play. Small starter positions with room to add on a broad tape flush.

Memory: still flat MU and SNDK. Watching for the reload setup — a session where fast money capitulates and the sell-side revisions catch up to the capex data. Alphabet's number pulls that revision forward.

Oil / energy: not chasing crude. XLE fade is the trade if the Strait de-escalates, XLE add is the trade only on an actual outage. Flat and watching, same as yesterday.

COIN: still on, still targeting $200-plus. Nothing about today's tape changes the stablecoin flywheel thesis.

SPY: 740 is the pivot, 745 is the mid, 750 is the tag on a Microsoft confirmation. Below 738 with volume, I hedge with QQQ puts rather than flatten longs.

Alphabet showed you that the AI capex is being converted into cloud revenue. Tesla showed you that not every capex story converts on the same timeline. Oil is telling you the macro regime is stagflation-lite until it isn't, and the Strait is the switch.

The trade is not 'buy Big Tech.' The trade is: own the arms race — silicon, memory, power, cooling, interconnects — and let the hyperscalers spend the money that flows down to those vendors 12-18 months from now. The trade is: respect $94 Brent as a Fed problem, not a supply problem, until proven otherwise. The trade is: run SPY 740 as the master pivot and let the next two hyperscaler prints decide whether we tag 750 or flush 720.

I am flat memory, long COIN, watching XLE and semis, and letting Microsoft, Meta, and Amazon tell me what the next month looks like. Discipline over conviction. Trade the arms race, not the headlines.

Everything on this page is my opinion based on publicly available reporting and my read of the tape as of publication on July 23, 2026. It is not investment advice and not a recommendation to buy, sell, or short any security or derivative. Positions and levels discussed can change without notice and can invalidate the same session. Verify independently before risking capital.