$GOOGL printed an EPS of $5.11 against a $2.63 consensus. On paper, that’s a 94% beat. In practice, it’s roughly zero. Welcome to the most misread earnings headline of the quarter.
The 94% beat that wasn’t
Strip out the noise and the picture changes fast. Alphabet’s net income included a $36.9 billion gain on equity securities that added $2.35 to diluted EPS. That’s a non-cash, non-operating mark-to-market on the company’s investment portfolio. Pull it out and you get an adjusted EPS of $2.62. The Street wanted $2.63. CNBC pegged adjusted EPS at $2.62 versus $2.63 expected. So on the metric most analysts actually model, $GOOGL came in line. Not double. Not 94%. In line.
Which means the financial press headlines you’ve been reading this week are technically true and structurally misleading. Most of the EPS jump was a portfolio gain, not operating leverage. That’s worth saying out loud before getting carried away.
But here’s the twist: the stock still ripped. $GOOGL closed Wednesday at $349.94 and ran 9.96% on Thursday to close at $384.80, an all-time high. Roughly $250 billion of market cap created in a single session. Market cap now sits around $4.4 trillion, more than double its 52-week low of $147.84 set on May 7, 2025. So the market didn’t reward the headline EPS, it rewarded what was sitting underneath it. That’s the part worth dissecting. (And if you’re still hazy on the GOOG vs GOOGL split, the share class structure is the place to start.)
Cloud is the actual story
Google Cloud did $20 billion in revenue, up 63%. That’s not a typo. Cloud revenue increased 63% to $20.0 billion, led by enterprise AI Solutions and AI Infrastructure on GCP. For comparison, $MSFT Azure printed +40%, which itself was above the +39% Street whisper. AWS at $AMZN has been growing in the high 20s. So in this quarter, Google Cloud accelerated harder than both rivals. That’s a first.
The number that matters more than the revenue print: backlog. Cloud backlog nearly doubled sequentially to $462 billion. Doubled. In one quarter. That’s roughly twenty-three quarters of current Cloud revenue locked in. Investors who’ve spent two years asking when Google Cloud would stop being a “scale at any cost” line item just got their answer. Operating margin hit 32.9%, up from 17.8% a year ago. The unit is now genuinely profitable at scale.
And then there’s the line management dropped on the call: Pichai said the company is compute constrained in the near term and that Cloud revenue would have been higher with more capacity. That’s not a complaint, it’s a flex. When demand exceeds your ability to ship, you have pricing power. Cloud GenAI model revenue grew nearly 800% year over year. New customer acquisition doubled. Multiple billion-dollar-plus deals closed in the quarter. This isn’t a soft thesis anymore.
This is the Cloud inflection point we’ve been tracking. The ROIC is finally showing up.
The bear case here isn’t gone. It’s just narrower. Cloud margins still trail AWS by a wide gap, and a backlog isn’t recognized revenue. Some of that $462B is multi-year, lumpy, and exposed to renegotiation if compute economics shift. Don’t read the doubling as cash in the bank. Read it as visibility, which is rarer and almost as valuable.
Search isn’t dying, it’s leveling up
Eighteen months ago the consensus on $GOOGL was that AI chatbots would slowly eat search advertising. ChatGPT was a king-killer. Perplexity was the new homepage. The Search moat was crumbling. That was the bear thesis. It is now in worse shape than it’s been in two years.
Google Services revenue grew 16% to $89.6 billion, with Google Search & other up 19% and YouTube ads up 11%. The Search number is the headline: 19% is the fastest growth in years for what is supposedly a dying franchise. AI Overviews, AI Mode, and Gemini integrations are not cannibalizing the ad load. They’re expanding it. That’s the inverse of the thesis bears were selling for the entirety of 2024.
For the longer dissection of why this was always more defensible than the chatbot crowd thought, see the work on the $175B search moat. The short version: distribution plus data plus a decade of query intent labeling is hard to dislodge with a chatbot, even a very good one.
Caveat. One quarter is one quarter. AI search displacement is a slow grind, not a cliff. The dynamic to watch isn’t whether queries decline. It’s whether monetization per query holds when more answers happen above the fold without a click. Q2 will tell us more.
The capex divide: why MSFT got punished and GOOGL got rewarded
Both companies raised capex on the same day. Both reported revenue and EPS beats. The market reaction was opposite.
- $GOOGL: 2026 capex guidance lifted to $180-190B, up from $175-185B. Management said 2027 capex will “significantly increase” again. Stock up nearly 10% on Thursday.
- $MSFT: 2026 capex guided to $190 billion, well above the $154.6B Visible Alpha consensus, with $25B attributed to higher component prices. Stock down almost 4% despite the beat.
- $META: Strong numbers, capex raised, and punished harder still. Down roughly 9%.
Same playbook, three different outcomes. What gives?
The market is now grading capex on visible return, not on ambition. $GOOGL’s case is straightforward: Cloud +63%, backlog $462B, Search reaccelerating, TPU revenue showing up as a third-party hardware line. The capital is producing measurable output. $MSFT has Azure +40% and an AI run rate of $37B, which is real, but the OpenAI dependency, the data center depreciation hit on margins, and a capex number 25% above consensus rattled investors. $META’s spend keeps going up while the AI revenue link is harder to draw on a single slide.
This is where the silicon substrate thesis matters. Google’s TPU v6/v7 stack is the only credible internal alternative to $NVDA accelerators at hyperscale, and it’s increasingly being sold as a service to third parties. The Q1 call confirmed something more aggressive: TPUs are now being delivered directly to select customers’ data centers, with those hardware agreements rolled into the Cloud backlog. That’s a structural shift. Google is no longer just renting TPU capacity inside its own walls, it’s becoming a chip vendor. Microsoft has Maia, Amazon has Trainium, but Google’s TPU is on its seventh generation, trained Gemini end to end, and is now sitting on someone else’s racks. That’s a different maturity curve.
One asterisk on Cloud margins: the Wiz acquisition closed in March and management flagged a low single-digit headwind to Cloud’s operating margin for the rest of 2026. Worth modeling, not worth panicking about.
Side note on capital return: $GOOGL also bumped the dividend 5% to $0.22 per share. Small in dollar terms, signal in posture terms. The full picture on Alphabet’s dividend policy sits alongside the buyback machine, which has been doing the heavy lifting on per-share metrics for years.
$NVDA: the pure-play at $5.2 trillion
You can’t talk about hyperscaler capex without talking about who’s getting paid. The answer remains, overwhelmingly, $NVDA.
$NVDA hit a $5.2 trillion market cap in late April 2026, with non-GAAP EPS of $1.62 versus $1.52 consensus on its prior print. The Q1 FY27 outlook calls for $78 billion in revenue, plus or minus 2%, with no Data Center compute revenue from China assumed. That’s the cleanest read on demand the market is going to get for another two weeks. Earnings are scheduled for May 20, with consensus skewed bullish: 57 Buys, 2 Holds, 1 Sell, average target $268.61.
The bull case writes itself when you see a $462B backlog at $GOOGL Cloud and a combined $190B capex line at $MSFT. Most of that flows through Jensen Huang’s order book, directly or indirectly. Total $NVDA supply commitments stand at $95.2 billion. The roadmap, Blackwell Ultra into Vera Rubin, keeps the performance-per-dollar story intact for another product cycle.
The bear case is shorter and worth saying. At $5.2T, $NVDA is priced for execution, not for surprise. The China zero-revenue assumption is conservative on guidance but it’s also baked into expectations. Any thaw on export controls is upside, any escalation is symmetrical downside. And the customer concentration risk is real: when four hyperscalers account for the bulk of your data center revenue, your earnings are downstream of their capex committee meetings.
May 20 is the catalyst. If $NVDA prints a clean beat with a clean Q2 guide above $80B, the cycle gets another leg. If guidance is in line and the China commentary turns cautious, expect chop.
The third pole read
Three things to walk away with.
- One. $GOOGL’s quarter was good but the headline EPS is a head-fake. The real beat was operating: Cloud accelerating, Search reaccelerating, margins expanding, backlog providing multi-year visibility, and management openly saying they’d sell more if they had the chips. The full-stack thesis (foundation model, custom silicon, cloud, distribution) finally produced a quarter where every layer was firing at once. That’s rare, and that’s why a stock at $4 trillion in market cap can still print +10% in a session.
- Two. $MSFT didn’t have a bad quarter. It had a quarter where capex anxiety overrode operating strength. The OpenAI relationship is starting to feel less like a strategic edge and more like a $190B annual obligation with margin pressure attached. Until investors can model the AI revenue against the data center depreciation curve, the stock will trade choppy on prints.
- Three. $NVDA is still the cleanest expression of this cycle, and the most expensive. At $5T+, you’re not buying optionality, you’re buying execution. The May 20 print will set the tone for the entire AI complex through summer.
The framing we use for all of this sits in the physics of capital. Capital flows where return is most visible. Q1 2026 was a quarter where $GOOGL made the visibility case better than its peers. Doesn’t mean it stays that way. Means right now, that’s where the gravity is.
