MY TAKE
I’ve spent all week trying to find the right word to describe what happened on Wednesday, April 29, and the truth is I’m now certain we’re in a new paradigm.
There’s a phrase attributed to Mario Conde, though I haven’t been able to verify it:
“When something doesn’t add up for you, it adds up for someone else.”
Someone smarter, perhaps, or with more information than you.
In four hours, Microsoft, Alphabet, Amazon and Meta presented Q1 results and, together, announced $665 billion of committed 2026 capex for AI infrastructure.
Microsoft around $190 billion. Alphabet $180–190 billion (revising upward). Meta $125–145 billion (also revising upward).
Amazon doesn’t break it out, but in Q1 alone it invested $21.9 billion. The annualized projection exceeds $80 billion for AWS alone.
We’re talking about a figure larger than the military budget of any country except the United States. Larger than all US federal R&D investment. Larger than the GDP of Belgium. In a single year.
Across four companies, Rick.
But the story isn’t the amount — yes, that too — it’s the market’s reaction.
Alphabet announced $180 billion, presented proof of monetization (Google Cloud at $20 billion, +63% year over year; Gemini Enterprise +40% quarter over quarter in paid MAUs) and the stock rose 7% in after-hours.
Its best month since 2004.
Meta announced a similar figure, showed record revenue ($56.31 billion, EPS of $10.44 against $6.67 estimated) and the stock fell 9%.
JP Morgan downgraded it to Neutral the next day. The conclusion seems to make sense: the market no longer rewards spending on AI. It rewards proving you’re making money on it.
That’s a regime change many CEOs still haven’t internalized.
What obsesses me as a CTO is the corollary. If the public market is splitting the hyperscalers into two categories — those who monetize AI and those who bet on it — what happens when the same yardstick reaches the mid-market?
When your CFO asks you next year what the ROI is on the $12 million you’ve spent on GPUs and models, what are you going to show them?
Sundar Pichai’s answer this week was a concrete metric: users paying for enterprise agents, growing 40% quarter over quarter.
The answer from arch-villain Zucky was yet another round of “it will be transformational.” And the market sanctioned it with -9%. It’s already happening.
And meanwhile, in another multiverse, Microsoft and OpenAI revalidated their agreement on April 27.
They killed the AGI clause — the one that gave Microsoft the right to decide when OpenAI reached general intelligence and to cut off its licensing tap.
They scrapped the Azure exclusivity. OpenAI can now deploy on AWS, on Google Cloud, anywhere. I keep thinking about how I put it two weeks ago: “Anthropic is now hostage to two hyperscalers.”
Well, OpenAI just did the opposite. It divorced its only hyperscaler. Microsoft keeps the equity and the license until 2032, but the cage is open. The question is whether OpenAI had the means to pay for the divorce. And the more interesting question: whether the new multi-cloud enterprise OpenAI can compete against Gemini Enterprise, which this week posted +40% quarter over quarter in paid MAUs.
And the last layer. Oracle. On Thursday, Time published a first-person piece by Jill, a technical writer with thirty years at the company, who was asked to exhaustively document her workflows.
That documentation trained the AI that replaced her.
More than 600 Oracle employees signed an open letter on April 17 demanding better severance. 30,000 layoffs. 18% of the global workforce. While Oracle reported its best quarterly growth in 15 years.
This is going to happen more, Rick.
What’s rattling around my head after all this: when the market no longer rewards capex but conversion, and when models are supplied as a utility through hyperscalers that just spent $665 billion to keep them locked in…
What is the sustainable competitive advantage for a company that is neither a hyperscaler nor a frontier lab?
I believe it’s what the Americans call applied AI — that domain-specific application layer that understands the business. But the hyperscalers are building that layer too, via Gemini Enterprise, Copilot Studio, AgentCore. And that’s where things get interesting. Or uncomfortable. Depends on where you’re sitting.
Really, this investment only holds up if AI ends up being the operating system…
Of the universe?
Thanks for reading. If you want more detail, below you have what my minions wrote — and what I read to give you my opinion.
And you know: sharing is life… and commenting is the best.
THE BOMBSHELL OF THE WEEK
$665 billion of AI capex in one week. And the market starts separating winners from losers.
On Wednesday evening, April 29, in a four-hour window, four of the world’s five largest listed companies presented Q1 2026 results simultaneously. The aggregate AI capex figures committed for 2026, once the upward revisions are added up:
- Microsoft: approximately $190 billion (includes $25 billion of component-cost inflation)
- Alphabet: $180–190 billion (revised up from 175–185)
- Meta: $125–145 billion (revised up from 115–135)
- Amazon: Q1 capex $21.9 billion; AWS annualized trajectory exceeds $80 billion
Total committed: approximately $665 billion.
The timeline:
- April 27: Microsoft and OpenAI announce the new phase of the partnership (removal of the AGI clause and the Azure exclusivity)
- Morning of April 29: Alphabet raises guidance. CFO Anat Ashkenazi indicates 2027 capex “will increase significantly”
- Afternoon of April 29: Microsoft, Amazon and Meta report back to back
- April 30: Apple reports $111.2 billion in revenue (fiscal Q2), +17% YoY, authorizes $100 billion in buybacks. JP Morgan downgrades Meta. Ternus debuts on the earnings call
- May 2: Meta closes a $25 billion bond issue to finance the capex
What the mainstream press didn’t tell you: the stock-market divergence is the only signal that matters. Alphabet rose 7% in after-hours and closed its best month since 2004 (+34% in April). Meta fell 9%. JP Morgan cut its rating to Neutral citing a “challenging path” to ROI. The difference, per the analyst consensus, is concrete: Alphabet showed conversion (Google Cloud grew 63% year over year to $20 billion; users paying for Gemini Enterprise +40% QoQ). Meta showed a bet without proof.
And confirmation came from the bond market. Meta closed a $25 billion bond issue on Friday to finance the capex. All six tranches priced at wider spreads than the October 2025 issue. The peak order book was $96 billion, versus $125 billion for the previous offering. Institutional fatigue with Meta’s AI narrative is measurable and growing.
The figure everyone should have on the table: Andy Jassy revealed that AWS’s contracted backlog stands at $364 billion, not including Anthropic’s $100+ billion commitment. Microsoft has a remaining commercial performance obligation of $627 billion. That’s not revenue — it’s future spending already contracted by their customers. The AI infrastructure build-out is locked in for 3–5 years regardless of macro conditions.
The signal: The public market has just bifurcated the hyperscalers into two categories — those who monetize AI and those betting on the future. The metric separating one from the other isn’t total revenue. It’s the growth rate of the AI products specifically, and their marginal contribution to the P&L. Any company, hyperscaler or not, that has to defend AI capex over the next four quarters needs a conversion narrative with concrete metrics. The era of “it will be transformational” is over.
POWER MOVES
John Ternus debuts on his first earnings call. And his words say more than they seem to.
Tim Cook handed the microphone to his successor on April 30, on the first earnings call after the transition announcement. Ternus had a couple of minutes. His words were surgically chosen: he praised Cook’s “financial discipline,” promised continuity. And then he said this: “it’s the most exciting time in my 25 years at Apple to build products and services.”
What the press overlooked: the phrase “build products and services” from an engineer who has spent two and a half decades in hardware is not rhetoric. Ternus is signaling that the roadmap he’s about to inherit — and which he has presumably been shaping for some time — is substantial. Separately, he confirmed to Fortune that he will continue “the company’s tradition of secrecy.” That’s a deliberate message: no analyst should expect more informational openness with the CEO change.
The question no analyst asked: the Services segment reached $30.98 billion this quarter with a record gross margin of 49.3%. It’s already the main margin engine. A hardware-trained CEO inheriting a financial model dominated by services creates an organizational tension that will surface over the next 6–18 months. I’d bet on a reorganization of the Services area within Ternus’s first six months.
Sources: 9to5Mac - Ternus Q2 Earnings Call | Apple Newsroom
Anat Ashkenazi vs Susan Li: the capex narrative war was won on the CFO’s side
On Wednesday, April 29, two CFOs presented structurally similar figures within hours of each other: AI capex revised upward, record spending, massive commitments. Anat Ashkenazi (Alphabet’s CFO since July 2024, ex-Eli Lilly) announced an upward revision of 2026 capex to $180–190 billion, and explicitly warned that 2027 capex “will increase significantly” from that base. The stock rose 7% in after-hours. Susan Li (Meta’s CFO) announced the same magnitude of upward revision, to $125–145 billion. The stock fell 9%.
The difference wasn’t in the figures. It was in the supporting metric. Ashkenazi presented three concrete numbers: Google Cloud grew 63% year over year to $20 billion, Gemini Enterprise paid MAUs rose 40% quarter over quarter, and the remaining commercial performance obligation hit record levels. Every dollar of capex could be justified with a dollar of AI-traceable revenue. Li had no equivalents. Meta’s revenue is advertising — and the conversion of AI capex into incremental margin beyond what the existing business already generates was not demonstrated with data.
The power signal: after three years of CEOs selling AI narrative, the public market has gone back to auditing via the CFO. Ashkenazi has just become the most credible voice in the sector on hyperscaler capex. Every other CFO defending AI capex over the next four quarters will be compared against her April 29 benchmark. That’s a regime change in how these companies communicate.
Sources: Bloomberg - Alphabet Outpaces Meta | CNBC - Investors Trust Google More
Cristiano Amon emerges from the OpenAI smartphone deal with a complete Qualcomm pivot
On April 27, Ming-Chi Kuo’s report confirmed that Qualcomm and MediaTek are jointly designing the custom chip for OpenAI’s agent-first smartphone, with Luxshare Precision as manufacturer. Declared target: 300–400 million annual units in 2028. Qualcomm’s stock rose 7% in the session.
What’s relevant about this move isn’t the deal itself — it’s what it says about the position of Cristiano Amon (Qualcomm’s CEO) right now. Qualcomm had spent two years in forced transition: the iPhone modem business ends with Apple’s own chip, Snapdragon’s patents are being squeezed by Apple Silicon chips in PCs, and the Snapdragon for Windows business still isn’t scaling. Amon needed a next-generation anchor partner. The shape of the design win with OpenAI — a custom chip, not standard Snapdragon — is the first time Qualcomm has been indispensable in a platform play since 2018.
The competitive implication: if the OpenAI smartphone ships at scale, Qualcomm structurally replaces its dependence on Apple-iPhone as a critical platform customer. And positionally, it becomes the only mobile application chip partner capable of competing with Apple Silicon in the post-app-store scenario. That rewrites Qualcomm’s investment thesis. And MediaTek gets on the same train.
Sources: TechCrunch - OpenAI Phone | CNBC - Qualcomm +7%
MONEY TALKS
Microsoft and OpenAI rewrite the deal. Goodbye AGI clause. Goodbye Azure exclusivity.
This was the structurally most significant event of the week, and the mainstream press hasn’t given it the weight it deserves. On April 27, Microsoft and OpenAI formalized an amendment to their partnership. The changes:
- AGI clause eliminated: the original agreement gave Microsoft the right to determine when OpenAI reached AGI and to cut off its technology license. That clause — a source of continuous legal friction — is gone. The license now runs to a fixed date in 2032, independent of capability milestones.
- End of Azure exclusivity: OpenAI can now deploy its full stack on AWS, Google Cloud, or any provider. Microsoft remains the “principal partner,” but can no longer block OpenAI from serving customers on competing infrastructure.
- Revenue-share restructuring: Microsoft stops paying revenue share to OpenAI (a significant cost reduction). OpenAI keeps paying revenue share to Microsoft until 2030, at the same percentage but with an undisclosed total cap. Microsoft retains its equity stake.
- The Amazon connection: TechCrunch reported that eliminating the AGI clause specifically resolved Microsoft’s legal exposure over the $50 billion deal between OpenAI and Amazon. The previous clause would have created a contractual conflict when OpenAI deployed workloads on AWS.
What this really means: OpenAI has structurally freed itself from Microsoft as a constraint. The company can now pursue the full enterprise market on whichever cloud infrastructure its customers prefer — and many Fortune 500 companies are primarily AWS or GCP, not Azure. For Microsoft it’s a face-saving retreat: they lose exclusivity, but retain the IP license until 2032 and keep the equity upside.
Sources: Microsoft Official - Next Phase | TechCrunch - AGI Clause Gone
Ineffable Intelligence: a $1.1 billion seed with no product, no revenue, no roadmap.
On April 27, David Silver — architect of AlphaGo and AlphaZero, a decade at DeepMind — closed the largest seed round in European history for a company founded in November 2025. The round was co-led by Sequoia and Lightspeed. Nvidia, DST Global, Index, Google and the UK Sovereign AI Fund participated. Valuation: $5.1 billion. Product: none. Revenue: zero. Published roadmap: nonexistent.
The thesis: Silver is betting that reinforcement learning — AI learning by trial and error without human-generated data — is the path to systems that exceed human performance across multiple domains. AlphaGo defeated the world’s best Go players in 2016. AlphaStar defeated StarCraft professionals in 2019. His track record is the entire investment thesis.
What this tells the market: when Sequoia and Lightspeed co-lead a $1.1 billion seed with no possible diligence on product or revenue, you’re witnessing a credentials market, not a product market. Top RL researchers at DeepMind, Anthropic and OpenAI are reaching Series C valuations the moment they announce their departure. This compression of the credential cycle will produce large-scale failures in 2027–2028. But in April 2026, the LP that doesn’t put capital into these vehicles assumes it will be left out. Sand Hill Road’s incentive structure is optimized for not being left out, not for being right. We’ll be repeating this.
Sources: TechCrunch - David Silver $1.1B | Bloomberg - Sequoia/Nvidia Back Ineffable
True Anomaly: $650 million to build space interceptors for the Golden Dome program.
On April 28, True Anomaly closed a $650 million Series D at a $2.2 billion valuation. Co-led by Eclipse and Riot Ventures, with Paradigm, Atreides, G Squared, VanEck, Accel and Menlo participating. The company had been selected by the US Space Force to develop prototypes for the Golden Dome program — Trump’s missile defense initiative valued at $185 billion — covering space interceptors, ground systems and command infrastructure.
True Anomaly reaches $1 billion in total capital raised and will grow to 500 employees before the end of 2026. This is not software or AI. It’s manufacturing and systems engineering financed by private capital and anticipated government contracts. The Golden Dome budget allocation creates a demand floor that de-risks early-stage investment in a way purely commercial markets can’t. The signal for VCs: national defense as a venture capital thesis is back, and at scale. Not the “dual-use software” cliché of 2018–2022, but heavy aerospace hardware.
Sources: Axios - True Anomaly 650M | SpaceNews - $2.2B Valuation
PRODUCT SECRETS
OpenAI is preparing a smartphone with agents instead of apps. And a parallel Jony Ive device.
On April 27, analyst Ming-Chi Kuo reported that OpenAI is developing a smartphone where AI agents replace traditional applications, with Qualcomm and MediaTek jointly designing a custom chip and Luxshare Precision as manufacturer. Target: 300–400 million annual units in 2028. Qualcomm rose 7% on the rumor alone.
This is independent of the hardware program with Jony Ive (the $6.4 billion acquisition of io). The Ive project is developing a non-mobile device first: reportedly a smart speaker with a camera, followed by glasses, a lamp and earbuds.
What the press underestimates: OpenAI is simultaneously running two hardware roadmaps with different form factors, different supply chains and different manufacturing partners. It’s an extraordinary resource commitment for a company that had no hardware capability 18 months ago. The organizational complexity of managing two simultaneous hardware programs while also restructuring the enterprise business (Sora shutdown, triple executive exit, Fidji Simo on leave) is substantial. I doubt it has the organizational capacity for both.
The real threat model: if the agent-first smartphone ships at scale in 2028, it doesn’t compete with the iPhone on specs or ecosystem. It competes by making app stores structurally irrelevant — if AI agents complete tasks without launching discrete applications, the App Store’s 30% commission disappears as a business model. That’s a longer-term threat than the market is currently pricing.
Sources: TechCrunch - OpenAI Phone | CNBC - Qualcomm +7%
Amazon’s chip empire: $50 billion that almost nobody is covering properly.
Andy Jassy revealed on the April 29 earnings call that Amazon’s custom chip business (Trainium, Graviton, Nitro) already generates more than $20 billion in annualized revenue, growing at triple-digit rates year over year. And he said — verbatim — that if it were sold externally on the open market, the business would generate $50 billion a year. Amazon has $225 billion in contracted Trainium revenue commitments with customers. Trainium3 — launched in early 2026 — is “almost completely reserved.” AWS also announced an alliance integrating Cerebras inference chips alongside Trainium for low-latency AI workloads.
The number nobody covered correctly: Meta signed a multibillion-dollar deal for Graviton5 — the infrastructure of a direct competitor — because its $145 billion capex can’t be satisfied by a single supplier. Meta moving compute spend to AWS, instead of depending exclusively on Nvidia, is the structural shift. Nvidia’s pricing power depends on being the only viable option. The moment hyperscalers can substitute at scale, Nvidia’s margin structure changes. If I were an Nvidia analyst, I’d be discounting this.
Sources: TIKR - Amazon Chip Business | TechCrunch - Meta Amazon Chip Deal
Gemini everywhere in one week: cars, TVs, Mac, file generation, enterprise.
Google announced a dense cluster of Gemini extensions in the last week of April, deliberately coinciding with earnings. Full deployment of Gemini in vehicles with Google built-in (tens of millions of cars), replacing Google Assistant. A native Gemini app for macOS. Generation of downloadable formatted files in more than 9 formats directly from the chat. A new Gemini Enterprise app with Agent Designer, long-running agents, Skills, Projects and an Inbox for managing agent activity. Generative tools on Google TV via a Gemini tab on TCL TVs in the US.
The strategic read: Alphabet’s CFO revealed that Gemini Enterprise paid MAUs grew 40% quarter over quarter. That’s the metric that justified Alphabet’s +7% while Meta took a -9% for spending the same. Google is demonstrating enterprise AI conversion faster than any other hyperscaler this quarter. And the past week’s extensions are, read together, a ubiquity strategy: put the agent on every surface where the user already is, before OpenAI or Anthropic can replicate it. It’s the distribution play Microsoft ran with Copilot in Office — but at consumer hardware scale.
Sources: Google Blog - Gemini April Drop | TechCrunch - Gemini in Cars
REAL NUMBERS
AWS backlog published April 29: $364 billion (Q1 close), excluding Anthropic’s $100+ billion commitment. Comparison revealed the same week: Microsoft’s remaining commercial performance obligation reaches $627 billion. This is not revenue — it’s future spending already under contract. The AI infrastructure build-out is locked in for 3–5 years regardless of the macro. It’s the most under-reported figure of the week.
Amazon’s chip business (revealed April 29 by Andy Jassy): current annualized revenue >$20 billion, growing at triple digits year over year. $225 billion in contracted Trainium commitments with customers. Trainium3 “almost completely reserved.” If sold externally: $50 billion a year. A figure that matches Nvidia’s data-center revenue from two quarters ago.
Meta bond sale closed May 1–2: $25 billion in six tranches. Wider spreads than the October 2025 offering. Peak order book of $96 billion, versus $125 billion for the previous issue. A concrete metric of institutional fatigue with Meta’s AI narrative. It’s the first time the bond market has repriced a Big Tech’s capex credibility downward.
Apple Q2 (published April 30): revenue $111.2 billion (+17% YoY), Services $30.98 billion with a record 49.3% gross margin, $100 billion buyback authorization. EPS $1.71 vs $1.63 estimated. Q3 guidance raised 14–17%.
Alphabet Q1 (published April 29): Google Cloud $20 billion (+63% year over year). Gemini Enterprise paid MAUs +40% quarter over quarter. Best month for the stock since 2004 (+34% for the full month of April).
Tech layoffs 2026 year to date (tracker updated May 1): 100,443 workers across 155 events. ~830 jobs lost per day. Q1 2026 closed at ~80,000 with AI cited as the explicit driver in ~20% (the real figure with implicit drivers exceeds 50%).
Sources: TIKR - Amazon Chip Business | Bloomberg - Meta $25B Bond | CNBC - Apple Q2 | Tom’s Hardware - Q1 2026 80K Layoffs
THE DRAMA
Oracle’s employees trained the AI that replaced them. The April 30 exposé.
The most significant human-interest piece of the week was published in Time on April 30, and it has strategic implications that go beyond the labor story.
Jill, a technical writer with 30 years at Oracle, recounted in first person how she and her team were required to exhaustively document their workflows. The internal framing was “process documentation” and “AI enablement.” That documentation was used to train Oracle’s AI systems. Afterwards, her entire team was laid off. Time’s reporting goes beyond Jill: multiple former Oracle employees described the same systematic pattern to the journalist — structured knowledge-extraction programs, framed as AI readiness, followed by role elimination. It’s the first time a mainstream outlet has documented the pattern with cross-corroborated testimonies.
Strategic implication for CTOs: knowledge-extraction programs described as “AI preparation” or “workflow documentation” are starting to be recognized as precursors to layoffs. Employees are beginning to refuse them or deliberately slow them down. If your company is implementing similar programs internally, this Time article will make your workforce substantially more suspicious of the intent. And the EU AI Act has specific provisions arriving in May–June on AI systems applied to HR (including dismissals). Watch this.
Sources: Time - Oracle Workers AI Replacement
The capex revolt: Meta falls 9% and JP Morgan breaks the consensus.
Meta posted Q1 revenue of $56.31 billion — a genuine beat. EPS of $10.44, crushing the $6.67 estimate. And the stock fell 9%. JP Morgan downgraded Meta to Neutral from Overweight on April 30, citing a “challenging path” to generating returns on the capex commitment. But JP Morgan’s analysis surfaced a specific issue: Meta is now spending what it would cost to acquire Ford, GM and Chrysler combined — on AI infrastructure — without demonstrating a monetization path separable from advertising.
Meta’s core revenue is advertising. The AI infrastructure primarily enables better ad targeting and content recommendation. The question investors are now asking is: what is the incremental revenue from the $145 billion of capex above what $60–70 billion would have generated? Zuckerberg hasn’t answered with data.
The bond market’s response is equally eloquent. The $25 billion issue closed on Friday, May 2, but all six tranches priced at wider spreads than the October 2025 offering. The peak order book was $96 billion, versus $125 billion previously. The institutional fixed-income market is quietly repricing Meta’s AI credibility. This is the start of a different conversation about Meta, not an isolated spike.
Sources: 247 Wall St - Meta 8% Drop | Fortune - Meta $145B Capex | CNBC - Investors Trust Google More
THE WEEK AHEAD
Thursday, May 7 — Microsoft’s Rule of 70 notifications: 8,750 eligible employees receive the formal voluntary buyout offer. Acceptance window until approximately June 6. Watch for: leaks of the package terms (the financial details haven’t been disclosed), employee responses on LinkedIn and Blind, and whether Microsoft expands eligibility or sweetens the package if acceptance falls below target. It’s the first voluntary retirement program in Microsoft’s 51 years — the outcome will set the precedent for how the rest of Big Tech manages workforce reductions.
Mid-May — Cerebras IPO roadshow and pricing: aiming for a Nasdaq listing in mid-May under the ticker CBRS, target valuation $23 billion. It has a $20 billion OpenAI commitment and a $1 billion partnership with AWS. It will be the first major AI hardware IPO of 2026. The pricing will set the reference for AI-specific chip companies in the public market.
May 20 — Meta’s layoffs begin (8,000 employees): the preceding weeks tend to generate internal leaks about which teams and functions are affected. Watch Glassdoor and Blind for early signals. The specific functions eliminated will reveal which parts of Meta’s AI strategy are being de-prioritized after the market’s April 30 punishment.
Throughout May — OpenAI’s formal enterprise reorganization: after the April 27 divorce from Microsoft (removal of the AGI clause, end of Azure exclusivity), expect an organizational announcement formalizing the new multi-cloud enterprise structure. Any communication about new enterprise B2B leadership will clarify whether the company is consolidating or resetting.
May — Upcoming AI-frontier earnings: watch Snowflake, Palantir, and CrowdStrike reporting results — they’re the first post-Big-Tech wave that will have to demonstrate AI-to-revenue conversion under the new Wall Street yardstick established on April 29. Any defense of AI capex without concrete metrics will be punished.
May–June — EU AI Act compliance deadlines for high-risk systems: several EU AI Act provisions affecting high-risk AI systems have implementation deadlines in May–June 2026. European companies deploying AI in HR (including hiring, performance evaluation and dismissal decisions) face specific requirements. The “Oracle trains AI then lays off” pattern from Time’s April 30 exposé may attract regulatory attention in EU jurisdictions.