The bet that can’t be unmade
Apr 30, 202613 min readNewsletter
Grow Smart Income
A weekly briefing on AI · Markets · Power · by Kaloian Parchev
WEEK 18
Apr 28–30, 2026
Apr 28–30, 2026
The Fed nobody believes anymore
Powell holds. Warsh waits. Oil spikes. OpenAI misses. And four of the world’s largest companies just committed $600B to a bet they can’t afford to get wrong.
S&P 500
7,137.90 +1.05%
Nasdaq
24,657.57 +1.64%
Bitcoin
~$78,568 +3.77%
WTI Crude
$106.88 +6.95%
Apr 29 close · NYSE/NYMEX · Source: Yahoo Finance, CNBC, Barchart
This week the market fears:
a Fed too divided to cut while oil rewrites inflation
Est. reading time: 7 min
Week in 60
The week that was supposed to be about AI earnings turned into something stranger. Four Magnificent Seven companies reported — all beat on revenue, all raised capex — and then the Fed held rates with the most internal dissents since 1992, Powell announced he’s staying on as Governor to block Trump’s next appointment, WTI broke $106 on a US-Iran blockade that shows no sign of ending, and a leaked WSJ report revealed OpenAI is missing its own user and revenue targets ahead of an IPO valued near $1 trillion. Big Tech is spending more. The anchor AI company is growing less. The Fed is splitting. The strait is closed. And $600B in infrastructure capital is committed regardless.
Big Idea
The Bet That Can’t Be Unmade
This was supposed to be the week that answered the question: is the AI spending boom justified? It answered it — just not the way bulls expected.
Alphabet raised its 2026 capex guidance to $180–190B after Google Cloud grew 63% year-over-year. Meta lifted its full-year infrastructure commitment to $125–145B. Microsoft guided Q4 capex above $40B and said it remains “compute constrained.” Amazon is tracking toward $200B for the year. Combined, the four hyperscalers are spending roughly $600B on AI infrastructure in a single calendar year — a number with no historical precedent.
Big Tech is spending $600B building the world’s most expensive bet on a company that’s missing its own targets.
That company is OpenAI. The WSJ reported Monday that OpenAI missed multiple monthly revenue targets in early 2026 after losing ground to Anthropic in coding and enterprise markets, and fell short of its internal target of 1 billion weekly ChatGPT users by year-end. CFO Sarah Friar privately warned colleagues the company might not be able to fund future compute agreements if revenue doesn’t accelerate. Oracle dropped 5.5%. SoftBank fell nearly 10% in Tokyo. The entire AI infrastructure trade wobbled.
Then Alphabet reported. Then Meta. Then Microsoft. And the capex kept going up. The hyperscalers aren’t waiting for OpenAI to figure itself out — they’re building for a future where demand has to arrive, because the cost of not building is higher than the cost of being early. Alphabet’s Google Cloud backlog nearly doubled to $462B this quarter. That’s not a spend. That’s a commitment. What changed this week isn’t that AI spending accelerated. It’s that the gap between infrastructure reality and demand reality is now visible — and nobody is slowing down because of it.
The Model
THE INFRASTRUCTURE DECOUPLING
Hyperscalers commit $600B+ in AI capex, anchored to projected demand
↓
OpenAI — the primary demand signal — misses user and revenue targets
↓
AI-linked stocks drop; markets briefly price in a crack in the thesis
↓
Hyperscalers report earnings — and raise capex guidance anyway
↓
Infrastructure build continues, decoupled from end-user demand
↓
Either demand catches up, or the largest capital misallocation in history becomes apparent
The hyperscalers aren’t betting on AI demand — they’re betting that refusing to bet is the bigger risk.
By the Numbers
AI INFRASTRUCTURE CAPEX — THE FOUR HYPERSCALERS, 2026 GUIDANCE
Full-year commitments as updated in Q1 2026 earnings · USD billions
The combined $600B+ committed this year exceeds the GDP of most European countries. Even if demand proves weaker than projected, the infrastructure will be built — which means the competitive moat goes to whoever fills it fastest.
WTI CRUDE OIL — PRICE TRAJECTORY DURING IRAN WAR
Selected data points, USD per barrel · Source: CNBC, Barchart, Investing.com
WTI has nearly doubled from its 52-week low since the Strait of Hormuz closed. That’s not a commodity trade anymore — it’s an inflation signal that is actively constraining what the Fed can do next.
INDIA DATA CENTRE CAPACITY — PRESENT VS PROJECTED (GW)
Installed and planned capacity · Source: Economic Survey 2026, Nomura, Google Cloud Press Corner
India is targeting a 6x increase in data centre capacity by 2030 — from 1.4 GW to over 8 GW. Google broke ground on its Visakhapatnam AI hub this week as part of a $15B commitment. The race for AI infrastructure isn’t just US vs China — it’s every large economy trying to own its own compute.
Signal vs Noise
Fed holds rates at 3.5–3.75% — most dissents since 1992, Powell stays on as Governor
In what was likely his final meeting as Chair, Powell held steady but revealed a 4-way split — three presidents opposing an easing bias, one favoring a cut. Kevin Warsh confirmed as successor, but the Fed he inherits is fractured. cnbc.com →
Trump confirms extended Iran naval blockade — WTI surges to $106.88
The Strait of Hormuz remains closed, cutting roughly 20% of global oil flows. Trump told Axios the blockade is “more effective than the bombing.” UAE simultaneously announced its OPEC exit. cnbc.com →
Kevin O’Leary’s 9 GW “Stratos” data centre approved in Utah — entirely off-grid on natural gas
MIDA approved the 40,000-acre campus in Box Elder County; Phase 1 targets 3 GW. At full build-out it would consume more than double Utah’s entire current power output. No hyperscale tenant has been named yet. tomshardware.com →
Google breaks ground on India AI hub in Visakhapatnam — gigawatt-scale, $15B commitment
Built in partnership with AdaniConneX and Nxtra by Airtel, the campus is Google’s largest India investment to date. India’s data centre capacity is projected to grow from 1.4 GW today to over 8 GW by 2030. googlecloudpresscorner.com →
Anthropic’s Patrick Dreisch keynotes in Sofia — first Anthropic appearance in Bulgaria
The BASSCOM AI Conference (500 attendees, Inter Expo Center) brought Anthropic’s GTM Lead to Bulgaria for the first time. Sessions covered Claude Code, on-prem AI coding, and AI-native engineering — all hands-on, no theory. basscom-ai-conf.com →
Macro Signals
UAE exits OPEC — first major departure in decades
Amid war-related damage to its energy infrastructure and a desire for pricing flexibility, the UAE announced it will leave OPEC next month. For oil markets already strained by Iran, it removes a moderating voice from the bloc.
Powell refuses to leave the Fed Board after his chairmanship ends May 15
Breaking with 113 years of tradition, Powell will stay on as a Governor — citing ongoing DOJ investigations and legal attacks on the institution. This denies Trump a fourth Board appointment and sets up a potentially adversarial dynamic with incoming Chair Warsh.
The structural read: two independent institutions — the Fed and OPEC — both fractured in the same week. When the frameworks that managed global money and oil break simultaneously, the next cycle is priced by whoever fills the vacuum first.
What I’m Reading
The Innovator’s Dilemma
I pulled this off the shelf after the OpenAI miss — because what the WSJ described is a Christensen scenario playing out in real time: the market leader optimizing for its existing users while challengers capture new segments from below. What I agree with is the core mechanism: incumbents are rational when they ignore disruptors, right up until they aren’t. What feels incomplete is that Christensen assumed the disruptor had time; in AI, the window between “niche threat” and “enterprise standard” is measured in quarters, not years. What I’m taking from this: if Anthropic continues winning in coding and enterprise while OpenAI defends the consumer chat market, we are watching a classic low-end disruption — just compressed to a speed Christensen never modeled.
One Number
$600B
Combined AI infrastructure capex committed by Alphabet, Microsoft, Meta, and Amazon for 2026 alone — announced in Q1 earnings this week. That’s more than the annual GDP of Sweden. It’s also committed regardless of whether OpenAI, the anchor demand signal, meets its revenue targets.
Source: CNBC / Fortune · Q1 2026 earnings calls · Apr 29, 2026
This Week I Noticed
There’s something telling about the way people talked after the BASSCOM AI Conference in Sofia. Not “interesting session” or “good networking” — but a specific kind of energy you get when people realize a gap is closing faster than they thought. Anthropic showing up in Bulgaria, in person, with hands-on sessions on Claude Code and production AI deployment, signals something I’ve been observing for a while: the serious technical work is no longer happening only in San Francisco. It’s happening here too, with real engineers building real things.
The OpenAI story this week got framed as a crisis. I think it’s the opposite — it’s a market correction inside a larger story that remains intact. What actually happened is that Anthropic gained share in coding and enterprise, Google’s Gemini grew faster than expected, and OpenAI had to compete for the first time. That’s not a red flag for AI. That’s what a maturing market looks like. The companies that were going to win by default are now going to have to earn it. That’s better for everyone building on top of these models — including us.
Quote
“In the short run, the market is a voting machine but in the long run, it is a weighing machine.”
— Benjamin Graham · The Intelligent Investor, 1949
This week the market voted against OpenAI’s partners on a leaked report, then voted for Alphabet after a strong earnings call — 24 hours apart. The weighing machine is still loading the data.
Three things to remember
Big Tech committed $600B to AI infrastructure in 2026 regardless of whether demand is keeping pace — the hyperscalers have decided hesitation is the riskier bet.
The Iran war has become the most consequential macro variable of the year: WTI above $106, the Fed paralyzed by 4-way dissent, inflation re-anchoring upward, and Powell refusing to leave his post.
The real AI geography story isn’t who has the best model — it’s who builds the most watts; India, Utah, and Bulgaria all made that point in the same week.
One Thing To Do
Run a 10-minute audit of any AI tools or SaaS subscriptions you use that depend on OpenAI models. If your workflow is entirely OpenAI-dependent, this week’s news is a prompt — not a panic — to test one Anthropic or Google alternative on your most-used task. Vendor concentration risk is real, and competition just made switching cheaper.
Support This Newsletter
Sponsor
Trading 212
No account yet? Use my link — you get a shot at free shares, and you help keep this newsletter going. No catch.
Open Trading 212 →
Enjoyed this article?
Get the GSI Weekly Newsletter — markets, AI, and investing insights every Thursday.
Subscribe for free →No spam. Unsubscribe anytime.