MY TAKE
What a week… this keeps accelerating…
SoftBank has borrowed $40 billion from five banks to bet on OpenAI — takes some nerve to borrow that money… and even more to lend it.
SpaceX is preparing a stock market debut at $1.75 trillion, and that one actually looks pretty good to me. Meanwhile, Amazon, Alphabet, Microsoft and Meta have committed $650 billion to AI infrastructure in 2026 alone.
There’s a pattern this week I see VERY CLEARLY.
Capital is concentrating at unprecedented speed into an ever-smaller number of bets. Huge, binary, irreversible bets. Taleb would not approve.
SoftBank isn’t diversifying — it’s putting everything on one card with a 12-month bridge loan whose only viable exit is an OpenAI IPO before March 2027.
Then SpaceX isn’t testing the market — it’s preparing the largest stock market debut in the history of technology. And the four big clouds aren’t “investing in AI” — they’re committing volumes of capital that will push Amazon into negative free cash flow this year, I believe for the first time in its history (I checked, roughly).
When the cost of being wrong is value destruction at industrial scale, and everyone pushes ahead anyway, there are only two possible explanations.
Either the AI thesis is correct and we’re facing the biggest economic transformation since the internet — or I’d go further… since the industrial revolution — or we’re facing the biggest capital allocation bubble in history, amplified by a market that rewards aggression and punishes prudence.
What worries me isn’t which of the two is right.
What worries me is that nobody seems to be hedging the possibility that it’s the second. SoftBank’s loans have no hedges. Big tech’s capex has no review clauses. The layoffs are executed as if they were permanent.
The entire system is operating as if option B didn’t exist.
And watch this: OpenAI just killed its e-commerce product, Instant Checkout, because it isn’t working… it reminds me of Google — you can trust the continuity of its products less than you can trust Torrente as president. (I’ve seen the movie and I laughed, but very even-handed… it is not).
The company worth between $730 billion and $1 trillion in secondary markets hasn’t managed to get people to complete a purchase inside ChatGPT.
Let me repeat that….
The company worth between $730 billion and $1 trillion in secondary markets hasn’t managed to get people to complete a purchase inside ChatGPT.
If the most valuable company in the AI ecosystem can’t solve the most basic use case in online commerce, maybe the distance between valuations and operational reality is greater than markets are willing to admit.
I’m not saying AI doesn’t work. I’m saying the speed at which capital is being committed is incompatible with the speed at which real problems are being solved.
For a while now, I’ve been spending part of my time on AI-driven process re-engineering at SME companies, and the reality is exactly that. Everyone talks about AI, everyone uses AI, but they haven’t rethought their processes from scratch, and the “human in the loop” is excessive.
Result: laughable productivity gains, because nobody acknowledges the current trilemma.
Between autonomy, usefulness and safety, you can only pick two today. I read that from Samuel Gil and I find it spot on.
And that’s it for my human side. If you want the detail on everything else, keep reading.
And don’t forget to share and comment… you help me with distribution that way.
THE BOMBSHELL OF THE WEEK
SoftBank borrows $40 billion to bet on OpenAI’s IPO
On March 27, SoftBank Group closed a $40 billion unsecured bridge loan — the largest uncollateralized corporate credit in technology history — arranged by JPMorgan Chase, Goldman Sachs, Mizuho, SMBC and MUFG. The explicit purpose: to finance its $30 billion commitment in OpenAI’s $110 billion Series round, completed in February.
The timeline:
- February 2026: OpenAI closes a $110 billion round. SoftBank commits $30 billion, raising its total exposure to $64.6 billion.
- March 25: Bloomberg and The Information report that SpaceX is preparing a confidential stock market debut. The market reads it as the IPO window for mega-tech opening up.
- March 27: SoftBank confirms the $40 billion bridge loan maturing in March 2027. SoftBank shares fall — the market buys the thesis but fears the execution.
What the press isn’t telling you. The structure of the loan is the news, not the amount. A 12-month unsecured bridge loan from five tier-1 banks is an instrument designed specifically to be repaid with IPO proceeds. The lending banks are implicitly pricing an OpenAI stock market debut before March 2027. OpenAI’s CFO, Sarah Friar, has publicly guided toward 2027. The loan structure suggests the operational calendar is Q4 2026.
The numbers that matter. SoftBank’s cumulative exposure to OpenAI reaches $64.6 billion, representing roughly 13% of the company. OpenAI’s implied valuation in secondary markets sits between $730 billion and $1 trillion — not the nominal $300 billion of the February round. If OpenAI goes public in 2026, it will be the largest tech IPO in history, surpassing Meta (2012) and Alibaba (2014).
The signal: The loan’s clock is ticking in public. If OpenAI doesn’t go public before March 2027, SoftBank will have to refinance $40 billion unsecured in a rate environment nobody can predict. Masayoshi Son has bet SoftBank’s balance sheet on a binary event with an expiration date.
Sources: TechCrunch, March 27 | Bloomberg, March 27 | Financial Content, March 27 | SoftBank Press Release
POWER MOVES
Atlassian guts its R&D and replaces its CTO with two successors
Atlassian announced 1,600 layoffs — 10% of its workforce. But the detail that matters isn’t the total number, it’s the distribution: more than 900 of those cuts — 56% of the total — come from software R&D. They aren’t cutting costs. They’re hollowing out the product engine.
CTO Rajeev Rajan leaves the company on March 31. He isn’t being replaced by a single CTO but by two promoted lieutenants with narrower mandates: Taroon Mandhana (CTO Teamwork) and Vikram Rao (CTO Enterprise and Chief Trust Officer). The restructuring will cost between $225 and $236 million — roughly $170 million in severance and $60 million in office space.
The pattern — CTO exit plus splitting the CTO mandate plus a majority of cuts in engineering — is a classic indicator of an “AI-replatforming play” where the existing product organization is structurally incompatible with the new direction.
The signal: When a company puts more than half of its cuts in R&D and fractures the CTO’s responsibility, it isn’t optimizing. It’s acknowledging that the team that built the current products isn’t the team that will build the next ones.
Sources: The Next Web | GeekWire | People Matters
Meta lays off 700 people the same day it grants $921 million in stock to each executive
On March 25, Meta eliminated approximately 700 positions across Reality Labs, recruiting, sales and the Facebook division — its second round of layoffs in 2026. The same day, SEC filings revealed that Meta granted stock options worth up to $921 million to each of four executives: CFO Susan Li, CTO Andrew Bosworth, CPO Christopher Cox and COO Javier Oliván. The options only vest if Meta reaches a $9 trillion market cap before March 2031 — roughly 6 times its current $1.5 trillion valuation. Zuckerberg, whose net worth exceeds $200 billion, was not included.
It’s the clearest illustration of the two-speed tech economy: rank-and-file employees eliminated to fund AI infrastructure, senior executives receiving lottery-ticket compensation conditioned on AI-driven hypergrowth. Fortune published a full analysis of Zuckerberg’s leadership noting he has eliminated 25,000 positions since 2022 while the stock has returned more than 600%.
The signal: When the same company lays off 700 people and grants a potential $3.684 billion in stock to four executives on the same day, the message to Wall Street is unambiguous: the company’s value lies in four people, not seven hundred.
Sources: CNBC, March 25 | Winbuzzer, March 27 | The Next Web | Fortune, March 27
T-Mobile runs the classic sequence: OpenAI contract first, layoffs second
T-Mobile confirmed additional layoffs this week described as “aligning our IT organization.” GeekWire sources put the figure at hundreds of positions. The context that completes the story: T-Mobile recently signed a multi-year contract with OpenAI to replace customer service agents with ChatGPT-powered interactions. The sequence is transparent — you sign the AI contract, then you cut the human equivalent.
The signal: Under new CEO Srini Gopalan, T-Mobile is running the playbook Meta established last week: AI as the operational justification for headcount restructuring.
Sources: GeekWire | Fierce Network
MONEY TALKS
SpaceX prepares the largest IPO in history: $1.75 trillion
On March 25, Bloomberg and The Information reported that SpaceX plans to file confidential IPO paperwork with the SEC “as soon as this week.” The target valuation: $1.75 trillion. It would be the most valuable company in the world at the moment of its debut.
The complexity nobody is explaining well: the recent all-stock merger with xAI folds Elon Musk’s artificial intelligence venture into SpaceX’s orbital infrastructure. And the Starship V3 orbital test is targeting an April window — probably designed as a pre-IPO marketing event.
The confidential filing means there won’t be an immediate public prospectus — the clock starts on a 15-day review window before the S-1 becomes mandatorily public.
The signal: If the Starship V3 test succeeds, expect it to serve as the IPO’s catalyst event. If it fails, the filing will be delayed. It’s a literal dependency between a rocket and a stock market debut.
Sources: Bloomberg, March 25 | CNBC, March 25 | SpaceX Stock
Shield AI raises $1.5 billion at a $12.7 billion valuation after Air Force contract
On March 26, defense startup Shield AI closed a $1.5 billion Series G co-led by Advent International and JPMorgan Chase’s Security and Resiliency Initiative. The valuation jumped 140% year over year to $12.7 billion. The catalyst: Shield AI won the contract to provide autonomous pilot software (Hivemind) for Anduril’s Fury fighter, selected by the US Air Force. Blackstone separately committed $500 million in preferred equity plus a $250 million facility. Part of the funds finance the acquisition of Aechelon Technology, a defense simulation company.
The signal: This is the quarter’s largest defense-tech financing event. The Air Force contract is the validation event that unlocked institutional capital at scale. The Fury/Hivemind integration represents a structural shift: the military is buying autonomous warfare capability from startups, not traditional contractors.
Sources: TechCrunch, March 26 | Fortune, March 26 | Washington Technology, March 26
Reflection AI negotiates $2.5 billion at a $25 billion valuation — pre-revenue, backed by Nvidia
On March 25, the Wall Street Journal revealed that Reflection, an AI lab founded by ex-Google DeepMind researchers (Misha Laskin and Ioannis Antonoglou), is negotiating a $2.5 billion round at a $25 billion valuation. Nvidia, which invested roughly $800 million in an earlier round at $8 billion, is an existing investor. JPMorgan Chase is negotiating to come in through its Security and Resiliency Initiative. The declared mission: to build open source LLMs and agentic AI for software development, explicitly positioned as the American answer to DeepSeek. The company has not generated significant revenue.
The signal: A $25 billion valuation for a pre-revenue company indicates investors are pricing geopolitical risk as a premium — the national security framing is doing more valuation work than the financial fundamentals. The Nvidia connection suggests vertical integration between chip infrastructure and open source model labs is the emerging dominant structure.
Sources: US News/WSJ, March 25 | TechFunding News | TechStartups, March 26
PRODUCT SECRETS
OpenAI quietly kills Instant Checkout — the reality behind “ChatGPT as Amazon”
This week OpenAI formally abandoned its “Instant Checkout” product — the in-ChatGPT purchase feature launched in September 2025 to compete with Amazon. The stated reason: low purchase completion rates despite high browsing engagement.
The reality, according to the reports: OpenAI had only integrated a dozen of Shopify’s millions of merchants, product data on the internet is too fragmented for reliable AI-powered checkout, and the company hadn’t built the sales tax collection infrastructure. The new approach redirects users to third-party apps to complete purchases.
The irony: Amazon recently invested $15 billion in OpenAI, turning the competitive dynamic explicitly cooperative. The relationship is contradictory and unstable — they are partners and investors at the same time as competitors in AI for commerce.
The signal: If the most valuable company in the AI ecosystem can’t solve e-commerce’s most basic use case, the distance between valuations and real operational capability deserves more scrutiny than it’s getting.
Sources: TechCrunch, March 24 | CNBC, March 24 | Modern Retail
Apple announces WWDC 2026 and confirms Siri is being rebuilt on Google’s Gemini
On March 23, Apple announced that WWDC 2026 will run June 8-12, with the debut of iOS 27, macOS 27 and the rest of the ecosystem. The real news isn’t the date but the confirmation: Siri will be rebuilt on Google’s Gemini models — the culmination of Apple’s deal with Google to license Gemini and finally deliver the Apple Intelligence features promised for nearly two years and still not shipped.
Apple has been unable to ship its own AI at competitive quality for two years. Licensing a competitor’s model to rescue its most prominent feature is a strategic admission that Apple’s internal AI development has fallen materially behind. This creates a dependency on a competitor (Google) for Apple’s most visible functionality, and adds antitrust and revenue-sharing complexity to Google’s already-contested default search deal.
The signal: When the most valuable company in the world needs to license a rival’s AI to make its voice assistant work, the question is no longer whether Apple is “late to AI” — it’s whether it can rebuild the internal capability or whether the Google dependency is structural.
Sources: 9to5Mac, March 23 | Bloomberg, March 23 | Apple Insider, March 23
REAL NUMBERS
59,959 tech layoffs in 2026 — 689 people a day
As of March 28, 2026, confirmed layoffs in the tech sector total 59,959 people across 198 separate events. The average is 689 jobs eliminated every day since January 1.
Of the 45,363 cuts confirmed through early March, exactly 20.4% (9,238 people) were explicitly attributed to AI and automation by the companies themselves — not by external critics but by the companies making the announcement. The remaining 79.6% likely includes AI-driven decisions not disclosed as such.
Breakdown of the March 21-28 week:
- Epic Games — 1,000 cuts (20% of workforce). Fortnite MAU -28% since 2023. Average playtime: from 29h to 15.4h/month. CEO Sweeney says AI is NOT the reason.
- T-Mobile IT — Hundreds of positions. Exact figure unconfirmed. Coincides with the OpenAI contract.
- Crypto.com — 180 cuts (12% of global workforce).
- Fiserv — 118 positions at its New Jersey hub.
Meta’s two-speed economy in a single day: 700 layoffs + $3.684 billion in stock for four people
Meta’s March 25 numbers:
- Employees laid off: ~700
- Executives with stock grants: 4
- Maximum value per executive: $921M
- Total potential executive value: $3,684M
- Target market cap for vesting: $9 trillion (6x current)
- Deadline to reach the target: March 2031
- Total layoffs since 2022: ~25,000
- Stock return since 2022: +600%
- AI capex 2026: $115-135B
The contrast is arithmetic: assuming an average cost of $250,000 per laid-off employee, the 700 cuts save $175 million a year. The potential stock grant for the four executives is 21 times larger. The market isn’t rewarding efficiency — it’s rewarding the concentration of value in an ever-smaller number of people.
Sources: CNBC, March 25 | Winbuzzer, March 27 | TechCrunch - Epic Games, March 24 | Network World
THE DRAMA
Anthropic vs. the Pentagon: a federal judge blocks the “national security risk” designation
This is the week’s most significant AI governance story. In late February, Defense Secretary Pete Hegseth designated Anthropic a “supply chain risk” — effectively disqualifying the company from all federal contracts. The underlying conflict: the Pentagon demanded “unrestricted access” to the Claude AI models for “all lawful uses.” Anthropic’s counteroffer: use within limits — no fully autonomous weapons, no mass domestic surveillance. The Department of Defense rejected the offer.
On March 26, a federal judge in San Francisco granted Anthropic’s preliminary injunction, ruling that “punishing Anthropic for drawing public attention to the government’s contracting position is classic unlawful retaliation under the First Amendment.”
The judge’s language — describing this as First Amendment retaliation — is extraordinary in the context of a national security designation. It sets a direct precedent on the boundary between AI companies’ usage policies and government access through public contracting.
In parallel: Anthropic’s $1.5 billion settlement in the Bartz v. Anthropic copyright case (training data sourced from LibGen) enters its final phase with a claims deadline of March 30. And Universal, Concord and BMG filed a new $3 billion lawsuit against Anthropic over alleged mass downloading of songbooks.
The signal: A federal judge has just established that an AI company can refuse to give the US government unrestricted access, and that punishing it for doing so publicly violates the First Amendment. Regardless of the appeal, this redraws the playing field for every tech company with government contracts.
Sources: CNBC, March 26 | Washington Post, March 26 | CNN, March 26 | Axios, March 26
Meta found liable for harming minors’ mental health: the “Big Tobacco moment” arrives
On March 25, a Los Angeles jury found Meta (Instagram) and Google (YouTube) liable for the mental health harms of a woman who started using their platforms at age 6, awarding $3 million in compensatory damages and $3 million in punitive damages. Meta bears 70% of the liability.
The amount isn’t the story. The story is that the trial produced a jury verdict on liability, which creates precedent for roughly 1,000 additional pending cases with the same fact pattern. Meta’s exposure across the pending case portfolio could be measured in billions.
In parallel: on March 24, a Santa Fe jury ordered Meta to pay $375 million for violating New Mexico’s consumer protection laws.
The signal: Two verdicts against Meta on two different days, in two different states, under two different legal frameworks. The liability net is closing in from multiple directions simultaneously.
Sources: NPR, March 25 | CNBC, March 25 | CNN, March 26
THE WEEK AHEAD
March 30 — Deadline for the Bartz v. Anthropic copyright settlement. Claims close on the $1.5 billion settlement. Expect media coverage on settlement mechanics and renewed debate about training data and copyright.
Week of March 31 — Rajeev Rajan’s effective exit as Atlassian’s CTO. The LinkedIn activity of departing R&D engineers will be an early indicator of the cultural damage. Watch internal communications about the dual-CTO structure.
Week of March 31 — Confirmation of SpaceX’s confidential IPO filing with the SEC. If Bloomberg and The Information are right, the filing has already happened or lands in the coming days. Once filed, there’s a mandatory 15-day window before the S-1 goes public. This will move every space and AI infrastructure stock.
April (window) — SpaceX’s Starship V3 orbital test. If it succeeds, it will serve as the IPO’s catalyst event. If it fails, the public filing will be delayed. A literal rocket-IPO dependency.
Imminent (days) — Remedies ruling in US v. Google (adtech). Judge Leonie Brinkema must issue her remedies decision before the end of Q1 2026 — that is, this week or next. The DOJ is asking for the forced divestiture of AdX (the ad exchange) and DFP (DoubleClick for Publishers). If she orders a structural remedy, it will be the biggest antitrust breakup of a tech asset since AT&T and will restructure the $600 billion+ programmatic advertising market.
Ongoing — Amazon’s response to OpenAI’s shopping pivot. Amazon invested $15 billion in OpenAI but simultaneously competes in AI search for commerce. Any statement on the scope of the partnership will be a leading indicator.
Ongoing — Meta’s appeal strategy in the mental health and consumer protection verdicts. If the legal strategy becomes public, it will signal how aggressively they intend to fight the Big Tobacco-style liability framework.
Thanks for reading.
Onwards.