On Thursday, April 23, Meta cut 8,000 jobs and Microsoft offered buyouts to roughly 8,750 more. Both cited AI. Neither was losing money. Microsoft’s program is the first voluntary buyout in its 51-year history, eligibility set by a “rule of 70” — age plus tenure must reach seventy, capped at senior director level, with notifications going out May 7. Meta’s chief people officer Janelle Gale told staff in a memo obtained by Bloomberg that the cuts are “to offset the other investments we’re making.” Severance is 16 weeks plus two per year of service; effective date May 20. The company is also closing 6,000 unfilled roles, removing 14,000 headcount positions from its 2026 plan.
The investments Gale referenced are the point. Meta’s 2026 capex guidance is $115–135 billion, up from $72 billion in 2025 — almost entirely data centres, Nvidia GPUs, custom silicon, and the infrastructure underneath the Llama and Superintelligence Labs ecosystem. Microsoft is on track for $145 billion in FY26 capex, with $37.5 billion in Q2 alone going mostly to data centres. The combined Big Four AI infrastructure budget for 2026 — Alphabet, Microsoft, Meta, Amazon — runs near $700 billion, roughly the entire annual revenue of the global SaaS industry.
What’s unusual isn’t the layoffs. Tech has cut over 92,000 jobs in 2026 to date, a 40% increase over the same period last year. Oracle did 30,000 in March. Snap did 1,000. Block did 4,000. Amazon restructured 16,000. What’s unusual is who’s cutting now and why. Meta booked $22.8 billion in net income in Q4 2025 alone. Microsoft’s Q2 capex of $37.5 billion is roughly the annual revenue of Workday. These are not distressed companies. They are not pivoting away from a broken business model. They are doing what Fortune put plainly: “spending on AI, rather than being replaced by technology, is what’s costing Microsoft workers their jobs.”
The framing matters because it’s the opposite of what the productivity story claims. Zuckerberg told the January earnings call that 2026 was “the year that AI starts to dramatically change the way that we work” and that “projects that used to require big teams now be accomplished by a single very talented person.” That reads as productivity displacement. The capex math reads as something else: with $700 billion in infrastructure spending to fund and gross margins on AI workloads still well below traditional software, the most cash-rich vendors in tech are extracting the difference from payroll. The replacement narrative is partly cover for a capital reallocation that was going to happen with or without the productivity gains.
This complicates the hypothesis more than it verifies it. The thesis this blog tracks holds that software loses value because AI can produce it cheaply. Meta and Microsoft are the test case from the other direction. Microsoft’s productivity software is not being commoditized — it is selling AI add-ons at premium prices and reporting record cloud growth. Meta has no SaaS pricing problem because it has no SaaS pricing. The cuts are not defensive against margin compression. They are an offensive bet that infrastructure scale is more valuable, dollar for dollar, than software operating margin. April 7’s Oracle Eats Itself was a vendor doing the same trade under duress, with the stock down 54% from peak and SaaS divisions cut 30%. Microsoft and Meta are doing it from peak profitability, voluntarily, with the equity market shrugging — Microsoft shares fell about 4% on the announcement and recovered.
The mechanism this evidence points at is not the one in the hypothesis. Software-as-license is not visibly losing value at the hyperscalers. What’s losing value is software-as-operating-cost — the headcount that builds, sells, and supports it. The vendors are signalling, with $280 billion in combined 2026 capex from these two companies alone, that they expect the next dollar of margin to come from infrastructure utilisation, not software gross margin. If that bet pays off, software’s share of total tech value compresses by choice rather than by commoditization. If it doesn’t — if AI inference revenue doesn’t compound into the depreciation curves these companies have just locked in — this is one of the largest capex misallocations in corporate history.
What to watch: Microsoft Q3 FY26 earnings on April 30 for Copilot seat count, segment margins, and any change in the capex pace; Meta Q1 2026 earnings April 30 for ad revenue and the first explicit ROI commentary on the $115B+ AI build; whether the next round of cuts hits revenue-attached roles (sales engineering, customer success) or stays in eng/G&A — the latter signals capital reallocation, the former signals demand softness; and whether the $700B Big Four capex figure is revised up or down by the July prints. The hypothesis score remains 62% verified. This week’s evidence neither confirms nor falsifies it. It reframes the question. Software’s value may not collapse. It may just get outspent.
Sources: CNBC — Microsoft plans first-ever voluntary buyout, CNBC — Meta will cut 10% of workforce, CNN — Meta to cut 10% of staff as it pours billions into AI, Fortune — Why Microsoft is doing buyouts, CNBC — 20,000 job cuts at Meta, Microsoft.