Tech Digest – February 27, 2026
Workforce & Market Repricing
Block Cuts Half Its Workforce to “Move Faster With AI” — Stock Jumps 24%
Block laid off over 4,000 employees — roughly half its workforce — with CEO Jack Dorsey framing the cuts as a move to “smaller teams using AI.” The market responded with a 24% after-hours spike. Dorsey confirmed the company is now targeting $2M+ gross profit per person, four times its pre-COVID efficiency. He attributed the over-hiring to a dual company structure (Square and Cash App) and complexity from lending, banking, and BNPL products — all of which have since been consolidated.
Note: The market didn’t just tolerate this — it rewarded it instantly and aggressively. That 24% signal travels fast through every boardroom with a cost structure to defend.
Sources: CNBC, Jack Dorsey (X)
Software Stocks Lose $1.6 Trillion as Investors Reprice Legacy SaaS
Components of the State Street software ETF have shed a combined $1.6 trillion in market capitalization this year. The sell-off reflects a broad investor repricing of traditional SaaS businesses against AI-native alternatives that promise to deliver the same functionality at a fraction of the cost and headcount.
Note: If your institution is mid-cycle on a multi-year SaaS contract, this repricing is relevant now. The vendors you’re paying may look very different — or may not exist — by the time renewal comes around.
Sources: Wall Street Journal
Burger King Deploys Voice AI That Scores Employees on “Friendliness”
Burger King is rolling out “Patty,” a headset-mounted voice AI that assists kitchen staff with meal preparation and simultaneously evaluates employee behavior, including scoring workers on “friendliness” in real time.
Note: Real-time AI performance scoring in fast food today, frontline public services tomorrow. Any organization with citizen-facing staff should be thinking about where the line between “assistance” and “surveillance” falls — before the technology decides for them.
Sources: The Verge
Infrastructure & Capital Flows
AI Infrastructure Spending Accelerates Across Every Layer
CoreWeave’s Q4 revenue grew 110% year over year. Dell forecasts AI server revenue to double in fiscal 2027. Meta has reportedly signed a multi-billion-dollar deal to rent Google’s TPUs, diversifying its chip supply away from NVIDIA. Japan’s Rapidus secured $1.7 billion in government funding to reach 2-nm mass production by 2028. And Eli Lilly partnered with NVIDIA to launch LillyPod — the world’s first DGX SuperPOD with B300 systems, packing 1,016 Blackwell Ultra GPUs and over 9,000 petaFLOPs of compute dedicated to drug discovery.
Note: Five data points, one direction: every major player is doubling down on AI compute simultaneously. The infrastructure layer is scaling faster than the institutions that will eventually depend on it. Procurement timelines set today are based on a supply landscape that won’t exist when the contracts land.
Sources: Reuters (CoreWeave), Reuters (Dell), The Information, Bloomberg, NVIDIA Blog
Smartphone Shipments Set for Biggest-Ever Decline as AI Cannibalizes Consumer Hardware
IDC projects smartphone shipments will fall 12.9% this year to a decade low — the largest decline the industry has ever recorded. The driver: AI-fueled demand for memory chips is pushing prices up, making consumer devices more expensive while capital and components flow toward data centers instead.
Note: The device that defined digital public services for fifteen years is fading. If your digital strategy still centers on “mobile-first,” it may be time to ask what comes after the phone.
Sources: Reuters / IDC
AI Governance Goes Geopolitical
Anthropic Refuses Weapons Work — Department of War Official Attacks Claude’s Values
Anthropic published a formal statement refusing to let its models power mass surveillance or autonomous weapons for the Department of War. In response, Under Secretary of War Emil Michael publicly attacked Claude’s constitutional principles for requiring sensitivity to non-Western cultural traditions, framing the AI’s value system as a national security liability. The exchange previews a new kind of regulatory conflict: system prompts and model constitutions as geopolitical battlegrounds.
Note: The values baked into an AI model are now a matter of public policy debate. For institutions evaluating AI vendors, this isn’t abstract — your choice of model is also a choice about whose values run your operations.
Sources: Anthropic, Under Secretary Emil Michael (X)
Norway’s $2 Trillion Sovereign Wealth Fund Uses Claude for Ethical Screening
NBIM, which manages Norway’s $2 trillion sovereign wealth fund — the world’s largest — now uses Anthropic’s Claude to screen investments for reputational and ethical risk. The fund is outsourcing moral judgment calls to AI at sovereign scale, applying it across its portfolio of holdings in over 8,700 companies.
Note: A sovereign institution trusting AI with ethical risk assessment sets a precedent that reaches well beyond finance. If the world’s largest fund thinks AI judgment is reliable enough for reputational screening, the pressure on smaller institutions to follow will be significant.
Sources: CNBC
The Architecture Is Compressing
Smaller, Faster, Cheaper: Three Advances Collapse AI’s Cost Curve
Three research results dropped in a single day, all pointing in the same direction. Researchers demonstrated that foundation models can be self-distilled into multi-token predictors that decode 3x faster at under 5% accuracy loss. Sakana showed it can compile entire documents directly into model weights via hypernetworks — giving language models durable memory without inflating context windows. And LM Provers released QED-Nano, a compact 4-billion-parameter model that writes Olympiad-level mathematical proofs approaching frontier performance.
Note: Last year’s frontier capability is becoming this year’s commodity. For anyone budgeting AI projects based on current pricing, the cost floor is dropping faster than most procurement cycles can track.
Sources: arXiv, Sakana AI, LM Provers / HuggingFace
Claude Gets a Work Calendar — AI Agents Move From On-Demand to Autonomous
Anthropic introduced scheduled tasks in Claude Cowork, enabling the AI to complete recurring jobs at set times — morning briefings, weekly spreadsheet updates, Friday team presentations. Separately, Amplifying is pointing Claude Code at thousands of GitHub repositories to extract what the model considers current best practices, letting AI systematically audit and learn from the code it is increasingly being used to write.
Note: The shift from “tool you invoke” to “colleague with a calendar” is a quiet but structural change. When the AI runs tasks on a schedule without being asked, it stops being a utility and starts being a workflow participant.
Sources: Claude (X), Amplifying
Machines at the Front Desk
Humanoid Robots Greet Patients at Chinese Hospital
Two AGIBOT A2 humanoid robots — named Zhen Zhen and Ru Ru — are now stationed in the lobby of Changzhou First People’s Hospital in China, greeting patients with handshakes and handling registration queries and wayfinding. AGIBOT has shipped over 5,000 A2 units globally and holds certifications for China, the US, and EU markets. The robots are already deployed across eight commercial sectors including reception, hospitality, and security.
Note: Humanoid robots handling patient intake at a public hospital — not as a demo, but as deployed infrastructure. The A2 is EU-certified and priced for scale. This is no longer a question of “if” for European public services, but “when” and “under what procurement framework.”
Sources: CyberRobo (X), AGIBOT, PR Newswire
Iris-Scanning Orbs Now Verify Shoppers at Gap Stores
Sam Altman’s World ID project has installed biometric verification orbs at a Gap store in San Francisco, scanning shoppers’ irises to confirm they are human. The deployment marks the entry of persistent biometric identity verification into everyday retail — a scenario that, as many have noted, was depicted in Minority Report as a vision of 2054.
Sources: Wall Street Journal
