Week 19 · May 3 – May 9, 2026

high signal

Named Customers

The agentic AI story shifted from "we can do X" to "named customers report Y at scale" in a single week. ServiceNow's Knowledge 2026 launched Autonomous Workforce across seven enterprise functions with named-customer production data — Docusign at 90 percent autonomous IT ticket resolution, City of Raleigh at 98 percent employee-request deflection, one global energy customer cutting threat containment time 97 percent and saving 1.2 million hours. Microsoft Agent 365 went GA the same week; Anthropic's Code with Claude shipped Outcomes (rubric-graded self-correction in beta), Routines (async PR automation), and Dreams — the first production-preview of autonomous agent memory consolidation. Palantir's Q1 cleared 100 percent US revenue growth for the first time, with US commercial up 133 percent and Rule of 40 hitting 145 percent. AXON Q1 AI revenue grew 700 percent year over year as "nearly all large US law enforcement agencies" now include AI in purchases. Anthropic's deal for SpaceX's Colossus 1 cluster transferred 220 thousand NVIDIA GPUs and 300 megawatts to Anthropic inference — Claude Code rate limits doubled the same week. Challenger Gray's April report named AI as the leading cited reason for layoffs for the second consecutive month, the first back-to-back in the report's history. NIST CAISI published its first independent DeepSeek V4 Pro evaluation: 8 months behind US frontier on the composite score, with the math benchmarks at near-parity. The data turned from anonymous percentages to specific names with specific numbers.

4 predictions updated4 milestones27 companies refreshed

Key Developments

Four major enterprise platforms shipped agentic GA in the same week — and the production deflection numbers came with named customers, not anonymous percentages

ServiceNow's Knowledge 2026 (May 5) launched Autonomous Workforce across seven enterprise functions — IT, HR, finance, legal, security, operations, and procurement — with customer deflection data tied to specific names: Docusign targeting 90 percent autonomous IT ticket resolution, City of Raleigh at 98 percent employee-request deflection, one global energy customer cutting threat containment time 97 percent and saving 1.2 million hours. Security and Risk ACV crossed $1 billion. Microsoft Agent 365 went GA on May 1 as the discovery, governance, and security orchestration layer for AI agents across Microsoft and third-party SaaS. Anthropic's Code with Claude (May 6) shipped Outcomes (multi-step rubric-graded self-correction, public beta), Routines (async automations that file PRs overnight), Multiagent Orchestration, and Dreams — a scheduled memory-consolidation process that reviews past agent sessions, extracts patterns, and restructures memory stores. Salesforce Agentforce remained at $800 million ARR (+169 percent YoY) with 29,000 enterprise deals. Four major enterprise platforms shipping agentic GA simultaneously is the structural confirmation P-021 (40 percent of enterprise apps featuring AI agents by end of 2026) needed.

Confirms P-021 ↗ 40% of enterprise applications will feature task-s...Confirms ↗ Confirms ↗

Counterpoint

The deflection numbers are vendor-curated case studies, not a representative sample. Docusign's 90 percent number is "targeting" — forward language, not measured. The City of Raleigh deployment is municipal IT, where the baseline is old ticketing systems, not the productivity bar that enterprise software's $700 billion 2026 capex is supposed to clear. Anthropic's Outcomes claim — +8.4 percent on docx tasks, +10.1 percent on PowerPoint tasks — is internal self-report on benchmarks Anthropic designed; Vals AI's Finance Agent benchmark at 64.37 percent for Opus 4.7 is third-party but Anthropic still markets itself as "industry leader" without independent ranking confirmation. P-022 (over 40 percent of agentic AI projects will fail by 2027) held at 0.78 because Deloitte's 95 percent pilot-failure finding has not been refuted; what's improving is tooling for the named, well-resourced 5 percent that already had working pilots — not the 95 percent that don't.

Source →

Palantir cleared 100 percent US revenue growth and printed Rule of 40 at 145 — the AIP bootcamp model is the first enterprise AI sales motion to demonstrably compound at scale

Palantir reported Q1 2026 revenue of $1.63 billion (+85 percent YoY) — beating consensus by $90 million. US revenue crossed 100 percent growth for the first time at +104 percent, with US commercial at $595 million (+133 percent). Rule of 40 hit 145 percent, up from 127 last quarter and the highest in company history. Two hundred and six deals larger than $1 million closed in a single quarter — 72 above $5 million, 47 above $10 million. US commercial customer count reached 615 (+42 percent YoY). FY2026 guidance was raised to $7.65 billion (+71 percent YoY), with US commercial guided above $3.224 billion (+120 percent). The AIP bootcamp conversion model — forward deployed engineers spending five days inside a customer's ontology — has now produced 1,300 plus completed bootcamps with the kind of multi-quarter compounding that historical enterprise software companies took years to demonstrate. The Hollowing applies in a different direction here: AIP bootcamps are absorbing the analyst layer of customer consulting work that Accenture and Deloitte used to monopolize.

Confirms P-021 ↗ 40% of enterprise applications will feature task-s...Confirms ↗

Counterpoint

One quarter of 130 percent-plus US commercial growth is not a multi-year trend yet. The base rate for enterprise software companies sustaining 130 percent growth past 2-3 quarters is low — Snowflake decelerated from 174 percent (Q4 FY21) to 50 percent within six quarters; UiPath decelerated from 90 percent to 30 percent within four. At ~60x forward revenue Palantir has zero margin for execution slip; the stock fell 7 percent on the print despite the beat, which is the market's way of saying valuation already discounted this print and then some. Government revenue concentration (~40 percent of total) is genuine political risk in an administration that periodically reviews Palantir contracts. And the ontology moat works against customer choice — Databricks Mosaic and Microsoft Fabric remain alternatives that have not lost ground in the same quarter, suggesting the market is still expanding rather than concentrating.

Source →

Anthropic leased SpaceX's Colossus 1 — 220 thousand NVIDIA GPUs and 300 megawatts moved tenants between former competitors, and Claude Code rate limits doubled the same week

Anthropic announced May 6 that it has signed a deal to use the full compute capacity of SpaceX's Colossus 1 data center in Memphis — 220,000 NVIDIA GPUs (mix of H100, H200, and GB200) across 300 megawatts. The cluster was originally built as an xAI training cluster; Musk acknowledged comfort with the arrangement because xAI had already moved training to Colossus 2. The operational consequence landed immediately: Claude Code 5-hour rate limits doubled for Pro, Max, and Enterprise tiers; peak-hours throttling was removed; Opus API rate limits raised "considerably." Anthropic also expressed interest in partnering with SpaceX on multi-gigawatt orbital compute. Dollar value of the deal was not disclosed, but the cluster represents roughly $4-8 billion in hardware value at current GB200 ASPs — the largest known infrastructure tenant transfer between frontier labs in the cycle. Anthropic's reported ARR trajectory ($30 billion+ as of March, with some sources citing $40 billion) and a potential $40-50 billion summer raise at $850-900 billion valuation provide the demand context for the compute side of the trade.

Confirms P-038 ↗ Approximately 1% of total US economic growth in 20...Confirms P-021 ↗ 40% of enterprise applications will feature task-s...

Counterpoint

The Anthropic-SpaceX arrangement introduces a concentrated governance risk that did not exist when Anthropic's compute was distributed across AWS, GCP, and OCI. Elon Musk's relationship with frontier labs has historically included public disputes (OpenAI lawsuit, xAI competitive posture), and a 300-megawatt single-tenant relationship with a counterparty who runs both a competing AI lab and a competing space-launch business is a different risk profile than a hyperscaler contract. Pentagon also froze Anthropic out of GenAI.mil access on May 1 (Impact Level 6/7 classified networks granted to OpenAI, Google, NVIDIA, Microsoft, AWS, Reflection, SpaceX) — the federal national-security business is now the gap on Anthropic's enterprise positioning, and SpaceX as compute landlord is one half of a relationship where the other half is a Pentagon vendor that Anthropic can't access. The capacity relief is real; the strategic dependence is the trade-off.

Source →

What the Evidence Moved

P-02140% of enterprise applications will feature task-specific AI...

ServiceNow Knowledge 2026 (May 5) launched Autonomous Workforce across 7 enterprise functions with named-customer production data: Docusign 90% autonomous IT ticket resolution, City of Raleigh 98% employee-request deflection, one global energy customer cut threat containment time 97% and saved 1.2M hours; Security & Risk ACV crossed $1B. Microsoft Agent 365 went GA May 1 as the discovery/governance/security orchestration layer. Anthropic Code with Claude (May 6) shipped Outcomes (multi-step rubric-graded self-correction in beta), Routines (async PR-filing automation), and Dreams (autonomous memory consolidation in research preview). Four major enterprise platforms shipping agentic GA simultaneously is the structural confirmation the trajectory needed.

87%→90%▲ +3pp

P-038Approximately 1% of total US economic growth in 2026 stems d...

Aggregate 2026 hyperscaler capex revised up to $725-830B (was ~$520B at year-start). Microsoft $190B FY27 (+130% YoY), Google $180-190B, Amazon ~$200B, Meta $125-145B. AMD Q1 2026 data center revenue $5.8B (+57%) and Q2 guide $11.2B confirm absorption on the supply side; Meta committed up to 6 GW of MI450 deployment. Cerebras IPO books 20x oversubscribed at $26-40B market cap. The capex is being absorbed in real hardware — supply-side validation, not vaporware. Risk now skews toward overshoot (1.5-2pp GDP) rather than undershoot.

70%→75%▲ +5pp

P-020AI-exposed industries will see job losses of ~20,000 per mon...

Challenger Gray April 2026: 88,387 total cuts (+38% from March), 21,490 AI-attributed = 26% of all cuts — AI led stated reasons for the second consecutive month, first time in the report's history. Tech YTD 2026 layoffs 85,411 (+33% vs same period 2025) while non-tech declined 50%, isolating tech as the AI-exposed outlier. BLS April: information sector -13,000 (16th consecutive month of net losses, payrolls at lowest since March 2021), finance -11,000. Goldman's 16K net AI-attributed jobs/month threshold now exceeded by April Challenger Gray attribution alone.

63%→66%▲ +3pp

P-002AI will disrupt 50% of entry-level white-collar jobs over 1-...

Yale Insights / Stanford HAI structural argument lands as independent academic confirmation of three converging mechanisms: (1) software developers aged 22-25 -20% employment since late-2022 peak; (2) software development job postings -53% since late-2022; (3) recent college-grad unemployment ~6% rising 2x faster than general workforce. The mechanism is hiring freeze, not announced layoffs — companies are not replacing junior workers and not hiring new ones. Goldman May 2026 productivity finding ('no meaningful relationship between AI and productivity at economy-wide level') removes the productivity-creates-new-jobs offset path.

65%→68%▲ +3pp

Company Impact

Palantir Technologies

Score change

Q1 2026 BEAT — revenue $1.63B (+85% YoY), US revenue +104% (first time crossing 100%), US commercial $595M (+133%), Rule of 40 hit 145% (highest in company history). 206 deals >$1M in single quarter (72 >$5M, 47 >$10M); 615 US commercial customers (+42%). FY26 guide raised to $7.65B (+71%), US commercial >$3.224B (+120%). md 9→10, dr 4→3; score 8.6→9.1. Approaches ceiling — first 9.1 in the dataset.

Palantir Q1 2026 PR · Investing.com PLTR slides

Key Developments

Four major enterprise platforms shipped agentic GA in the same week — and the production deflection numbers came with named customers, not anonymous percentages

Palantir cleared 100 percent US revenue growth and printed Rule of 40 at 145 — the AIP bootcamp model is the first enterprise AI sales motion to demonstrably compound at scale

Anthropic leased SpaceX's Colossus 1 — 220 thousand NVIDIA GPUs and 300 megawatts moved tenants between former competitors, and Claude Code rate limits doubled the same week

What the Evidence Moved

Company Impact

Palantir Technologies

Axon Enterprise

Advanced Micro Devices

ServiceNow

Datadog

Duke Energy

Figure AI

General Dynamics

Goldman Sachs

Anthropic

Capital One Financial

ASML

Micron Technology

Cadence Design Systems

Equinix

Vertiv Holdings

Vertex Pharmaceuticals

Schneider Electric

Funding Snapshot

Regulatory Calendar

Capability Milestones

Sources