← Back to All Research
Commission a Report
← Back to All Research
Commission a Report
Three days ago, a data leak exposed something Anthropic was not yet ready to announce: a new model tier they call Capybara, housing a model named Claude Mythos — described internally as "a step change" in AI performance, surpassing everything the company has ever released by a significant margin. Cybersecurity stocks plunged on the news. Enterprise software stocks were already nursing $1 trillion in cumulative losses from prior Claude releases. The AI "bubble" narrative — the dominant framing in sell-side research and mainstream financial press — cannot coherently account for any of this.
This report makes a specific, falsifiable argument: the AI bubble thesis is not just wrong, it is directionally inverted. The correct frame is a displacement cycle. One private company — Anthropic — has in approximately 14 months grown from $1B to $20B in annualized revenue, secured 70% of the Fortune 100 as customers, triggered over $1 trillion in equity market losses in the sectors it is replacing, and is now leaking a model so capable that its own internal draft calls it a potential cybersecurity "nightmare." This is not a bubble. This is what a platform shift looks like from the inside — and the NASDAQ 100 contains a dozen companies that will not exist in their current form by 2030.
The four core findings of this report:
On March 25–27, 2026, security researchers Roy Paz (LayerX Security) and Alexandre Pauwels (University of Cambridge) discovered approximately 3,000 unpublished assets in an unsecured, publicly-searchable data store belonging to Anthropic's content management system. Among them was a draft blog post describing a model that had not yet been publicly acknowledged.
Anthropic confirmed through a spokesperson that the model exists and is in early-access trials with a limited set of enterprise customers. The model is "not ready for general release" and is described as expensive to run at current infrastructure costs.
The leaked document establishes a new product hierarchy. Capybara is a tier name — a new classification above Opus, which was until now Anthropic's most powerful tier. Claude Mythos is the first model in that tier. The framing is deliberate: just as Haiku, Sonnet, and Opus represented cost/capability trade-offs within a generation, Capybara represents a generational ceiling-raise. The naming break (from poetic forms to animal names) signals that Anthropic views this as a category departure, not an incremental upgrade.
| Tier | Model Examples | Position |
|---|---|---|
| Haiku | Claude Haiku 4.5 | Fast / cost-efficient |
| Sonnet | Claude Sonnet 4.5, 4.6 | Balanced performance |
| Opus | Claude Opus 4.5, 4.6 | Frontier performance (until now) |
| Capybara | Claude Mythos | New ceiling — above Opus |
The leaked draft characterizes Mythos as showing "dramatically higher scores" compared to Claude Opus 4.6 across three specific domains: software engineering tasks, academic reasoning, and cybersecurity-related benchmarks. No precise numbers were included in the exposed draft, but the document specifically states Mythos is "currently far ahead of any other AI model in cyber capabilities."
That last claim is what sent cybersecurity stocks down sharply on March 27. The draft document — written by Anthropic itself — included a warning that Mythos "presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders." This is extraordinary: a company warning, in its own marketing copy, that its product poses unprecedented risks to the very cybersecurity industry that exists to protect enterprise infrastructure. The market reacted correctly: if the model can attack faster than defenders can respond, human-staffed cybersecurity operations at scale face a structural viability question.
| Model | Release Date | Key Capability |
|---|---|---|
| Claude Opus 4 / Sonnet 4 | May 22, 2025 | First Claude 4 generation |
| Claude Sonnet 4.5 | September 29, 2025 | 77.2% SWE-bench Verified; best-in-class agentic |
| Claude Haiku 4.5 | October 15, 2025 | Cost-efficient agents |
| Claude Opus 4.5 | November 24, 2025 | "Effort parameter" — 76% fewer tokens at parity |
| Claude Opus 4.6 | February 5, 2026 | Release that triggered $1T+ equity selloff |
| Claude Sonnet 4.6 | February 17, 2026 | Current production standard |
| Claude Mythos (Capybara) | Pending — early access | "Step change" above all prior models |
Note the cadence: major releases every 4–8 weeks. The model generation clock is running faster than any prior technology platform in history. By the time enterprise procurement teams complete an evaluation cycle for one model, two generations have shipped.
The AI bubble thesis is not held by fools. It is held by serious analysts at Goldman Sachs, Yale, the World Economic Forum, and Sequoia Capital. The argument has four pillars, and each is worth engaging directly:
Pillar 1: The $600B–$700B spending gap. Sequoia's David Cahn identified a gap between what infrastructure spending requires in revenue to be justified and what the AI ecosystem actually generates. Goldman Sachs stated that $700B in AI investment delivered "basically zero" U.S. GDP growth in 2025. The WEF projected a bubble burst playing out over 18–36 months.
Pillar 2: Enterprise ROI failure. MIT Media Lab found 95% of enterprises report zero return on GenAI investment. The National Bureau of Economic Research found 90% of firms report no measurable productivity impact. These are credible institutions with credible methodology.
Pillar 3: Circular financing and inflated valuations. AI companies receive cloud credits from investors (Microsoft → OpenAI, Google → Anthropic) which inflate apparent revenue. S&P 500 forward P/E at 23x echoes dot-com concentration risk. OpenAI is committed to $1.4 trillion in datacenter spending against $13B in revenue with no profitability roadmap.
Pillar 4: DeepSeek shock as deflation signal. In January 2025, Chinese lab DeepSeek released a model that matched GPT-4o at a fraction of the training cost. Nvidia dropped 17% in a single session. The implication: if frontier AI is cheap to build, the infrastructure moats justifying trillion-dollar valuations are illusory.
The bubble thesis makes a category error at every level of analysis. Here is where each pillar breaks down:
The internet's productivity contribution was unmeasurable by GDP statistics for roughly seven years after commercial deployment. Broadband infrastructure investment in the late 1990s preceded measured productivity gains by a full decade. GDP is a lagging, aggregate instrument that mixes signal with noise across all sectors simultaneously. The appropriate instrument for detecting platform shift is sector-level revenue displacement — and that signal is already deafening. A single Anthropic product launch (Claude Cowork, January 30, 2026) triggered $285 billion in enterprise software equity losses in a single trading session. That is not a GDP story. That is a market pricing a replacement cycle in real time.
The 95% zero-ROI figure from MIT covers all enterprises that "invested" in GenAI — a population that includes companies whose "investment" was buying a ChatGPT Team license and letting two employees experiment with it for a quarter. It does not isolate enterprises with structured deployment, operator-configured Claude agents, and measurable workflow automation. The Deloitte State of AI 2026 study, which does isolate structured enterprise deployers, finds 88% report revenue increases and 87% report cost reductions. The data is not contradictory — it is measuring different populations. The bubble narrative cites MIT; the displacement thesis cites Deloitte. Both are right about their respective populations.
The circular financing critique applies most forcefully to OpenAI, where Microsoft cloud credits historically constituted a significant share of reported revenue. It does not apply to Anthropic with the same force. Anthropic's $14–20B ARR is approximately 80–85% enterprise contracts — Fortune 100 companies paying for production Claude deployments. These are line items that appear in CFO-approved technology budgets. They are not speculative. They are cancellable and they are renewing. The fastest-growing software ARR cohort in history — customers spending $1M+/year — grew from approximately 12 to 500+ in 24 months. That is not circular financing. That is revenue.
The DeepSeek shock is universally cited as evidence against AI infrastructure investment. The correct reading is the opposite. DeepSeek demonstrated that inference cost curves fall faster than even optimists projected. Cheaper inference means higher adoption. Higher adoption means more tasks displaced. The deflationary effect on Nvidia GPU margins is real — but cheaper compute is the mechanism by which AI penetrates every enterprise in the Fortune 1000, not just the ones that can afford $100K/month in API costs. DeepSeek makes the displacement thesis more durable, not less.
| Period | Anthropic Annualized Revenue Run Rate | Growth Since Prior Period |
|---|---|---|
| December 2024 | ~$1B | Baseline |
| July 2025 | ~$4B | +300% (7 months) |
| December 2025 | ~$9B | +125% (5 months) |
| February 2026 (Series G close) | ~$14B | +56% (2 months) |
| March 2026 | ~$19–20B | +43% (1 month) |
Year-over-year growth rate: approximately 1,167%. No B2B software company in history has compounded at this rate from a $1B base. For context, Salesforce took 13 years to reach $10B ARR. Anthropic did it in roughly 14 months from $1B.
Claude Code, Anthropic's coding agent launched publicly in May 2025, reached $2.5B in annualized revenue by February 2026 — nine months from launch. That single product, if spun out, would rank as one of the fastest-growing standalone software businesses ever created. Boris Cherny, Claude Code's creator, stated publicly in February 2026 that the job title "software engineer" could cease to exist by end of 2026, replaced by what he called "builders" who direct AI agents rather than writing code themselves. A Google senior engineer described Claude Code as having "recreated a year's worth of work in an hour."
Perhaps the most analytically significant data point in this entire report is not a revenue number — it is a product decision. Microsoft, which has invested approximately $135 billion in OpenAI, chose Claude as the model powering Microsoft Copilot Cowork, its primary enterprise agent product, integrated into the new Microsoft 365 E7 licensing tier. Microsoft is the most sophisticated enterprise software buyer on earth. It evaluated both models under competitive conditions and chose Claude for the task that matters most: enterprise agentic work. That choice is a stronger signal about Claude's enterprise advantage than any benchmark.
The following analysis maps which NASDAQ 100 members and adjacent large-cap companies face structural displacement from Anthropic's products specifically — not from "AI" generically, but from Claude's deployed capabilities as of Q1 2026. We organize by sector and assess displacement depth on a three-tier scale: Existential (core value proposition directly threatened), Structural (significant revenue streams at risk), and Adaptive (threat is real but company has integration path).
| Company | Market Cap (approx.) | Displacement Tier | Mechanism |
|---|---|---|---|
| ServiceNow (NOW) | ~$160B | Existential | Claude Cowork automates IT service management workflows natively; ServiceNow's primary value is workflow automation — Claude does this in natural language without a workflow builder |
| Intuit (INTU) | ~$170B | Existential | TurboTax and QuickBooks face direct substitution by Claude agents that process financial documents, file taxes, and generate accounting entries via computer use |
| Workday (WDAY) | ~$50B | Structural | HR workflows, payroll processing, and workforce planning are high-value Claude use cases; Workday's UI layer becomes optional when Claude can operate it directly |
| Salesforce (CRM) | ~$230B | Adaptive | Named Claude as core model for Agentforce — pursuing integration, but risks ceding margin to Anthropic as CRM becomes an API layer rather than a product |
| Snowflake (SNOW) | ~$50B | Structural | Data querying, pipeline generation, and analytics interpretation are all Claude use cases; the SQL interface abstraction becomes a commodity |
Market evidence is not theoretical: on January 30, 2026 — the single day Claude Cowork launched — ServiceNow fell 23%, Salesforce fell 22%, Snowflake fell 20%, and Intuit fell 33%. Thomson Reuters fell 31%. Combined, these moves represented a $285B single-session equity destruction event driven by one Anthropic product announcement. That is the market's real-time pricing of displacement risk, not analyst opinion.
| Company / Firm | Revenue Scale | Displacement Tier | Mechanism |
|---|---|---|---|
| Accenture (ACN) | ~$65B revenue | Structural | Cut 11,000 roles while pledging $3B in AI investment; entry-level analytical work is directly displaced; C-suite advisory persists but margin compresses |
| McKinsey (private) | ~$16B revenue | Structural | Headcount fell from 45,000 (2022) to 40,000 (mid-2025); further 10% cuts announced December 2025; ~150 former senior consultants contracted to train AI models to do entry-level consulting work |
| Cognizant (CTSH) | ~$20B revenue | Adaptive | Aligned 350,000 employees with Claude models — attempting transformation but the transformation itself reduces headcount requirements |
| Infosys (INFY) | ~$19B revenue | Structural | Software development services — Anthropic's core use case — is the most displaced segment; coding agent adoption at clients reduces engagement scope |
The consulting displacement follows a specific pattern: the firms are not being replaced at the relationship layer. McKinsey partners still fly to client sites. What is being replaced is the analytical substrate — the hundreds of analyst hours that produce the slide decks those partners present. BCG's "Deckster" drafts client presentations from raw datasets. McKinsey's "Lilli" retrieves and synthesizes prior project work. These tools were built to save headcount costs. They work. The headcount savings are the first-order effect. The second-order effect is that clients observe the efficiency and ask why they are paying Big 3 rates for work a Claude deployment could do internally.
| Company | Market Cap (approx.) | Displacement Tier | Mechanism |
|---|---|---|---|
| Thomson Reuters (TRI) | ~$70B | Existential | Westlaw's legal research product is a curated database of precedent retrieved by Boolean query — Claude Mythos performs this natively in natural language at a fraction of the subscription cost; -31% stock decline on Cowork launch is market confirmation |
| FactSet (FDS) | ~$18B | Adaptive | Integrated as a Claude Cowork plugin (February 2026) — data infrastructure persists but the analytical layer above it commoditizes |
| Verisk Analytics (VRSK) | ~$38B | Structural | Insurance risk analytics, property data synthesis — high-value Claude use cases that reduce reliance on third-party data services |
| MSCI (MSCI) | ~$45B | Structural | Index construction methodology and ESG scoring are vulnerable to AI replication; data licensing less so |
Thomson Reuters is the starkest case. The company built a $70B market cap on the premise that legal research requires expert human curation of an indexed database. Westlaw charges law firms $500–$3,000/month per attorney for access. Claude Mythos, trained on legal texts, performs legal research in natural language for a small fraction of that cost — and Claude Cowork can then draft the brief, review the contracts, and flag the compliance issues in the same session. The value proposition that took Thomson Reuters decades to build erodes in a single product cycle.
This is the most quantitatively documented displacement in Anthropic's own data. The Anthropic Economic Index, updated March 2026, found that approximately 1 in 2 U.S. jobs now has at least a quarter of associated tasks appearing in Claude usage data — up from 1 in 3 a year ago. For software engineering specifically, the evidence is more direct:
NASDAQ 100 companies exposed here include every enterprise software company that employs large software engineering staffs: Microsoft, Google, Meta, Amazon, Salesforce, Oracle, and others. The exposure is not to their products — it is to their cost structure. Companies that currently employ 30,000 software engineers face a world in which 10,000 can produce the same output. That is a labor cost compression story that benefits the hyperscalers in operating leverage but punishes the consultancies and service firms that bill by the engineering hour.
This is the newest and most alarming displacement vector — and the one the Mythos leak made suddenly legible to equity markets. The cybersecurity industry's business model rests on two assumptions: that human expertise is required to detect threats, and that the attack-defense equilibrium can be maintained with sufficient investment in skilled analysts.
The Mythos leak document directly challenged both assumptions. If the model is "currently far ahead of any other AI model in cyber capabilities" and "can exploit vulnerabilities in ways that far outpace the efforts of defenders," then the economic logic of maintaining large human analyst staffs is undermined. Not because AI will replace the analyst immediately — but because the threat surface is expanding faster than human response capacity can scale.
Companies in the blast radius include CrowdStrike, Palo Alto Networks, Fortinet, Zscaler, and others. None of these are NASDAQ 100 members at current weights, but they represent a combined $300B+ in market capitalization built on the labor-intensive model of human threat analysis. The structural compression here will play out over 2–4 years, not overnight — but the Mythos leak was the first hard signal that Anthropic intends to dominate this space rather than merely assist it.
| Sector | Combined Market Cap at Risk (est.) | Displacement Timeline | Current Market Signal |
|---|---|---|---|
| Enterprise Workflow Software | ~$700B | 2–4 years | $285B single-session selloff (Jan 30, 2026) |
| Professional Services / Consulting | ~$400B (public firms) | 3–6 years | Headcount cuts already underway; 10–15% reduction |
| Financial & Legal Data | ~$200B | 2–4 years | Thomson Reuters -31%, FactSet adapting |
| Software Engineering (labor) | N/A (labor, not equity) | 1–3 years | Claude Code $2.5B ARR; 1 in 2 jobs in usage data |
| Cybersecurity (analyst-layer) | ~$300B | 3–6 years | Sharp selloffs post-Mythos leak (March 27, 2026) |
| Total addressable equity displacement | ~$1.6T | >$1T already realized in selloffs |
The displacement analysis above is not speculative — it is grounded in deployed capabilities. Understanding how Claude does this requires understanding the agentic stack Anthropic has built.
On March 23–24, 2026, Anthropic shipped Computer Use capabilities to Claude in preview. Claude can now open files, operate browsers, navigate macOS interfaces, fill spreadsheets, run developer tools, and complete multi-step computer tasks autonomously. CNBC described the user experience as follows: "Users can message Claude a task from their phone, and the AI agent will then complete that task by opening apps on your computer." This is not a chatbot that answers questions. This is a worker that operates software.
The Claude Agent SDK, released publicly in 2025, enables developers and enterprises to deploy multi-agent systems where orchestrator agents spawn, coordinate, and synthesize results from specialist subagents running in parallel. A production-tested configuration — 2–5 specialist agents, 5–6 tasks each — can simultaneously handle what previously required a team of analysts. Anthropic discovered (or admitted) that a fully-implemented multi-agent system had been hidden in Claude Code's compiled binary for months before official activation in February 2026. This suggests the multi-agent architecture was production-ready before it was policy-approved.
Claude Cowork, launched January 30, 2026, is the product that triggered the $285B single-session selloff. It enables non-technical employees to delegate multi-step knowledge work — research, analysis, document synthesis, communication drafting — via natural language. Plugins added February 24, 2026 connect Cowork to Google Drive, Gmail, DocuSign, FactSet, and more. Microsoft chose Claude to power Copilot Cowork, integrated into the new M365 E7 enterprise tier. This means Microsoft is distributing Claude's capabilities to its entire enterprise customer base — a channel advantage that no prior AI lab has had.
Anthropic has built a three-tier trust hierarchy (Anthropic → Operators → Users) that allows enterprise customers to deploy Claude with the same compliance posture as their existing enterprise software. VPC isolation via AWS Bedrock and Google Vertex AI, SAML 2.0/OIDC SSO, Zero Data Retention options, and the Auto Mode safety classifier (March 24, 2026) that evaluates each agent action before execution — these are not features. They are the prerequisites for regulated industry adoption. Healthcare, financial services, and legal are the remaining large holdouts. Their barriers are eroding.
Bloomberg and The Information reported on March 27, 2026 that Anthropic is targeting an IPO as early as October 2026, aiming to raise over $60 billion. Goldman Sachs, JPMorgan, and Morgan Stanley are in preliminary discussions for lead roles. At its current $380B private valuation, Anthropic would rank among the 20 most valuable publicly traded companies globally upon listing.
| Funding Round | Date | Amount Raised | Post-Money Valuation |
|---|---|---|---|
| Series E | March 2025 | $3.5B | $61.5B |
| Series F | September 2, 2025 | $13B | $183B |
| Series G | February 12, 2026 | $30B | $380B |
| IPO (target) | ~October 2026 | ~$60B raise | TBD |
At $380B private valuation, Anthropic is already larger than many NASDAQ 100 members it is displacing:
This is not a bubble dynamic. Bubbles involve a company trading at a premium to unrelated fundamentals. This is a company trading at a premium to the companies it is replacing — while generating real revenue, holding real enterprise contracts, and posting real growth rates that no software company in history has matched.
European Business Magazine described Anthropic as having "walked into $5 trillion of market cap" when it launched Claude Cowork. That number represents the combined addressable market of enterprise software, professional services, financial data, and legal information sectors in Anthropic's direct displacement path. Anthropic's current $380B valuation against a $5T displacement target represents a 13x multiple on potential capture. The question is not whether the displacement is real — it is happening and the equity markets have priced $1T+ of it already. The question is what share of that $5T Anthropic captures before competition, regulation, or capability ceilings intervene.
This report argues against the bubble thesis but not in favor of naive AI triumphalism. The following risks are material:
The Mythos leak's cybersecurity claims will attract regulatory attention. If policymakers determine that a single private company's AI capabilities pose systemic risk to digital infrastructure defense, capability restrictions, mandatory licensing, or national security review could impose constraints on deployment speed. The EU AI Act's high-risk category framework is already structurally primed for this outcome.
Healthcare, financial services, and government procurement operate on 12–36 month evaluation cycles. Even if Claude Mythos is as capable as leaked documents suggest, the largest remaining revenue pools are behind compliance and liability walls that take years to negotiate. The revenue growth rate of 1,167% will almost certainly compress as the easiest enterprise segments saturate.
DeepSeek demonstrated that capability moats compress faster than valuations adjust. Google, Meta, and OpenAI are not standing still. If a rival reaches Mythos-tier capability within 12 months — plausible given current lab trajectories — pricing power erodes. Anthropic's 80–85% enterprise revenue share is its structural defense, but it is not a permanent moat.
Google's cloud credits contribution to Anthropic's compute costs is real, even if smaller than OpenAI's equivalent arrangement. At IPO, public market investors will scrutinize the revenue quality question. Any revenue that effectively nets against cloud costs rather than representing incremental economic activity will face discount.
Leaked internal documents are marketing drafts, not technical audits. "Dramatically higher scores" on Anthropic's internal benchmarks may not replicate in independent third-party evaluation. The Mythos performance claims should be treated as directionally credible but not as established fact until independent evaluation data exists.
PRZC Research does not make individual security recommendations in this format. The following represents sector-level analytical framing for investment committee consideration.
Companies whose primary value proposition is knowledge work synthesis — legal research, financial data, workflow automation, IT service management — face structural margin compression over a 2–4 year horizon. The compression is not binary (zero vs. full displacement) but gradient: pricing power erodes before revenue does, revenue erodes before market cap adjusts fully, and the adjustment is lumpy rather than smooth (as the January 30 session demonstrated).
Companies that have structurally positioned themselves as orchestration layers above Claude — Salesforce via Agentforce, Microsoft via Copilot Cowork, FactSet via plugin — face a different risk profile. Their risk is margin capture: they retain customers but cede the economic rent to Anthropic. Revenue holds; P/E multiples compress as software premium erodes toward service-business multiples.
Nvidia, AWS, GCP, and Azure benefit from the displacement cycle regardless of which AI company wins. Inference demand is inelastic to model brand. As Mythos and its successors deploy to enterprise customers, compute consumption scales. Infrastructure providers are not in the displacement basket — they are in the tax-on-displacement basket.
For institutional investors with access to secondary market Anthropic shares or Series G allocation rights, the October 2026 IPO timeline creates a 6-month liquidity window. The core risk is whether public market pricing at IPO can be sustained against a $380B private entry. At $20B ARR growing at even 50% (a dramatic deceleration from current rates), a 20x ARR multiple — consistent with high-growth SaaS comps — implies a $600B valuation at IPO. That represents meaningful upside from current secondary pricing. The downside scenario is regulatory intervention or a major competitive capability release before the IPO closes.
The AI bubble narrative will persist. It is a coherent story, it is supported by credible institutions, and it is useful as a check on naive capital allocation. But it is wrong about what is happening at Anthropic specifically, and it will be proven wrong by the weight of accumulated evidence as the next 24 months unfold.
What Claude Mythos represents — if the leaked characterization is accurate — is the crossing of a threshold: a model so capable in software engineering, reasoning, and cybersecurity that its own creators feel compelled to warn about the risks in the document announcing it. A model tier above Opus, above everything that triggered a trillion dollars in equity destruction, arriving while Anthropic is still growing revenue at four digits annually.
The NASDAQ 100 was built over 40 years on the premise that knowledge work — legal research, financial analysis, workflow automation, software engineering, consulting deliverables — requires human expertise organized in firms. Anthropic is not disrupting that premise. It is replacing it, contract by contract, enterprise by enterprise, at a revenue growth rate that no software company in history has achieved.
The bubble thesis asks: is AI overvalued relative to its economic impact? The anti-bubble thesis asks: is the NASDAQ 100 undervalued relative to its displacement risk? The market, in three key sessions in January and February and March 2026, has been answering the second question. The answer is not complete. But the direction is clear.
"We are watching, in real time, a redistribution of value from the companies that organized human knowledge work to the company that learned to automate it."
— PRZC Research, March 2026
| Metric | Value | Date |
|---|---|---|
| Anthropic valuation (Series G) | $380B | February 2026 |
| Anthropic ARR | ~$19–20B | March 2026 |
| ARR growth YoY | ~1,167% | Dec 2024 → March 2026 |
| Fortune 100 customers | 70% | Q1 2026 |
| Claude Code ARR | $2.5B | February 2026 (9 months old) |
| Customers spending $1M+/year | 500+ (from ~12 two years prior) | Q1 2026 |
| Claude Cowork single-session selloff | $285B | January 30, 2026 |
| Cumulative equity losses triggered by Claude releases | >$1T | Jan–March 2026 |
| U.S. jobs with 25%+ tasks in Claude usage data | ~1 in 2 | March 2026 |
| Enterprises reporting AI revenue increase (Deloitte) | 88% | 2026 |
| Goldman Sachs: AI GDP impact in 2025 | "Basically zero" at macro level | February 2026 |
| Goldman Sachs: AI productivity in targeted use cases | 30% gains | March 2026 |
| Anthropic IPO target | ~October 2026, ~$60B raise | March 27, 2026 |
| Claude Mythos leak date | March 25–27, 2026 | March 2026 |
Disclaimer: This report is produced by PRZC Research for informational and analytical purposes only. It does not constitute investment advice, a solicitation, or a recommendation to buy or sell any security. All figures cited from third-party sources are attributed to those sources and have not been independently verified by PRZC Research. Past market reactions are not predictive of future outcomes. Readers should conduct their own due diligence before making any investment decision.