By Thorsten Meyer — May 2026
On April 29, an analyst on Meta’s Q1 2026 earnings call asked Mark Zuckerberg about signs of return on the company’s AI investment. Meta is spending $125-$145 billion on AI infrastructure in 2026 alone — more than 2024 and 2025 combined, on AI, in one year. Zuckerberg’s answer was “that’s a very technical question.”
The stock dropped 6% in after-hours trading. Meta still posted $56.3 billion in revenue, up 33% year over year. Operating income rose 30%. Profits grew 61%. By any pre-2025 measure, this would have been a triumphal quarter. By the standard the AI capex cycle has now imposed, “very technical question” is the moment Wall Street’s patience inflected.
This dispatch is about what Q1 2026 earnings season — JPMorgan, Goldman Sachs, Meta, Alphabet, Bank of America, Microsoft — actually disclosed about AI ROI, what the disclosure language tells you about what management actually knows, and the four-quarter pattern that emerged once the data was assembled. The pattern is not hostile to AI. It is hostile to the version of AI being sold on earnings calls.
The gap between what executives are claiming on earnings calls and what is showing up in the financial statements is now four quarters wide. The dispatch on the productivity gap called this from the macro side; the dispatch on the AI-washed layoff narrative called it from the labor side. Q1 2026 is the quarter the gap entered the financial statements directly.
Three numbers anchor everything else. Goldman Sachs analyst research found that 90% of companies discussing AI on earnings calls use qualitative language rather than quantitative metrics. The NBER survey of 6,000 executives across four countries found 90% reporting zero AI productivity impact over three years. BCG’s own CEO survey found 80% more optimistic about AI ROI than a year ago. Three numbers, three populations, three directions. The stock-price reaction in Q1 was the market starting to weight the first two over the third.
The earnings call gap.
Q1 2026 was the quarter the market started pricing in disclosure quality.
On April 29 an analyst asked Mark Zuckerberg about ROI on Meta’s $145 billion of AI capex. He called it “a very technical question.” The stock dropped 6% — on a quarter with revenue up 33% and profits up 61%. The market spent two years tolerating qualitative AI language. Q1 2026 is when it stopped.
April 29, 2026. Six percent.
An analyst asks about visible evidence that $145B of capex is producing proportional value. The CEO answers in venture-stage uncertainty language. The stock drops six percent on a quarter with revenue up 33%. The market just told public-company AI capex it has to be auditable now.
That’s a very technical question. I don’t think we have a very precise plan for exactly how each product is going to scale month over month, or anything like that, but I think we have a sense of the shape of where these things need to be.
AI for DevOps Engineers: Master AIOps, Kubernetes Automation, and Cloud Infrastructure Monitoring
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Same quarter. Different disclosure. Different stock reaction.
The market is now able to distinguish — and is starting to weight — disclosure quality. Companies that produced specific AI-attributable revenue or cost numbers were rewarded. Companies that produced qualitative statements were punished. The same quarter. Different disclosure quality. Different stock reaction.
Artificial Intelligence for HR: Use AI to Support and Develop a Successful Workforce
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What execs say on calls. What execs see in their orgs.
Two surveys. Two populations. Two findings — both at 90%. Together they describe the gap between the AI narrative on earnings calls and the AI experience inside the operating businesses underneath them.
Companies use qualitative language about AI on earnings calls.
The 10% using quantitative language are concentrated in: hyperscalers reporting cloud revenue, software companies with AI-revenue-attributable products, and a small handful of regulated-industry leaders who made disclosure a strategic differentiator.
Executives report zero AI productivity impact over three years.
n=6,000 across four countries. Three years of cumulative deployment, training, change management, and capex — with no measurable productivity impact at the executive’s own company. Lines up with Deloitte: 37% “surface level,” only 25% “transformative.”
Output to Outcome: An Operating Model for the Age of AI
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The JPMorgan format, scaled appropriately. Five elements.
The disclosure that wins through 2026 is a five-element format — small enough to fit in two paragraphs of prepared remarks, complete enough for analysts to model. Whatever the company decides, decide it before the IR team improvises on the call.
The disclosure that survives Q2 2026.
The CFO who publishes this format in Q2 2026 will be early. The CFO who publishes it in Q4 2026 will be on time. The CFO who has not published it by Q2 2027 will be experiencing the qualitative-language discount as a structural feature of the company’s valuation.
Total tech budget
The denominator — total spend within which AI sits
AI-specific incremental
The portion of incremental spend attributable to AI
AI value · projected
Annual AI-attributable business value · disclosed
Use-case count
With qualitative shape of where value concentrates
YoY comparison
Versus a prior baseline so analysts can model
The earnings call gap is now four quarters wide. Q1 2026 was the quarter the market started pricing it in. The CFOs who publish a number in Q2 will be early. The ones who don’t by Q2 2027 will be discounted structurally.
Lonestar Tracking Oyster3 Hidden GPS Tracker – 4G & 5G Waterproof Vehicle & Trailer Tracker – Anti-Theft Car, Long Battery Life, Multi-GNSS Precision, Geofence & Recovery Mode Real-Time Asset Tracking
HIDDEN GPS TRACKER WITH LONG BATTERY LIFE: The Oyster3 lasts up to 10 years with once daily alerts…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Four assignments. By role.
Decide your Q2 disclosure posture by mid-June.
The benchmark is JPMorgan’s five-element framework: tech budget, AI-specific incremental, AI-attributable business value (projected), use-case count, year-over-year comparison. Whatever you decide, decide it before the IR team improvises on the call.
Run the Goldman 90% screen on your own four prior calls.
If you’re in the qualitative-language 90%, you have one quarter to build the measurement infrastructure — workflow telemetry, productivity baselines, AI-attributable revenue/cost categorization — that lets you exit it.
Re-screen your portfolio for disclosure quality.
Pull each holding’s Q1 2026 transcript. Count quantitative versus qualitative AI mentions. Above 50% quantitative = positioned for the inflection. Below 20% = forward exposure to the qualitative-language discount.
Re-pitch around auditability, not transformation.
Customers who can publish JPMorgan-style disclosures will pay a premium. Customers who cannot are about to enter a price war on commodity capabilities. The product-marketing claim that wins in 2026–2027 is “auditable,” not “transformational.”
Executive Summary · Q1 2026 AI ROI Reality Check
| Source | What was disclosed | Disclosure type |
|---|---|---|
| Meta | $125-145B capex 2026 (raised from $115-135B); revenue $56.3B (+33%); profits $26.8B (+61% incl. $8B tax benefit). Zuckerberg on AI ROI: “very technical question.” | Capex hard, ROI qualitative. Stock -6% after-hours. |
| Alphabet | Q1 cloud revenue $20B+ (first time), grew 63%. AI products up 800% YoY. $460B+ backlog. New customer acquisition doubled. | Quantitative. Stock up after-hours. |
| JPMorgan | Tech budget $19.8B (+10% YoY), ~$1.2B incremental for AI/modernization. Public projection: $1.5-$2B annual AI-generated business value. 400+ production AI use cases. | Hard dollar disclosure. Unusual for the sector. |
| Goldman Sachs | 48% surge in IB fees, $17.55 EPS. Internal: 3-4× productivity gains from autonomous coding agents. No public dollar figure. | Productivity ratio, no $. |
| Bank of America | Healthy consumer spending, trimmed full-year NII guidance. Erica AI: 3 billion cumulative interactions, embedded across 95% of employee base. | Usage metrics, no productivity $. |
| Lloyds Bank (Q4 2025 baseline) | £50M documented AI value 2025; targeting £100M 2026. | Rare before/after dataset. |
| Danske Bank | Q1 2026: ROE 13.3% in 2025 → above 14.5% by 2028. CEO: “productivity gains from technology and AI more than offsetting inflation.” | Bundled with cost takeouts. AI portion not isolated. |
| NBER survey (n=6,000 executives, 4 countries) | 90% report zero AI productivity impact over three years. | The denominator the optimistic surveys leave out. |
| Goldman Research | 90% of companies use qualitative language about AI on earnings calls. | The disclosure quality signal. |
| BCG CEO survey | 80% of CEOs more optimistic about AI ROI than a year ago. 65% say accelerating AI is a top-3 priority. | Self-report from owners of the budget. |
The pattern in one sentence: companies that disclose hard numbers (Alphabet, JPMorgan) are seeing flows; companies that defer to “very technical question” (Meta) are seeing stock-price punishment. The market just started pricing in the disclosure language.
1. The “Very Technical Question” Moment
Zuckerberg’s answer on April 29 deserves to be read in full context. The question was not abstract. The analyst was asking about visible evidence that $145 billion of unprecedented spending was producing proportional value. Zuckerberg’s full response: “That’s a very technical question. I don’t think we have a very precise plan for exactly how each product is going to scale month over month, or anything like that, but I think we have a sense of the shape of where these things need to be.”
A “sense of the shape of where these things need to be” is the language venture capital uses about a Series A startup, not the language of a public company spending $145 billion of shareholder capital in a single year. The market understood the substitution. The 6% drop in after-hours trading was the recognition that Meta has imposed a venture-stage uncertainty distribution onto a public-company capital allocation framework, and the spread between those two frames is widening.
Compare Alphabet’s same-day disclosure. Sundar Pichai: cloud revenue grew 63% to $20B+ in Q1. AI products built on Gemini grew “nearly 800% year-over-year.” New customer acquisition “doubled.” $100M-$1B deal count “doubled year-on-year.” Backlog “nearly doubled quarter-on-quarter to over $460 billion.” Customers outpaced their initial commitments by 45%. The numbers are specific. They are auditable. They are quantitative. Alphabet’s stock went up after the same earnings cycle that punished Meta’s stock.
The lesson is not that Alphabet is winning AI and Meta is losing AI. The lesson is that the market is now able to distinguish — and is starting to weight — disclosure quality. Companies that can produce specific AI-attributable revenue or cost numbers are being rewarded. Companies that produce qualitative statements about “leading labs” and “the shape of where things need to be” are being punished. The market spent two years giving every AI investment story the benefit of the doubt. In April 2026, that two-year window closed.
2. The Goldman 90% Finding
Goldman Sachs equity research published a screen of all S&P 500 earnings call transcripts from Q1 2025 through Q4 2025 examining how companies discuss AI ROI. The screen distinguished quantitative language (specific dollar values, basis points, percentage productivity gains, headcount avoided, revenue per employee, cost-to-income ratio improvements directly attributed to AI) from qualitative language (“transformative,” “reimagining,” “leveraging,” “AI-powered,” “next generation”). The result: across the population, 90% of S&P 500 companies discussing AI on earnings calls use qualitative language. 10% use quantitative language.
The 10% that use quantitative language are concentrated in three groups. Hyperscalers reporting cloud-revenue growth (Microsoft Azure, Alphabet GCP, Amazon AWS Bedrock) where AI is the line of business. Software companies with directly AI-revenue-attributable products (NVIDIA chip sales, Palantir, ServiceNow). And a small handful of regulated-industry leaders (JPMorgan with the $1.5-$2B disclosure, Lloyds with the £50M-to-£100M target) who have made the disclosure a strategic differentiator. The other 90% are not lying. They are deferring. They are saying things that will be true if AI works without committing to numbers that will be falsifiable if AI does not.
This is the disclosure-language version of the productivity gap. It exists because the underlying ROI work is, in most companies, not yet legible at the financial-statement level. Productivity gains accrue to individual workflows, individual teams, individual products. They do not yet roll up cleanly into the consolidated income statement. The CFO knows this. The IR team knows this. The qualitative language is the rational response to a measurement problem that has not yet been solved.
The market is no longer accepting that response. Q1 2026 is the quarter the qualitative-language tax started pricing in.
3. The NBER 90% Finding
While Goldman’s finding describes what executives say on earnings calls, NBER’s 6,000-executive survey describes what executives privately observe in their own organizations. The result is the more uncomfortable of the two. 90% of executives reported zero AI productivity impact over three years.
Three years. Not three months. Not three quarters. Three years of cumulative deployment, training, change management, and capex, with no measurable productivity impact at the executive’s own company.
The survey is large enough that it cannot be dismissed as methodological noise. n=6,000, four countries, conducted by NBER. The result lines up with Deloitte’s State of AI in the Enterprise 2026, which found that 37% of leaders are “using AI at surface level with little or no change to existing processes” and only 25% report “transformative effect.” The 75% middle is the qualitative-language population — companies using AI somewhere, sometimes, with no consolidated measurable impact yet.
Bill Briggs, Deloitte CTO, captured the underlying constraint: “Companies are allocating 93% of their AI budgets to technology and only 7% to the people expected to use it. This incrementalism is a hard trap to get out of. Organizations are buying ingredients without learning the recipe.”
That is the sentence that explains the 90/90 gap. Companies are spending capital on infrastructure they do not yet know how to deploy through human workflows. The infrastructure spend shows up immediately in capex line items. The productivity return depends on workflow redesign, change management, and skills development that has not been funded at remotely the same scale. The capex hits the income statement as depreciation and operating costs in 2026. The productivity gain — if it materializes — will hit the income statement in 2028 at the earliest.
Q1 2026 is the period in which the capex started compounding into expense without the offsetting productivity yet showing up. That is the actual mechanism behind Meta’s stock reaction. The market is now pricing in the gap between when the capex shows up and when the productivity has to.
4. The Banks Disclosure Spectrum
Financial services is the sector where AI ROI disclosure has gone furthest, and it produces a useful spectrum.
JPMorgan: full hard disclosure. $19.8B 2026 tech budget. ~$1.2B incremental year-over-year specifically for AI and modernization. Public projection of $1.5-$2B in annual AI-generated business value. 400+ production AI use cases on the bank’s enterprise ML platform, processing $10 trillion in daily transactions. This is unusual transparency for the sector and was clearly engineered as a strategic positioning move. JPMorgan’s disclosure is auditable: the projected $1.5-$2B is roughly 0.85-1.13% of trailing twelve-month net income. It is a number small enough to be plausible and large enough to matter. It is the gold standard the rest of the sector is being compared to.
Goldman: productivity-ratio disclosure, no dollar figure. The bank reports 3-4× productivity gains from autonomous coding agents internally, but does not publicly translate that into headcount avoided or net income contribution. The 48% surge in investment banking fees and $17.55 EPS are the headline numbers. The AI productivity is a private metric tied to engineering compensation decisions. Goldman gets the “AI is working” narrative without committing to a number that would let analysts model the contribution.
Bank of America: usage-metric disclosure, no productivity figure. Erica’s 3 billion cumulative interactions and 95% employee embedding are usage statistics, not financial impact statistics. Bank of America trimmed full-year NII guidance — meaning the macro environment is more important to the income statement than any AI impact would be. The AI disclosures are framing, not modeling.
Lloyds: rare before-after dataset. £50M documented AI value 2025, £100M target 2026. This is the format that Goldman’s research finding was implicitly asking for: a baseline year, a target year, an AI-attributable dollar figure. Lloyds is small enough that the £50M is a meaningful percentage of net income (~3%). It can be verified against future financial statements. It is the analog of JPMorgan’s disclosure for a smaller bank.
Danske Bank: bundled disclosure. Q1 2026 call: ROE 13.3% in 2025 → above 14.5% by 2028, with the CEO saying “productivity gains from technology and AI more than offsetting inflation and the cost of growth.” Notice the bundling: AI productivity, technology productivity, inflation absorption, and cost of growth are combined into a single non-decomposable claim. Investors cannot back out the AI-specific contribution. It is consistent with the qualitative-language pattern Goldman’s research identified, dressed in the language of strategic clarity.
The spectrum is its own data point. JPMorgan’s disclosure quality is structurally different from the other four because Jamie Dimon decided it would be. There is no technical reason the other four cannot produce JPMorgan-style disclosure. There is a strategic reason they are not yet doing so: the moment they publish the number, the number becomes the baseline they are measured against. Most CFOs are not yet willing to take that bet.
5. The Pattern Across Four Quarters
The full four-quarter dataset (Q2 2025, Q3 2025, Q4 2025, Q1 2026) shows the inflection clearly. Through Q2 2025, AI mentions on earnings calls correlated positively with stock-price reaction regardless of disclosure quality. Through Q4 2025, the correlation weakened — quantitative AI disclosures were modestly rewarded, qualitative AI disclosures stopped moving the stock. In Q1 2026, the correlation flipped negative for qualitative-language outliers. Meta’s “very technical question” was the most visible example. It was not the only one.
The pattern is consistent with how every prior capex cycle resolved. The 1996-2000 telecom buildout traveled the same arc: enthusiasm, capex acceleration, qualitative disclosure language, the “show me the cash flow” inflection, repricing of the qualitative-language outliers. The companies whose capex was in service of legible revenue (Cisco’s products, Lucent’s exports) survived the inflection. The companies whose capex was in service of “leading the next era” (Global Crossing, WorldCom) did not. The 2000s broadband buildout repeated the cycle on shorter timescales. The current AI cycle is now somewhere between the qualitative-language phase and the “show me the cash flow” inflection. Q1 2026 is the start of the inflection, not the end of it.
The companies that handle the inflection well share three characteristics. First, AI capex is in service of an existing revenue line, not an aspirational one. Alphabet’s $20B+ Q1 cloud revenue is the existing line. Microsoft’s Azure run-rate is the existing line. The capex compounds into a measurable customer-facing product. Second, the AI-attributable productivity is disclosed in a format analysts can model — basis points of cost-to-income improvement, dollars of headcount avoidance, percentage uplift in revenue per employee. JPMorgan’s $1.5-$2B is the model. Lloyds’s £50M-to-£100M is the model at smaller scale. Third, capex is paced against disclosure milestones — capex acceleration is tied to disclosed productivity progress, not to “leading lab” rhetoric.
The companies that mishandle the inflection share three opposite characteristics. First, AI capex is in service of an aspirational revenue line (“AGI,” “leading lab,” “the metaverse next layer”). Second, productivity disclosure is qualitative or bundled. Third, capex is paced against competitive-positioning rhetoric — “we cannot be left behind” — rather than against measured productivity progress.
The first set of companies will be rewarded for the next two years. The second set will be repriced. Q1 2026 is when the repricing started.
6. The CFO’s Q2 Disclosure Decision
Every CFO of a public company that discusses AI on earnings calls now has a Q2 2026 disclosure decision to make. The decision is not whether to mention AI. The decision is whether to publish a number.
The arguments against publishing a number are familiar. Once published, the number is auditable. If the number does not materialize in the financial statements, the gap becomes a future earnings call problem. The CFO who chooses qualitative language preserves optionality. The CFO who publishes a number gives that optionality up.
The arguments for publishing a number are now structural. Q1 2026 demonstrated that the market is starting to discount qualitative AI language, not just decline to reward it. The CFO who continues with qualitative language in Q2 2026 is no longer playing for upside; they are accepting a discount. The discount may be small in any given quarter, but it compounds.
The right disclosure format is the JPMorgan format, scaled appropriately. (a) Total tech budget. (b) AI-specific incremental within that budget. (c) AI-attributable business value, projected and disclosed. (d) Use-case count, with rough qualitative shape of where the value is concentrated. (e) Year-over-year comparison versus a prior baseline. The five elements together give analysts enough to model. They are also small enough that the disclosure can be made in two paragraphs of the prepared remarks.
The CFO who publishes this in Q2 2026 will be early. The CFO who publishes it in Q4 2026 will be on time. The CFO who has not published it by Q2 2027 will be experiencing the qualitative-language discount as a structural feature of the company’s valuation.
7. What This Means for the Three Audiences
Public company shareholders. The Goldman 90% finding is the screen to run on your own portfolio. Pull the Q1 2026 earnings call transcript for each holding. Count quantitative AI metrics. Count qualitative AI mentions. The ratio is your forward exposure to the disclosure-language discount. Companies above 50% quantitative are positioned for the inflection. Companies below 20% quantitative are at risk.
Private company executives and CFOs. You have one quarter, possibly two, to construct the AI-attributable disclosure framework before it becomes mandatory in your industry’s investor communications. The framework itself is not hard. The hard part is building the underlying measurement infrastructure — workflow telemetry, productivity baselines, AI-attributable revenue and cost categorization — that lets you produce the disclosure auditably. Most companies do not have this infrastructure. The companies that build it in 2026 will have it as a competitive advantage in 2027 and 2028.
AI vendors. Your customers are about to get serious about ROI measurement. The ones who already are (JPMorgan, Lloyds, the hyperscalers) are your reference accounts. The ones who are not are your sales pipeline. The product-marketing claim that wins in 2026-2027 is not “transformational” — it is “auditable.” If your platform produces the workflow telemetry and productivity baselines that let a CFO publish the JPMorgan-style five-element disclosure, that is now a procurement criterion. If it does not, you are in a price war.
What to Do This Quarter
1. CFOs of public companies: Decide your Q2 2026 disclosure posture by mid-June. The quantitative-disclosure benchmark is JPMorgan’s five-element framework: tech budget, AI-specific incremental, AI-attributable business value (projected), use-case count, year-over-year comparison. Whatever you decide, decide it before the IR team improvises on the call.
2. Senior corporate officers: Run the Goldman 90% screen on your own company’s last four earnings calls. If you are in the qualitative-language 90%, you have one quarter to build the measurement infrastructure that lets you exit it. The infrastructure work — workflow telemetry, productivity baselines, AI-attributable revenue and cost categorization — is the prerequisite to the disclosure.
3. Public-equity investors: Re-screen your portfolio for disclosure-quality. Companies above 50% quantitative AI language are positioned for the inflection. Companies below 20% are at risk. The ratio is your forward exposure to the qualitative-language discount.
4. AI vendors: Re-pitch your platform around the auditability of the productivity claim it enables. The customers who can publish JPMorgan-style disclosures will pay a premium. The customers who cannot are about to enter a price war on commodity capabilities.
The Strategic Read
Q1 2026 is the quarter the AI capex cycle stopped being free. From 2023 through 2025, every company that announced AI investment was rewarded by the market regardless of whether the investment was producing measurable returns. The reward came from the narrative — leading lab, AI-native, transformational — and the narrative did not require any specific evidence. The reward function was structurally biased toward qualitative language because there was no penalty for it.
The penalty arrived on April 29 when Meta’s stock dropped 6% on a quarter that included revenue growth of 33% and profit growth of 61%. The penalty was for “very technical question.” The market spent two years tolerating that kind of answer. It is no longer tolerating it.
The companies that handle the next four quarters correctly will be the ones whose AI capex is in service of disclosed, auditable, financial-statement-legible value creation. Alphabet is one. JPMorgan is one. Lloyds at smaller scale is one. The companies that handle it incorrectly will be the ones whose capex is in service of competitive-positioning rhetoric without the disclosure infrastructure underneath. Meta’s $145B is the largest example. There are dozens of smaller ones in every industry vertical.
The deeper signal is that AI is now mature enough as a budget category that the discipline applied to it has to match the discipline applied to other large budget categories. Cloud capex went through this transition in 2018-2019 — the “cloud-first” rhetoric stopped being its own justification, and CFOs started disclosing cloud spend with the same rigor they disclosed real estate or supplier spend. AI is reaching the same maturity threshold five years faster, because the absolute dollar volumes are larger and the macro stakes are higher. The CFO disclosure framework is the next layer of infrastructure to build.
The companies that build it in 2026 will own a structural advantage through the rest of the decade. The companies that defer it will spend the rest of the decade explaining why their stock trades at a discount to peers with similar fundamentals but better disclosure.
The earnings call gap is the most important risk variable in public-equity AI exposure right now. It is not a question of whether AI works. It is a question of whether the company can prove it does — auditably, on the income statement, in a format analysts can model.
The companies that can will outperform. The companies that cannot will be repriced. Q1 2026 was the inflection. Q2 will tell you who learned and who didn’t.
The earnings call gap is now four quarters wide. Q1 2026 was the quarter the market started pricing it in. The CFOs who publish a number in Q2 will be early. The ones who don’t by Q2 2027 will be discounted structurally.
About the Author
Thorsten Meyer is a Munich-based futurist, post-labor economist, and recipient of OpenAI’s 10 Billion Token Award. He spent two decades managing €1B+ portfolios in enterprise ICT before deciding that writing about the transition was more useful than managing quarterly slides through it. More at ThorstenMeyerAI.com.
Related Dispatches
- Your AI Vendor’s AI Vendor — agent supply chain compromise (Vercel × Context AI)
- Single Digits — the April 2026 open-weight inflection
- AI-Washed — the 47.9% / 9% layoff narrative gap
- The 27% Problem — Anthropic’s enterprise lead and Google’s $750M check
- The Bubble Is Not in Valuations — the productivity gap
- The Agent Trap — why 90% of AI launches are infrastructure liars
- The Channel Move — Anthropic × Wall Street PE channel acquisition
Sources
- Fortune, Meta is spending up to $145 billion this year on AI. When asked about signs of ROI, Zuckerberg said ‘that’s a very technical question’ (2026-04-29)
- Google Blog, Q1 2026 earnings call: Remarks from our CEO (2026-04-30)
- Wall Street Horizon / Advisor Perspectives, Q1 Earnings Kick Off: Strong Results and Record CEO Confidence (2026-04-21)
- Investing.com, Earnings call transcript: Danske Bank Q1 2026 (2026-04-30)
- NeuralCoreTech, Agentic AI in Banking & Finance 2026: How Wall Street Is Deploying Autonomous Systems (2026-03-20)
- DeepHumanX, 2026: The Year AI ROI Gets Real, or Your Board Stops Believing (2026-02-17)
- BCG, As AI Investments Surge, CEOs Take the Lead (2026-01-12)
- WNDYR, 2026: The Year AI ROI Gets Real and Forces a Strategic Fork in the Road (2026-04)
- NBER survey of 6,000 executives across four countries (cited in DeepHumanX, 2026)
- Goldman Sachs equity research (90% qualitative-language finding, 2026)
- Deloitte State of AI in the Enterprise 2026 (n=3,235 leaders globally)
