A 24- to 26-Month Outlook for the Commercial Maturation of Artificial Intelligence
By Jeffrey A. Newman Esq. MBA, with the help of AI
I have been studying the significant advances of AI technology with interest. The speed of this Advancement is obvious, but the cost has been enormously high and there are valid questions as to how businesses using and providing these services will make up that cost and gain profits. There is also a topic of equal significance and that is one of timing. The reason the subject is so sensitive is that AI has held up our economic system and its front, the stock market, for many months. That position cannot be sustained for too long. However, if the pattern and pathway to profitability emerges from the fog of the AI battles to survive, this phenomenon has the potential not only to lift up our nation and the economy, but also to buoy the global economy. If I were a betting person, which I am not, I’d lay my money that it will all come together. The probability is that it will pay off over time. The key is how much time. I am not an engineer, computer scientist, or economist. However, I’ve had my good share of analyzing companies in my 43 years in my law practice. I represent whistleblowers, and I review, read SEC filings, analyze corporate governance, regulations violations, funding, andtaxes. I also invest in the atock market and research companies for that daily. The issue about AI finding its place and succesding to profit is highly relevant to the well being of our nation and I want our country to succeed. I want AI to help us reduce our nation’s national debt TO ZERO. I also want it to help us solve the other issues we face. So I asked AI to help examine AI’s pathway to profitability. This report is just the beginning.
The core question is this: can the artificial intelligence industry generate profits commensurate with the trillion-dollar scale of investment being made into it over the next two years — has a more nuanced answer in mid-2026 than it did even six months ago. Real revenue is now arriving at scale: Microsoft’s annualized AI revenue reached approximately $37 billion (up 123% year over year)[1], The four largest U.S. hyperscalers have committed roughly $725 billion in 2026 capital expenditure[2], and the two leading model laboratories — OpenAI and Anthropic — together exceed roughly $70 billion in run-rate revenue between them[3] [4]. At the same time, J.P. Morgan calculates that the industry must produce approximately $650 billion in new annual revenue just to deliver a 10% return on the infrastructure already in flight[5], and an MIT study found that 95% of enterprise generative-AI initiatives produced no measurable profit-and-loss impact within six months of pilot completion[6]. The gap between what has been spent and what has been earned is the central financial question of the next 24 to 26 months.
There is no question that a profitable commercial layer of AI is in fact emerging — but it is narrower, more concentrated, and more fragile than the headline capex numbers suggest. Profits are arriving first and most reliably in : (1) the chip and hardware vendors, principally Nvidia; (2) the hyperscale cloud providers whose existing enterprise relationships allow them to sell AI as an add-on to durable contracts; and (3) the frontier model laboratories selling agent-style and coding-style products to enterprise developers. Outside these categories, the profit picture is largely vague and unproven, resting on financing arrangements that will be stress-tested in 2027. There’s a lot of lending going on among the top players, and the loans run in a circle from one to the next to the next and back to the lender, and that’s definitely not a basis for the kind of building block we need for a reliable, vibrant, successful business foundation.
The present setup is too fragile. It is so fragile that the single largest risk is not a technical failure of AI, but rather a financing-and-accounting reckoning. About $800 billion of private credit is expected to fund AI infrastructure between 2025 and 2028[13], much of it sitting in special-purpose vehicles, GPU-collateralized loans, and 15-to-20-year offtake contracts whose underlying revenue assumptions are large and unverified. Q4 2026 is widely identified as the first quarter in which year-one revenue attribution becomes auditable, which is when the first capital-expenditure impairment tests are expected to land on hyperscaler balance sheets[14]. A second, related risk is that inference prices are collapsing at roughly 50× per year for equivalent model performance[15], which is excellent news for AI users but corrosive to the revenue assumptions baked into the long-term lease commitments that financed the build-out.
The Revenue That Is Actually Arriving
It is no longer accurate to say AI revenue is speculative. The numbers are large and growing rapidly, but they are unevenly distributed across the stack. The cleanest way to look at the pathway to profitability is to examine each layer separately, because each one has very different economics.
A. The chip layer — already profitable, at extraordinary margins
Nvidia reported GAAP gross margins of 74.9% in its most recent quarter and is guiding to roughly 75% on a full-year basis[16]. Data-center revenue is now the dominant contributor, with deployments anchored by long-dated commitments from OpenAI (10 gigawatts), Anthropic (1 gigawatt initial commitment), Microsoft, Google, Oracle, and xAI. For an investor focused on “where is the actual income,” the chip layer is the unambiguous answer: Nvidia is earning, in real cash, what an oil major would earn in a generational supply squeeze, with the additional advantage that the buyers are all investment-grade or near-investment-grade counterparties.
The vulnerability of the chip layer is not demand in the next 24 months — that is already contracted — but customer concentration. Microsoft, Meta, Amazon, Alphabet, and Tesla together account for more than 40% of Nvidia’s revenue[17]. If any one of those buyers were to pause or stretch its capex schedule, the impact on Nvidia’s growth rate would be immediate, even if margins on shipments already made remained intact.
B. The hyperscale cloud layer — profitable as a category, with AI as accelerant
Microsoft, Google Cloud, and AWS are the second clear winner. Microsoft’s annualized AI revenue reached approximately $37 billion in its most recent quarter, with Azure cloud services growing 40%[18]. Google Cloud revenue grew 63.4% year over year — a record pace — which is unusual at that scale. Importantly, this AI revenue is being sold into customer relationships that already exist for non-AI workloads. That is what makes it durable: the AI line item is being added to enterprise commitments that the customer was making anyway. It is not a new product category trying to find its market; it is a margin expansion on an existing one.
The credit-quality distinction matters here. The Tier-1 hyperscalers — Microsoft, Google, Amazon, and Meta — are funding their AI capex predominantly out of operating cash flow rather than debt[19]. Their AI build-outs do not depend on capital markets remaining open at the current pace.
C. The model laboratory layer — racing toward, or away from, break-even
Here the picture diverges sharply between the two leaders. Anthropic’s reported annualized run rate climbed from roughly $9 billion at the end of 2025 to approximately $30 billion in April 2026[20] and approximately $47 billion in mid-May 2026[21]. More importantly, Anthropic is projecting an operating profit of approximately $559 million in Q2 2026 — which, if it materializes, would be the first profitable quarter in the history of frontier AI[22]. The mechanism is unit economics: compute cost per dollar of revenue is reported to have fallen from $0.71 in Q1 to a projected $0.56 in Q2 — a 21% improvement in a single quarter, driven principally by the Claude Code product line, which alone reached $2.5 billion in run-rate revenue by February 2026[23].
OpenAI’s trajectory is the mirror image. Annualized revenue is approximately $25-35 billion depending on the measurement methodology[24] [25], but gross margins fell from roughly 40% in 2024 to approximately 33% in early 2026, inference costs quadrupled year over year, and projected losses for full-year 2026 are approximately $14 billion[26]. Each new dollar of OpenAI revenue is, on the company’s own projections, more expensive to produce than the last. This is the inverse of what one wants to see in a maturing business.
The cause appears to be the mix of consumer subscriptions versus enterprise contracts. Anthropic’s growth is enterprise-led and concentrated in coding workflows where customers can directly attribute revenue or labor savings to the tool. OpenAI’s base remains weighted in consumer subscriptions, where retention is meaningfully weaker: enterprise customers retain at 88% after one year, team customers at 68%, and consumer Plus subscribers at only 59%[27].
D. The enterprise application layer — real, but narrow
Outside the hyperscalers themselves, genuine third-party enterprise AI revenue is still small. UBS estimates total third-party AI product revenue across all listed software companies at approximately $2.5 billion, with Microsoft alone accounting for more than 80%[1]. That figure is striking: stripped of hyperscaler-to-hyperscaler sales and internal deployments, the genuine external enterprise software AI category is, today, smaller than a single mid-tier SaaS company. This is the most important data point in the entire report and something everyone should consider. The infrastructure build-out is sized for an enterprise software market that has not yet materialized at scale outside of Microsoft’s own franchise. Salesforce (Agentforce), ServiceNow, Workday, Adobe, SAP, and the major vertical-SaaS incumbents are positioned to deliver that market over the 24-26 month horizon, but the actual P&L contribution remains modest in mid-2026
How the Money Will Be Made Over the Next 24-26 Months
Setting aside data-center construction, four monetization pathways are sufficiently visible today that one can have a defensible view on them by mid-2028.
A. Seat-based enterprise subscriptions and AI add-ons
This is the dominant monetization model in mid-2026 and is the most likely to scale. Microsoft 365 Copilot, ChatGPT Enterprise, Claude for Enterprise, GitHub Copilot, and the Agentforce-style products from Salesforce and ServiceNow all sell on a per-seat-per-month basis, attached to existing enterprise contracts. The math is favorable because the customer is comparing the cost against a fully-loaded employee — typically several thousand dollars per seat per year against a labor cost an order of magnitude higher. ChatGPT alone reported 9 million-plus paying business users by February 2026, a 4× increase in five months, and ChatGPT Enterprise seats grew 9× year over year in 2025[1].
The strength of this pathway is that it is already producing real revenue with real retention. The vulnerability is that it depends on customers continuing to believe the seat is generating value. The MIT NANDA study[2] and IBM’s Institute for Business Value (which measured a 5.9% ROI on enterprise-wide AI initiatives against a 10% capital investment)[3] suggest the majority of buyers are not yet seeing measurable returns. If those measurements harden over the next 24 months, seat counts could plateau even where overall AI usage continues to grow.
B. Vertical, task-specific AI products
The clearest, most profitable, sizeable niche is software engineering. Claude Code grew from launch to $2.5 billion in annualized revenue in approximately a year[4], and the broader category — Copilot, Cursor, Windsurf, Tabnine — together represents the first genuinely commodity-pricing-tolerant AI application: developers pay for results because the time savings are auditable. Adjacent task-specific categories — legal-document review, medical-claims processing, customer-service deflection, regulatory-compliance automation — are following the same pattern but lag by 18-24 months. ry.
B. API and inference-as-utility
Here is a large problem. Per-token API pricing is the layer most exposed to commoditization. Anthropic has cut prices by roughly 67%, Google by 70-80%, and OpenAI has repeatedly reduced costs across successive models. Epoch AI estimates that inference prices for equivalent model performance are falling at a median of 50× per year[5]. This is excellent for application builders — it means the model layer accounts for a small share of the total cost of ownership (labor and integration typically account for 60-75% of project cost) — but it also means the model laboratories cannot rely on API revenue per se as a durable profit source. They must climb the stack into agentic products and seat subscriptions, which both leaders are doing.
C. Agentic and autonomous products billed by task outcome
The most recent monetization innovation is per-task or per-outcome pricing for autonomous agents that complete defined work — drafting a brief, processing a claim, scheduling a fleet, reconciling an account. Salesforce Agentforce, ServiceNow’s AI agents, and the agent product lines from both Anthropic and OpenAI now bill in this way for at least some customers. The economics are appealing because the buyer pays only when the work product is delivered, which neutralizes the ROI-measurement problem identified in the MIT study. The question over the 24-26 month horizon is whether reliability has improved enough to make outcome-based pricing scalable beyond carefully bounded use cases. As of mid-2026, this remains the most exciting and the most genuinely uncertain part of the commercial picture.
D. Margins Are a Major Issue
The reason a 95% enterprise-pilot failure rate[1] and a $30+ billion run-rate model laboratory can coexist is that AI economics are bifurcating. A small number of players have crossed into improving unit economics. Yetthe largest cohort of buyers is still paying for capacity they cannot use productively. The next 24-26 months will likely further separate these two cohorts.
E. Compute-cost ratio
Anthropic’s reported compute cost of $0.56 per dollar of Q2 2026 revenue, down from $0.71 in Q1[2], is the cleanest published unit-economics datapoint in frontier AI. If sustained and not the product of methodology choices, it implies that gross margins on Anthropic’s revenue base will reach roughly 44% in Q2 and continue to expand. There is not yet a software company gross margin (mature SaaS runs at70-80%), but it is structurally better than the model laboratories were achieving even nine months earlier.
The contrasting OpenAI data — gross margin contraction from 40% to 33% with inference costs quadrupling year over year — illustrates the central tension. As models become more capable, they tend to consume more compute per query, which can offset price-per-token deflation. The economic question is not whether inference is getting cheaper (it is) but whether usage intensity per dollar of revenue is rising faster than the cost curve is falling. For Anthropic, the answer in mid-2026 is no. For OpenAI, it is yes.
Depreciation Problem
Hyperscalers depreciate AI data-center hardware over five to seven years. Architectural obsolescence in GPUs, however, occurs roughly every 18 months[3]. The gap between accounting life and economic life is something auditors will be forced to address once year-one revenue from these assets becomes verifiable, which is widely expected to begin in Q4 2026. If revenues attributable to those assets fall short of the capacity assumptions baked into the depreciation schedule, a material-adverse-change clause triggers an impairment review and a non-cash write-down hits the income statement.
Important nuance: this is an accounting event, not a cash event. The hardware does not disappear. But for the share price and the cost of new capital, impairment cycles matter a great deal — they reset the price of debt for the next round of construction. This is the most identifiable event over the next 24 months that could materially change the AI investment narrative, even as the underlying technology continues to advance.
3.3 A simplified view of the 24-26 month margin pathway
| Layer | Today (Mid-2026) | Likely by Mid-2028 | Direction |
| Chips (Nvidia) | ~75% gross margin | 60-70% gross margin | Slight compression |
| Hyperscale cloud | Mid-60s% on cloud overall; AI accretive | Stable; AI a growing share | Durable |
| Frontier model labs | 33-44% gross margin, diverging | 45-55% for leaders; losses for laggards | Bifurcating |
| Enterprise AI SaaS | Embedded; small standalone P&L | Becomes a real category | Improving |
| Neoclouds / GPU lessors | Negative free cash flow, heavy debt | Consolidation; some defaults | Deteriorating |
Notes: Figures are illustrative directional ranges drawn from cited sources, not point forecasts. Margin compression at Nvidia reflects expected competition from custom hyperscaler silicon (Google TPU, Amazon Trainium, Microsoft Maia) and the historical pattern that gross margins in any chip cycle normalize as supply catches demand
Circular financing the crux of increased risk
IDC and others have documented a pattern in which the same capital functions simultaneously as vendor payment and equity stake — Nvidia invests in OpenAI and sells GPUs to OpenAI; OpenAI buys compute from Oracle and CoreWeave, which in turn buy chips from Nvidia; Oracle’s $300 billion contract with OpenAI underpins Oracle’s own data-center debt[1] [2]. None of this is fraudulent. It is, however, a structure in which a meaningful share of reported AI demand is funded by the same balance sheets that book the supply. The Bank for International Settlements, in its March 2026 Quarterly Review, described it as “shadow borrowing”: obligations economically akin to debt but residing outside corporate balance sheets[3].
The practical risk this creates over your 24-26 month window is not a 2008-style banking crisis. It is, instead, the prospect of a slow repricing of Tier-2 and Tier-3 credit (Oracle, CoreWeave, Lambda, Crusoe) as the offtake contracts underlying their debt are tested against actual end-customer revenue. Financial Times reporting indicates that JPMorgan, Morgan Stanley, and SMBC have been trying for over six months to offload approximately $38 billion of construction debt tied to Oracle data-center projects in Texas and Wisconsin, with some banks accepting discounts to clear inventory[4]. When the banks closest to the underwriting are willing to take losses to reduce exposure, it is the most diagnostic signal available about where the actual risk sits.
The revenue-to-capex gap
The math from J.P. Morgan and Bain remains the largest single bear-case argument. J.P. Morgan estimates the industry must generate approximately $650 billion in new annual revenue just to deliver a 10% return on the infrastructure being built. Current AI-attributable revenue, even under generous assumptions, sits somewhere between $50 billion and $150 billion. That is a 4× to 13× gap[5]. Goldman Sachs CEO David Solomon has stated publicly that “a bunch of the capital being deployed in AI will actually not produce any returns,” and Goldman’s chief economist concluded that AI contributed essentially zero to U.S. economic growth in 2025[6].
I do not believe the gap implies that AI fails commercially. It implies that the timeline for the infrastructure to earn its cost of capital is longer than the lease tenors and the depreciation schedules assume — which is exactly the condition under which impairment cycles and credit re-ratings occur.
Model commoditization
The performance gap between open-source and proprietary frontier models compressed from approximately 8% to 1.7% in a single year, per Stanford’s 2025 AI Index[7]. If that compression continues, the long-run revenue available to closed-weight frontier laboratories will be smaller than current valuations imply — not zero, but smaller, with profits concentrated in vertical and agentic products rather than raw model access. This is the principal threat to OpenAI and Anthropic’s long-run pricing power, and the principal reason both companies are working aggressively to ascend into agent and seat-based pricing.
Power and the hardware horizon
I asked the AI to assume the data center construction problems are solved. The issues faced in completing the data centers, however, are not solved and are rising. Demand for electricity and water and draining resources has not been reconciled. PJM Interconnection failed for the first time in its history to procure enough electricity to meet its summer 2027 reserve target, falling approximately 6,600 megawatts short; 94% of the projected load growth that drove that shortfall came from AI data centers[8]. IEA projects global data-center electricity consumption to roughly double by 2030, reaching approximately 945 TWh — comparable to Japan’s current electricity consumption[9]. Even if your assumption holds, electricity costs will rise materially in the corridors where data centers are concentrated, which will compress hyperscaler operating margins and accelerate the impairment cycle described above.
I am not predicting an AI crash. The data simply does not support it. The technology is being adopted, real revenue is being generated, and the most disciplined operators are converging on profitability. What is being suggested is that the path is narrower than the trillion-dollar capex commitments imply, that the financing structure underlying those commitments will be tested before mid-2028, and that the difference between strong and weak credits within the AI stack will be repriced materially in 24-36 months. Given the rate of costs thats a long time. However, as mentioned, if a pathway starts to emerge that may be a time frame the lenders will tolerate.
Thoughts
So, based on what I see, the profit pathway exists and is already producing real cash returns at the chip and hyperscale-cloud layers, is on the cusp of producing them at the leading frontier laboratory (Anthropic), and remains aspirational across most of the enterprise application layer. The financial pressure points over the window are concentrated in the financing structure rather than in the technology itself: vendor-financed circularity, stretched investment-grade credits (principally Oracle), the GPU-collateralized neocloud segment, and the gap between revenue assumptions baked into 15-to-20 year leases and the actual rate at which enterprise customers can put AI capacity to profitable use.
AI is, in mid-2026, simultaneously the most genuinely profitable technology category of the past two decades at certain points in the stack, and the most financially leveraged technology category at others. Both statements are true at the same time, and an investment view of the industry that emphasizes only one of them will miss the most important developments of the next two years.
[1]FullStack Labs, “Generative AI ROI: Why 80% Fail” (May 2026), citing IBM Institute for Business Value, https://www.fullstack.com/labs/resources/blog/generative-ai-roi-why-80-of-companies-see-no-results
[1]IDC, “Circular Financing in AI: Why Enterprise Apps Matter” (June 2026), https://www.idc.com/resource-center/blog/circular-financing-has-muddied-the-ai-story-watch-the-application-layer-instead/
[2]Inc. Magazine, “Oracle and the AI Boom’s Hidden Debt Bomb” (June 2026), https://www.inc.com/fast-company-2/oracle-ai-boom-hidden-debt-nvidia-jensen-huang/91357055
[3]LinkedIn (Joe Toppe), “The $800 Billion Bet: How AI Data Centers Are Being Financed” (May 2026), https://www.linkedin.com/pulse/800-billion-bet-how-ai-data-centers-being-financed-where-joe-toppe-rfaoe
[4]LinkedIn (Joe Toppe), “The $800 Billion Bet: How AI Data Centers Are Being Financed” (May 2026), https://www.linkedin.com/pulse/800-billion-bet-how-ai-data-centers-being-financed-where-joe-toppe-rfaoe
[5]Anomaly Investment Partners, “This Obviously is an AI Bubble. The Math Says So” (June 2026), https://anomalyinvestments.substack.com/p/this-obviously-is-an-ai-bubble-the
[6]Anomaly Investment Partners, “This Obviously is an AI Bubble. The Math Says So” (June 2026), https://anomalyinvestments.substack.com/p/this-obviously-is-an-ai-bubble-the
[7]Apptitude, “AI Model Costs Are Collapsing” (May 2026), citing Epoch AI and Stanford 2025 AI Index, https://www.apptitude.io/blog/ai-model-costs-collapsing-where-value-lives/
[8]Knowledge at Wharton, “AI’s Supply Chain Problem” (May 2026), https://knowledge.wharton.upenn.edu/article/ais-supply-chain-problem/
[9]Schneider Electric Perspectives, “The AI Power Crunch” (May 2026), citing IEA data center projections, https://perspectives.se.com/latest/the-ai-power-crunch-what-companies-need-to-know-about-electricity-price-risk
[10]JanBask Blog, “Generative AI ROI: Why Enterprise Projects Fail on Paper” (June 2026), citing MIT Project NANDA, The GenAI Divide, https://www.janbask.com/blog/enterprise-generative-ai-roi-measurement-gap/
[11]FullStack Labs, “Generative AI ROI: Why 80% Fail” (May 2026), citing IBM Institute for Business Value, https://www.fullstack.com/labs/resources/blog/generative-ai-roi-why-80-of-companies-see-no-results
[1]JanBask Blog, “Generative AI ROI: Why Enterprise Projects Fail on Paper” (June 2026), citing MIT Project NANDA, The GenAI Divide, https://www.janbask.com/blog/enterprise-generative-ai-roi-measurement-gap/
[2]The Synthesis, “The Margin” (May 2026), https://thesynthesisai.substack.com/p/the-margin
[3]LinkedIn (Marko Markovic), “AI Capex Impairment Tests Hit Hyperscalers in Q4 2026” (May 2026), https://www.linkedin.com/posts/marko-markovic-msc-mba%F0%9F%87%B7%F0%9F%87%B8-330b52b_q4-2026-is-when-the-first-ai-capex-impairment-activity-7459557938894901248-lvCU
[1]Master of Code, “ChatGPT Statistics in Companies” (June 2026), https://masterofcode.com/blog/chatgpt-statistics
[2]JanBask Blog, “Generative AI ROI: Why Enterprise Projects Fail on Paper” (June 2026), citing MIT Project NANDA, The GenAI Divide, https://www.janbask.com/blog/enterprise-generative-ai-roi-measurement-gap/
[3]FullStack Labs, “Generative AI ROI: Why 80% Fail” (May 2026), citing IBM Institute for Business Value, https://www.fullstack.com/labs/resources/blog/generative-ai-roi-why-80-of-companies-see-no-results
[4]VentureBeat, “Anthropic says it hit a $30 billion revenue run rate” (May 2026), https://venturebeat.com/technology/anthropic-says-it-hit-a-30-billion-revenue-run-rate-after-crazy-80x-growth
[5]Apptitude, “AI Model Costs Are Collapsing” (May 2026), citing Epoch AI and Stanford 2025 AI Index, https://www.apptitude.io/blog/ai-model-costs-collapsing-where-value-lives/
[1]Anomaly Investment Partners, “This Obviously is an AI Bubble. The Math Says So” (June 2026), https://anomalyinvestments.substack.com/p/this-obviously-is-an-ai-bubble-the
[1]Stocktwits, “Microsoft, Meta And Google Just Silenced AI Spending Critics …” (June 2026), https://stocktwits.com/news-articles/markets/equity/microsoft-meta-and-google-just-silenced-ai-spending-critics-in-one-earnings-night-as-big-tech-capex-swells-to-725-b/cZBtCIgReEx
[2]The Business Engineer, “The AI Capex Map & The State of AI Hyperscalers” (May 2026), https://businessengineer.ai/p/the-ai-capex-map-and-the-state-of
[3]VentureBeat, “Anthropic says it hit a $30 billion revenue run rate” (May 2026), https://venturebeat.com/technology/anthropic-says-it-hit-a-30-billion-revenue-run-rate-after-crazy-80x-growth
[4]KuCoin Blog, “Anthropic’s IPO Filing Signals a New Era for AI Investing” (June 2026), https://www.kucoin.com/blog/anthropic-ipo-filing-signals-a-new-era-for-ai-investing
[5]Anomaly Investment Partners, “This Obviously is an AI Bubble. The Math Says So” (June 2026), https://anomalyinvestments.substack.com/p/this-obviously-is-an-ai-bubble-the
[6]JanBask Blog, “Generative AI ROI: Why Enterprise Projects Fail on Paper” (June 2026), citing MIT Project NANDA, The GenAI Divide, https://www.janbask.com/blog/enterprise-generative-ai-roi-measurement-gap/
[7]NVIDIA Investor Relations, “Q1 FY2027 Financial Results” (May 2026), https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-Financial-Results-for-First-Quarter-Fiscal-2027/default.aspx
[8]Stocktwits, “Microsoft, Meta And Google Just Silenced AI Spending Critics …” (June 2026), https://stocktwits.com/news-articles/markets/equity/microsoft-meta-and-google-just-silenced-ai-spending-critics-in-one-earnings-night-as-big-tech-capex-swells-to-725-b/cZBtCIgReEx
[9]The Synthesis, “The Margin” (May 2026), https://thesynthesisai.substack.com/p/the-margin
[10]The Synthesis, “The Margin” (May 2026), https://thesynthesisai.substack.com/p/the-margin
[11]Inc. Magazine, “Oracle and the AI Boom’s Hidden Debt Bomb” (June 2026), https://www.inc.com/fast-company-2/oracle-ai-boom-hidden-debt-nvidia-jensen-huang/91357055
[12]JanBask Blog, “Generative AI ROI: Why Enterprise Projects Fail on Paper” (June 2026), citing MIT Project NANDA, The GenAI Divide, https://www.janbask.com/blog/enterprise-generative-ai-roi-measurement-gap/
[13]LinkedIn (Joe Toppe), “The $800 Billion Bet: How AI Data Centers Are Being Financed” (May 2026), https://www.linkedin.com/pulse/800-billion-bet-how-ai-data-centers-being-financed-where-joe-toppe-rfaoe
[14]LinkedIn (Marko Markovic), “AI Capex Impairment Tests Hit Hyperscalers in Q4 2026” (May 2026), https://www.linkedin.com/posts/marko-markovic-msc-mba%F0%9F%87%B7%F0%9F%87%B8-330b52b_q4-2026-is-when-the-first-ai-capex-impairment-activity-7459557938894901248-lvCU
[15]Apptitude, “AI Model Costs Are Collapsing” (May 2026), citing Epoch AI and Stanford 2025 AI Index, https://www.apptitude.io/blog/ai-model-costs-collapsing-where-value-lives/
[16]NVIDIA Investor Relations, “Q1 FY2027 Financial Results” (May 2026), https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-Financial-Results-for-First-Quarter-Fiscal-2027/default.aspx
[17]LinkedIn (Joe Toppe), “The $800 Billion Bet: How AI Data Centers Are Being Financed” (May 2026), https://www.linkedin.com/pulse/800-billion-bet-how-ai-data-centers-being-financed-where-joe-toppe-rfaoe
[18]Stocktwits, “Microsoft, Meta And Google Just Silenced AI Spending Critics …” (June 2026), https://stocktwits.com/news-articles/markets/equity/microsoft-meta-and-google-just-silenced-ai-spending-critics-in-one-earnings-night-as-big-tech-capex-swells-to-725-b/cZBtCIgReEx
[19]LinkedIn (Joe Toppe), “The $800 Billion Bet: How AI Data Centers Are Being Financed” (May 2026), https://www.linkedin.com/pulse/800-billion-bet-how-ai-data-centers-being-financed-where-joe-toppe-rfaoe
[20]VentureBeat, “Anthropic says it hit a $30 billion revenue run rate” (May 2026), https://venturebeat.com/technology/anthropic-says-it-hit-a-30-billion-revenue-run-rate-after-crazy-80x-growth
[21]KuCoin Blog, “Anthropic’s IPO Filing Signals a New Era for AI Investing” (June 2026), https://www.kucoin.com/blog/anthropic-ipo-filing-signals-a-new-era-for-ai-investing
[22]The Synthesis, “The Margin” (May 2026), https://thesynthesisai.substack.com/p/the-margin
[23]VentureBeat, “Anthropic says it hit a $30 billion revenue run rate” (May 2026), https://venturebeat.com/technology/anthropic-says-it-hit-a-30-billion-revenue-run-rate-after-crazy-80x-growth
[24]Second Talent, “ChatGPT Statistics 2026: Users, Revenue, and Enterprise” (May 2026), https://www.secondtalent.com/resources/chatgpt-statistics/
[25]AI Automation Global, “Anthropic Hits $965B Valuation, Overtakes OpenAI” (May 2026), https://aiautomationglobal.com/blog/anthropic-965b-valuation-surpasses-openai-series-h-2026
[26]The Synthesis, “The Margin” (May 2026), https://thesynthesisai.substack.com/p/the-margin
[27]Second Talent, “ChatGPT Statistics 2026: Users, Revenue, and Enterprise” (May 2026), https://www.secondtalent.com/resources/chatgpt-statistics/
Jeffrey Newman, JD, MBA, a former prosecutor, is a whistleblower lawyer whose firm represents physicians and other healthcare providers who become whistleblowers in healthcare fraud cases. The firm also takes cases involving tariff fraud and export control fraud. Whistleblower laws in the U.S. allow individuals with information about export control violations or tariff fraud to report it under the False Claims Act, which, if successful, awards the whistleblower a percentage of the amount collected. The Firm’s website is www.JeffNewmanLaw.com. Attorney Newman can be reached at Jeff@Jeffnewmanlaw.com or at 617-823-3217. For other blogs, see: http://JeffNewmanLaw.com
