Bloomberg reported on 13 August 2026 that OpenAI’s annualised revenue run-rate had exceeded $40 billion. The figure is extraordinary even by the standards of the current AI cycle. It is also only half of the financial story.
OpenAI is trying to convert unprecedented demand for machine intelligence into an economic model that may require hundreds of billions of dollars of compute spending, large-scale data-centre construction and power capacity measured in gigawatts. Its prospective IPO therefore matters far beyond the valuation of one private technology company. It could become the first public-market test of whether frontier AI can combine software-like monetisation with acceptable returns on industrial-scale physical capital.
The central issue is not whether OpenAI can sell AI. It clearly can. The harder question is whether each additional dollar of revenue can eventually produce durable free cash flow after inference, training, model development, energy, financing and the continuing cost of remaining at the technological frontier.
Executive Summary
- OpenAI’s revenue run-rate has exceeded $40 billion according to Bloomberg, but it should not be described as verified ARR because the underlying revenue includes usage-sensitive and non-contractual economic activity.
- The latest major private financing valued OpenAI at $852 billion post-money, roughly 21.3 times the reported $40 billion run-rate and about 65.5 times approximately $13 billion of actual 2025 revenue.
- The 2025 cost base remains the strongest challenge to a conventional software valuation. Spending was about $34 billion, underlying operational losses were approximately $8 billion and adjusted gross margin was reported near 33%.
- OpenAI’s reported compute spending plan of roughly $600 billion through 2030 and Stargate’s $500 billion infrastructure ambition show that capital access may be as strategically important as model capability.
- A $1 trillion public valuation implies 25 times a $40 billion run-rate and requires a far more demanding cash-flow case than a simple extrapolation of revenue growth.
- Anthropic, Nvidia, hyperscalers, data centres, power markets and private credit are connected to the same AI monetisation assumption, creating cross-asset concentration risk.
- For UHNW investors and family offices, exposure may be indirect through venture funds, Microsoft, Nvidia, Nasdaq exposure, infrastructure credit and utilities.
How OpenAI Reached a $40 Billion-Plus Revenue Run-Rate
OpenAI’s commercial acceleration has been unusually steep, but the chronology requires disciplined definitions. Reported 2024 annual revenue was approximately $3.7 billion, while the year-end revenue run-rate was about $5.5 billion. By June 2025, annualised revenue had reached roughly $10 billion, according to company information reported by Reuters. A month later, Reuters reported approximately $12 billion.
The full-year 2025 figure was different. According to Financial Times reporting based on audited financial figures, actual 2025 revenue was approximately $13 billion. By year-end, Reuters reported annualised revenue of about $21.4 billion, while the Financial Times reported monthly revenue of approximately $2 billion, which would annualise to roughly $24 billion. Those figures should not be artificially reconciled because they may reflect different dates, classifications or monthly volatility.
By the end of February 2026, Reuters reported annualised revenue above $25 billion. Bloomberg’s 13 August report then placed the run-rate above $40 billion, implying growth of at least 60% in roughly five and a half months. The rise from approximately $20 billion to $40 billion is a doubling, or 100% growth.
Revenue composition is now as important as scale. By March 2026, business products represented about 40% of revenue, while OpenAI reported more than 900 million weekly active users and over 50 million subscribers. Consumer distribution remains a major advantage, but enterprise monetisation is increasingly central to the IPO thesis.
The revenue engine spans consumer and business subscriptions, API consumption, agents and coding products, with advertising and commerce still emerging. Each pool carries different retention and margin characteristics. API revenue can be deeply embedded in customer applications while remaining sensitive to token volume and price. Agents may expand revenue per workflow, but they can also multiply the inference steps needed to complete a task.
The $40 billion threshold matters because frontier AI demand has become a major commercial category.
The Revenue Quality Question
What does OpenAI’s $40 billion annualised revenue mean for investors? It demonstrates exceptional monetisation velocity, but it does not establish the contractual recurrence or margin structure associated with traditional SaaS.
A revenue run-rate annualises a recent period. ARR is a more specific recurring-revenue concept. OpenAI’s subscriptions may have SaaS-like recurrence, but API usage, enterprise consumption and agent workloads can expand or contract with customer activity. They may be economically sticky without being contractually recurring in a narrow sense.
Pricing creates the second complication. OpenAI cut API prices in July 2026 and linked those reductions to efficiency gains. Lower inference cost per task can improve unit economics, but competition can transfer the benefit to customers before it reaches shareholders. The relevant spread is between the decline in cost per unit of intelligence and the decline in price charged for it.
Model commoditisation therefore matters financially. If OpenAI, Anthropic, Chinese laboratories and open-weight models continue to compete on performance per dollar, customers gain bargaining power even as volume rises.
Enterprise adoption is the counterweight. Large deployments can become embedded in security architecture, data governance, coding environments and operating workflows. If pricing shifts from tokens towards completed tasks or outcomes, willingness to pay can rise even while token prices fall. The strongest IPO evidence would therefore be rising enterprise mix, high retention and gross profit per customer expanding faster than usage costs.
The Profitability Problem Behind the Growth
The sharpest challenge to the OpenAI IPO valuation is the 2025 cost structure. Approximately $13 billion of actual revenue sat against roughly $34 billion of spending, including about $19 billion of research and development and nearly $6 billion of sales and marketing. The company was simultaneously building models, acquiring compute and scaling a global commercial platform.
The headline attributable net loss was approximately $39 billion, equal to roughly three times 2025 revenue. Yet that figure included an estimated $30 billion non-cash structural charge linked to the prior investor structure. Treating the full loss as recurring operating cash burn would materially misstate the economics.
The more useful measure is the approximately $8 billion underlying operational loss identified in the research. Against $13 billion of revenue, that equates to about 61.5%. Reuters also reported adjusted gross margin near 33% in 2025, down from roughly 40% in 2024 as inference expenditure increased.
For a company seeking a valuation normally associated with high-growth software, a gross margin in the low thirties changes the underwriting problem. Scaling usage can expand both revenue and the physical cost of serving that revenue.
There is a credible counter-thesis. Cost per task can fall through model routing, caching, quantisation, custom kernels, alternative accelerators and better utilisation. Enterprise pricing and agent monetisation may raise revenue per workflow. If gross margin recovers while operating expenditure grows much more slowly than revenue, operating leverage could emerge sharply.
That outcome remains plausible rather than proven. OpenAI’s valuation already capitalises a significant portion of that future improvement.
Why Frontier AI Looks More Like Digital Industry Than Traditional Software
The industrial character of frontier AI becomes clearest when training is separated from inference. Training is episodic and strategically necessary to maintain frontier capability. Inference is continuous and expands with customer usage. Every consumer interaction, API call or agent step consumes compute.
OpenAI’s infrastructure strategy reflects that reality. Microsoft remains the primary cloud partner, but capacity has broadened across AWS, Oracle, CoreWeave and others. Amazon’s strategic relationship includes a $50 billion investment commitment and a $100 billion expansion of an existing multi-year compute agreement over eight years, including roughly 2 GW of Trainium capacity. CoreWeave has a five-year OpenAI contract worth approximately $11.9 billion.
Stargate gives the capital cycle its physical form. The programme was announced with a $500 billion four-year ambition and an objective of approximately 10 GW of US AI infrastructure. Later announcements placed planned capacity near 7 GW and investment above $400 billion over roughly three years. Separately, OpenAI’s reported aggregate compute spending through 2030 is about $600 billion.
Those figures are not interchangeable. The compute estimate is cumulative, Stargate is a programme objective and some infrastructure may be financed or owned by partners. The conclusion is nevertheless clear: frontier AI depends on an ecosystem whose investment requirements resemble a large industrial build-out.
Capacity utilisation will be decisive. A GPU cluster running near full utilisation against high-value enterprise demand can support attractive economics. Underutilised capacity financed at elevated cost can destroy value quickly, particularly when hardware depreciates economically faster than buildings or power assets.
OpenAI’s 2026 financing underlines why access to capital may itself be a moat. The company secured $122 billion of committed capital at an $852 billion post-money valuation. The commitment equals roughly 3.05 times the reported $40 billion annualised revenue figure and about 14.3% of the post-money valuation. Committed capital is not necessarily funded cash, but few competitors can mobilise financing at this scale.
In frontier AI, financial endurance is becoming part of technological competition.
Can OpenAI Justify a $1 Trillion Valuation?
At the $852 billion post-money private valuation, OpenAI is valued at approximately 21.3 times Bloomberg’s reported $40 billion revenue run-rate. At $1 trillion, the multiple becomes 25 times. At $1.25 trillion it rises to 31.25 times, and at $1.5 trillion to 37.5 times.
| Valuation | Multiple on $40bn run-rate | Sustainable FCF at 25x |
| $750bn | 18.75x | $30bn |
| $850bn | 21.25x | $34bn |
| $1.0tn | 25.00x | $40bn |
| $1.25tn | 31.25x | $50bn |
| $1.5tn | 37.50x | $60bn |
Can OpenAI justify a $1 trillion valuation? The answer depends less on one sales multiple than on terminal revenue, sustainable free cash flow and the capital required to reach both.
At 10 times sales, a $1 trillion value requires roughly $100 billion of revenue. At 8 times, $125 billion. At 6 times, approximately $167 billion. At 5 times, $200 billion. Those remain substantial sales multiples for a mature company with heavy infrastructure dependencies.
A cash-flow framework is more revealing. At 25 times sustainable free cash flow, $1 trillion requires approximately $40 billion of annual FCF. That equals a 26.7% margin on $150 billion of revenue, 20% on $200 billion and roughly 14.3% on the reported 2030 revenue ambition of more than $280 billion.
The 2030 forecast can therefore support a trillion-dollar valuation without software-like terminal margins if the revenue is achieved. The harder question is how much cumulative capital must be deployed first and who bears the financing burden.
Illustrative DCF scenarios in the research expose the sensitivity. A bull case using $75 billion of 2027 revenue, approximately $256 billion by 2031, FCF margins rising from 5% to 32%, a 9% cost of capital and 4% terminal growth produced an illustrative value near $1.25 trillion. A base case with $60 billion of 2027 revenue, approximately $158 billion by 2031, margins moving from negative 5% to positive 22%, a 10.5% cost of capital and 3.5% terminal growth produced about $359 billion. A bear case produced approximately $41 billion.
These are analytical scenarios, not price targets. The dispersion is the point. OpenAI is a terminal-assumption security. Small changes in the duration of growth, eventual margins, reinvestment requirements and discount rates can transform present value.
AGI optionality should be treated separately. Investors may assign value to the possibility that frontier systems create markets much larger than today’s subscriptions, APIs and enterprise software. A disciplined valuation should keep that long-duration option distinct from cash flows already visible in the business.
OpenAI Versus Anthropic and the Emerging Frontier-AI Market
Anthropic has removed any comfortable assumption that OpenAI is the only frontier laboratory capable of monetising at exceptional scale. In May 2026, Anthropic raised $65 billion at a $965 billion post-money valuation after reporting that its revenue run-rate had crossed $47 billion. That implies approximately 20.5 times run-rate sales, close to OpenAI’s 21.3 times at the $852 billion mark.
The business mix differs. OpenAI retains a much larger disclosed consumer distribution footprint, while Anthropic has shown strong enterprise penetration, including more than 1,000 business customers spending over $1 million annually by April 2026. Both are diversifying compute across strategic partners and alternative silicon.
Closely sequenced listings could create a public frontier-AI sector with live relative valuations for growth, gross margin, enterprise mix, cash burn and compute intensity.
DeepSeek, Moonshot AI and open-weight models add another pressure point. The risk is not simply that Chinese models are always cheaper. The deeper issue is that acceptable intelligence can increasingly be routed across providers according to cost, performance, sovereignty and latency. US semiconductor export controls may limit access to leading accelerators, but they can also accelerate domestic optimisation and technology fragmentation.
What OpenAI’s Growth Means for Nvidia, Hyperscalers and the AI Supply Chain
Nvidia generated approximately $215.9 billion of fiscal 2026 revenue, with a 71.1% GAAP gross margin and about $130.4 billion of operating income. The company sits at the most valuable point in the accelerator supply chain, but the durability of that value depends on downstream customers earning adequate returns on Nvidia-powered infrastructure. OpenAI’s public filings could provide some of the clearest evidence yet.
Efficiency has two effects. Lower compute cost per task can reduce hardware intensity for a given workload, but cheaper intelligence can stimulate far more usage. The relevant question for Nvidia is total compute consumed, not the cost of one token in isolation.
Microsoft remains OpenAI’s most embedded hyperscaler relationship, while Azure now sits alongside growing capacity across AWS, Oracle and CoreWeave. Microsoft’s model and product IP licence remains non-exclusive through 2032, while revenue-sharing payments continue through 2030 subject to a cap. For Microsoft investors, an IPO could crystallise value around its economic interest while exposing how much of the AI narrative depends on infrastructure spending and cloud utilisation.
Amazon is both an investor and infrastructure provider. Oracle is exposed through Stargate and large AI infrastructure commitments. CoreWeave provides the most visible analogue for pure AI compute intensity, with 2026 capex guidance of approximately $35 billion to $39 billion and the $11.9 billion OpenAI contract.
The wider hyperscaler spending envelope is extraordinary. The research places Microsoft’s 2026 spending context near $190 billion, Alphabet at $175 billion to $185 billion, Amazon near $220 billion and Meta at $130 billion to $145 billion. Using midpoints where required produces approximately $727.5 billion.
That figure is broader corporate capex, not pure AI spending. Even so, OpenAI’s $40 billion run-rate equals only about 5.5% of the combined envelope. OpenAI and Anthropic together, at roughly $87 billion of headline run-rate revenue, equal about 12%. The comparison is not an ROI calculation, because hyperscaler spending also supports cloud, advertising, search, internal workloads and long-lived infrastructure. It simply shows why frontier-lab revenue alone cannot validate the entire AI capex cycle.
The transmission chain reaches Nvidia and AMD accelerators, Broadcom custom silicon and networking, TSMC fabrication, SK Hynix, Micron and Samsung memory, Arista networking, data-centre construction, cooling, transformers, utilities and power generation. Private credit then finances part of the physical stack.
Energy is becoming a genuine macro constraint. The IEA estimates global data-centre electricity consumption at roughly 415 TWh in 2024 and approximately 945 TWh by 2030, with data centres accounting for nearly half of incremental US electricity-demand growth through 2030. That creates demand for transmission, generation, transformers and grid capacity, while intensifying capital deepening and potentially influencing local power prices and real investment. Longer term, AI productivity gains could offset part of that inflationary impulse.
OpenAI’s economics therefore sit upstream of a large share of the current technology and infrastructure capital cycle.
The IPO Could Reprice More Than OpenAI
OpenAI said in June 2026 that it had confidentially submitted a draft registration statement for a US IPO, without disclosing final size or timing. Reuters later reported that Sam Altman expected the company to become public within roughly a year, while separate reporting discussed a valuation of up to $1 trillion. None of those reports constitutes a confirmed IPO price or timetable.
Governance will matter alongside valuation. OpenAI operates as a public benefit corporation under the control of the OpenAI Foundation, creating a less conventional shareholder-rights structure than a standard listed technology company. Senior executive turnover ahead of a flotation adds another execution variable.
A large offering could become a benchmark for public and private technology markets. SpaceX’s 2026 listing established that markets can absorb a mega-IPO on exceptional scale, but an OpenAI deal would still require substantial risk capacity.
The impact could work in both directions. During bookbuilding, a large offering absorbs investor attention and capital. A successful aftermarket can then reopen technology issuance, strengthen venture exit assumptions and deepen crossover demand.
The most immediate valuation effect may occur in private markets. OpenAI carries an $852 billion post-money mark. Anthropic was valued at $965 billion in May. A public OpenAI valuation materially above its private mark could support late-stage AI NAVs and secondary pricing. A listing below that mark could force funds to reconcile stale private valuations with a live public comparable.
Future index inclusion could deepen the effect, although a mega-cap listing would not automatically enter major indices immediately. Profitability, seasoning, free float and index-specific eligibility rules matter. If OpenAI eventually qualifies, passive ownership could increase AI concentration across portfolios that never made an active decision to own frontier-model risk.
Credit is another channel. Public equity could eventually support corporate bond issuance if revenue becomes predictable and cash generation improves. That would add operating-company debt to an AI financing ecosystem already using strategic equity, infrastructure leverage, customer prepayments and private credit. Hardware depreciation and refinancing risk would become increasingly important to lenders.
Regulation may also enter valuation more visibly after listing. EU AI Act enforcement for general-purpose AI models is now live, with potential penalties of up to 3% of worldwide annual turnover in relevant circumstances. Copyright litigation remains unresolved across jurisdictions, while public disclosure would increase scrutiny of governance, data use and related-party obligations. The research provides no credible aggregate legal-liability figure, so none should be inferred.
The IPO would create more than liquidity. It would create a public reference price for frontier-AI economics.
The Hidden AI Concentration Inside Wealthy Portfolios
For UHNW investors and family offices, direct access to pre-IPO shares is only the first layer. The same economic thesis can appear repeatedly across portfolios that look diversified by asset class.
First-order exposure includes direct OpenAI or Anthropic holdings, venture capital, growth equity, secondary funds and frontier-AI software. Second-order exposure includes Microsoft, Nvidia, Amazon, Oracle, CoreWeave, semiconductors, memory, networking and data-centre developers. Third-order exposure includes utilities, power generation, electrical equipment, infrastructure funds, private credit, technology indices and structured products.
A family office might own a venture fund with OpenAI exposure, a legacy Microsoft holding, Nvidia through a technology manager, data-centre debt through private credit, a utility infrastructure fund and a Nasdaq ETF. Those positions occupy different allocation buckets. Economically, they can share one assumption: AI monetisation must grow fast enough to justify the capital being installed around it.
In an AI price-war scenario, lower model pricing compresses expected margins at frontier laboratories. Lower expected returns can reduce willingness to pay for compute. Hyperscaler spending assumptions weaken. Semiconductor and AI-cloud valuations reprice. Data-centre development expectations fall, and infrastructure credit spreads widen. No single asset needs to become impaired for correlations to rise.
Private-market valuation lag can hide the concentration temporarily. Listed Nvidia or Microsoft may fall immediately while a private OpenAI or venture-fund NAV remains unchanged until the next valuation cycle. Apparent diversification can therefore reflect stale pricing rather than lower economic risk.
Liquidity mismatch adds another layer. Public positions can be sold or hedged rapidly. Private AI stakes may face transfer restrictions and uncertain secondary liquidity. A family office facing collateral calls or tax obligations can discover that its most valuable technology exposure is also its least liquid.
Hedging is imperfect. Nasdaq futures, Nvidia or Microsoft can reduce broad AI beta, but none is a direct hedge for OpenAI. Microsoft has diversified cash flows, Nvidia is a supplier rather than a model company and public indices reflect hundreds of businesses. Basis risk can rise precisely when the model, cloud and semiconductor layers begin to trade differently.
A more sophisticated framework is look-through exposure rather than sector-label diversification. AI stress loss can be conceptualised as the sum of each position value multiplied by its sensitivity to an AI monetisation or capex shock, then overlaid with liquidity, leverage, pledged assets, derivative collateral and manager overlap.
For multi-asset investors, this is where Bancara’s broader cross-market perspective is relevant. The key issue is not identifying a single AI winner, but understanding how one demand assumption can propagate across equities, indices, commodities, currencies and other exposures. The exercise is portfolio risk mapping, not a directional recommendation.
A family office can become economically overexposed to OpenAI’s success without owning a single OpenAI share.
Bull, Base and Bear Paths Through the IPO
The useful scenarios are different regimes for the entire AI capital cycle, not point estimates for OpenAI shares.
| Scenario | Core assumption | Market transmission |
| AI Monetisation Supercycle | Enterprise agents scale, usage compounds and efficiency gains lift margins | OpenAI valuation support strengthens, Nvidia and hyperscaler demand remain robust, infrastructure utilisation rises and private AI marks expand |
| Growth Normalisation | Revenue growth slows but remains strong while capex stays elevated | Valuation becomes more earnings-sensitive, semiconductors remain supported but multiple expansion fades |
| AI Price War | Model providers compete aggressively on price and open models improve | Usage may rise while revenue per task falls, model margins compress and correlated AI assets reprice |
| Infrastructure Bottleneck | Power, GPUs, networking or financing constrain capacity | OpenAI growth becomes supply-limited, scarce infrastructure gains value and delayed projects create financing stress |
| Public-Market Repricing | Higher real rates or equity risk premia compress long-duration multiples | OpenAI and private AI marks fall even with solid operations, technology valuations de-rate and financing costs rise |
Premium private valuations broadly assume the AI Monetisation Supercycle, where enterprise agents scale, inference efficiency lifts margins and infrastructure stays highly utilised. Growth Normalisation can still produce exceptional operational growth while disappointing shareholders if the entry multiple assumes too much duration.
The AI Price War is primarily a margin scenario rather than a demand collapse. Infrastructure Bottleneck shifts value towards scarce power and contracted capacity. Public-Market Repricing is the macro case, where higher real yields or a higher equity risk premium compress the present value of distant AI cash flows.
Across six months, the focus is filing progression, revenue velocity and pricing. Over 12 to 24 months, gross margin, enterprise mix and cash burn become decisive. Over three to five years, sustainable free cash flow and returns on deployed compute dominate.
What Markets May Be Mispricing
The first potential mispricing is assuming exceptional AI revenue growth automatically deserves SaaS economics. OpenAI may continue to compound revenue rapidly while requiring far more physical capital than a conventional software platform.
The second is the opposite error. Capital intensity does not automatically destroy value. If utilisation remains high, cost per task falls rapidly and the platform earns persistent economic rents, infrastructure investment can create formidable scale advantages.
Third, capital access may itself be a moat. A company able to raise $122 billion of committed financing and participate in a $500 billion infrastructure programme can survive competitive cycles that smaller laboratories cannot.
Fourth, cheaper inference is not automatically bullish for margins. If pricing falls as fast as or faster than unit cost, customers rather than shareholders capture the efficiency gain.
Finally, AI concentration extends beyond technology-sector labels. Utilities, industrial equipment, private credit and indices can all carry the same underlying AI-capex factor.
What Investors Should Monitor Next
The first disclosure to watch is revenue quality. Monthly and annualised revenue should be considered alongside enterprise mix, subscription growth, API pricing and agent monetisation. Rising embedded enterprise usage would strengthen the case that the revenue base is becoming more durable.
The second is gross margin and underlying cash burn. Public investors need evidence that inference efficiency is improving faster than price compression and that operating losses are growing much more slowly than revenue.
Third is infrastructure execution. Compute capacity, GPU availability, power delivery, data-centre timing and long-term purchase commitments will show how much capital is required to sustain growth.
Fourth is the IPO filing itself. Offering size, governance rights, related-party relationships, cloud commitments, Foundation control and any disclosed path to profitability could materially change the valuation debate.
External indicators include Anthropic revenue growth, hyperscaler capex, Nvidia data-centre demand, AI infrastructure credit spreads and technology equity multiples. Together, they show whether OpenAI’s commercial success is converting into sustainable returns through the supply chain.
This IPO Could Put The Entire AI Capital Cycle On Trial
OpenAI’s reported $40 billion-plus revenue run-rate proves that frontier AI can monetise at a scale few companies have ever reached so quickly. It does not yet prove that frontier AI will generate the free-cash-flow profile implied by an $852 billion private valuation or a potential $1 trillion public value.
The company may ultimately combine falling inference costs, deep enterprise adoption, agent monetisation and high infrastructure utilisation into substantial operating leverage. It may also discover that competition continuously transfers technical efficiency to customers while capital requirements remain exceptional.
A public listing would force those narratives into the same financial statement. Gross margin, cash burn, purchase commitments, governance and capital intensity would become visible alongside revenue growth.
For Bancara’s sophisticated private and institutional audience, the significance is cross-asset. OpenAI’s economics now touch semiconductors, cloud platforms, private AI valuations, data centres, electricity, credit and eventually passive index exposure.
The prospective IPO is therefore not merely a flotation of an AI company.
It could become a public referendum on the economics of the global AI capital cycle, and on whether markets can finally distinguish AI adoption from AI profitability.
Works Cited
- https://www.bloomberg.com/news/articles/2026-08-13/openai-s-revenue-run-rate-tops-40-billion-ahead-of-ipo?srnd=homepage-americas
- https://openai.com/index/scaling-ai-for-everyone/
- https://openai.com/index/announcing-the-stargate-project/
- https://openai.com/index/five-new-stargate-sites/
- https://openai.com/index/stargate-advances-with-partnership-with-oracle/
- https://openai.com/index/next-phase-of-microsoft-partnership/
- https://openai.com/index/advancing-the-price-performance-frontier-with-gpt-5-6/
- https://www.reuters.com/technology/openai-tops-25-billion-annualized-revenue-last-month-information-reports-2026-03-05/
- https://www.reuters.com/technology/openai-sees-compute-spend-around-600-billion-by-2030-cnbc-reports-2026-02-20/
- https://www.ft.com/content/e15b0d7e-ff6b-4f16-ba7a-4068feddb828
- https://www.anthropic.com/news/series-h
- https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-Financial-Results-for-Fourth-Quarter-and-Fiscal-2026/
- https://investor.oracle.com/investor-news/news-details/2026/Oracle-Announces-Record-Q4-and-FY-2026-Results-Driven-by-Cloud-Infrastructure–Cloud-Applications/
- https://www.iea.org/reports/energy-and-ai/executive-summary
- https://www.renaissancecapital.com/IPO-Center/Stats/Proceeds