Facing $14B losses in 2026, OpenAI is now seeking $100B in funding. But can it ever turn a profit?

Nvidia, Microsoft, and Amazon are reportedly in talks to inject as much as $60 billion into OpenAI as part of a potential up to $100 billion funding round, according to The Information. The deal could be among the largest private capital raises ever attempted, valuing OpenAI at up to $830 billion, according to the Wall Street Journal, more than the GDP of Argentina. WSJ had a separate story on SoftBank being in talks to invest up to $30 billion. The company is also pursuing around $50 billion from Abu Dhabi sovereign wealth funds and others, TechCrunch reported.

The burn rate is among the fastest of any startup in history, with internal projections showing $14 billion in losses for 2026 alone, according to an analysis of OpenAI’s financial documents. The company expects cumulative losses of $115 billion through 2029 before reaching profitability sometime in the 2030s. For comparison: the Manhattan Project cost roughly $30 billion in today’s dollars; Apollo, $288 billion over 13 years.

Funding has lagged behind the spending. The March 2025 SoftBank-led $40 billion Series F, at a $300 billion valuation, remains the largest private funding round ever. An October 2025 secondary sale valued the company at $500 billion. The company also maintains a $4 billion revolving credit facility arranged by JPMorgan that remains largely undrawn.

Despite hitting $12 billion annualized revenue in July 2025, a milestone that took Google seven years and Facebook six, as SaaStr noted, and surpassing $20 billion by year-end, according to CFO Sarah Friar, the company must sustain unprecedented growth to hit its internal target of roughly $100 billion in annual revenue by 2029. In December 2025, Altman declared a “Code Red“, delaying advertising, shopping agents, and other revenue initiatives to marshal resources toward improving ChatGPT after Google’s Gemini 3 topped it on key benchmarks and enterprise market share fell to 27% while Anthropic rose to 40%, according to Menlo Ventures.

Stargate’s $500 billion promise versus $52 billion reality

The Stargate infrastructure project, announced at a White House event in January 2025 with a headline commitment of $500 billion over four years, exemplifies the gap between OpenAI’s announcements and confirmed capital. Actual committed equity totals $52 billion: $19 billion each from SoftBank and OpenAI, $7 billion each from Oracle and MGX (Abu Dhabi), according to The Information. The remaining $448 billion must come from debt financing not yet secured, and progress has been slow. SoftBank’s CFO conceded the project was “taking longer than anticipated,” Bloomberg reported in August 2025.

Oracle’s credit profile has deteriorated under Stargate’s weight, Barclays downgraded the stock to “underweight” and warned the credit rating could approach junk status if spending continues outpacing revenue conversion.

The contagion is spreading. On January 29, Microsoft’s stock plunged 12%, erasing $440 billion in market value, after disclosing that 45% of its $625 billion cloud backlog is tied to OpenAI. The “circular” economics of AI, where Microsoft invests in OpenAI, which commits to buy Microsoft cloud services, is prompting more questions from investors as the AI wave continues.

The circular financing problem runs deeper than Microsoft. Nvidia has committed up to $100 billion to OpenAI, money that, as OpenAI’s CFO Sarah Friar acknowledged, “will go back to Nvidia” in GPU purchases. Nvidia is a prominent investor in CoreWeave, which supplies cloud capacity to OpenAI and has spent billions buying Nvidia chips.

The competitive moat is eroding faster than revenue is growing

Similarweb’s Global AI Tracker shows ChatGPT’s web traffic share fell from 86.7% in January 2025 to 64.5% in January 2026, a 22-percentage-point dip in 12 months. Google Gemini captured much of the loss, growing from 5.7% to 21.5%. The January 2026 Apple partnership, integrating Gemini into Apple Intelligence and future Siri upgrades, accelerates this shift; analysts estimate the deal at $5 billion.

Anthropic presents another competitive threat in enterprise markets. The company disclosed that revenue grew from $1 billion ARR at end-2024 to more than $5 billion by August 2025. TechCrunch reported projections of $20–26 billion for 2026. The company raised $13 billion in September 2025 at a $183 billion valuation.

The talent war compounds OpenAI’s competitive pressure. Fortune reported that engineers at OpenAI are eight times more likely to leave for Anthropic than the reverse. CTO Mira Murati and cofounder Ilya Sutskever are among a wave of senior departures, including cofounders, research leads, and alignment researchers, many of whom have left to start competitors or join Anthropic. The Wall Street Journal reported that OpenAI expects to spend $6 billion on stock-based compensation in 2025, nearly half of projected revenue, to stanch the outflow.

Meanwhile, Chinese competitors pose an existential pricing challenge. DeepSeek V3.2, released December 2025, matches GPT-5-level performance on elite reasoning benchmarks, scoring 96.0% on AIME 2025 versus GPT-5-High’s 94.6%. Inference costs run 10–30x cheaper: DeepSeek at $0.21–$0.28 per million input tokens versus OpenAI’s $2.50+. Kimi K2.5 from Moonshot AI, released January 27, 2026, is a 1-trillion-parameter open-source model that beats GPT-5.2 on tool-augmented reasoning while costing roughly 1/20th as much.

User sentiment dips and Musk lawsuit wildcard

OpenAI’s consumer-facing models have swung wildly on personality. One variant of GPT-4o was slammed as a sycophantic “yes man” that endorsed harmful ideas, forcing rollbacks in April 2025. GPT-5’s August 2025 launch overcorrected into “cold” and “robotic” territory, sparking social media backlash. Iterations through GPT-5.2 drew complaints of degraded writing. At the January 27, 2026 developer town hall, Altman admitted: “I think we just screwed that up” on writing quality.

The Musk lawsuit, set for trial starting March 30, 2026, has produced damaging discovery. Unsealed Greg Brockman diary entries from November 2017 show the co-founder writing: “Cannot say that we are committed to the non-profit” and “it’d be wrong to steal the non-profit from [Musk]. To convert to a b-corp without him. That’d be pretty morally bankrupt.” The judge ruled there are “substantial grounds” for believing Musk was misled. OpenAI values its exposure at roughly $38 million (Musk’s donation amount); Musk seeks up to $134 billion in wrongful gains plus punitive damages.

One estimate pegs the settlement probability at 50-60%, with potential damages ranging from Musk’s $38 million donation to tens of billions if a jury finds OpenAI’s valuation was built on fraud.

The bull case requires everything to go right

OpenAI’s path to profitability hinges on premium pricing for advanced AI agents. The company has shifted toward usage-based and outcome pricing (per-task or revenue share), with tools like Codex (20x growth since August 2025) and domain-specific agents as priorities. SoftBank committed $3 billion annually to deploy OpenAI tech, including agent products like Operator and Deep Research.

Scientific breakthrough monetization offers the biggest upside. CFO Sarah Friar has discussed “value-sharing” arrangements, taking a percentage of profits when clients generate breakthroughs using OpenAI tech. Partnerships with Thermo Fisher Scientific, Sanofi, and Formation Bio position the company in what JPMorgan estimates as a $700+ billion AI-in-pharma market by 2030.

Enterprise adoption backs the bull case. OpenAI’s December 2025 report shows 92% of Fortune 500 companies using ChatGPT. Enterprise seats grew 9x year-over-year.

Meanwhile, commoditization looms. Stanford’s AI Index shows the cost to achieve GPT-3.5-equivalent performance dropped 280x, from $20 per million tokens in November 2022 to $0.07 by October 2024.

Potential scenarios

The bull case: “The AI operating system”

Agentic AI works: OpenAI starts executing labor, justifying enterprise pricing approaching hundreds of dollars per seat. Stargate scales to full capacity. Proprietary “reasoning data” from hundreds of millions of users creates an unassailable advantage. Revenue exceeds $100 billion by 2029, and the company justifies a valuation approaching Big Tech scale.

The base case: Premium brand, infrastructure constraints

OpenAI becomes the premium consumer brand but loses the backend infrastructure war to Azure, AWS and open-source alternatives. Stargate’s 10-gigawatt buildout faces the same grid constraints and permitting friction that could slow any hyperscale project.

The bear case: “The Netscape trajectory”

Models become commodities. Open-weight alternatives match state-of-the-art capabilities for essentially free, and inference margins collapse toward zero. Value accrues to the application layer or the chip layer, not the model provider. Revenue plateaus at $15–20 billion while burn continues. Microsoft, which already owns significant profit rights, absorbs the team in what amounts to a debt-to-equity conversion.

The tail risk: Regulatory crackdown or catastrophic failure

The precedents are accumulating: algorithmic trading halts, health-insurance AI denying claims at industrial scale, chatbot hallucinations wiping out $100 billion in market cap. California and New York have enacted frontier AI laws with million-dollar penalties; the EU AI Act takes full effect in August 2026. A sufficiently damaging incident could trigger federal “model licensing” requirements that freeze frontier deployment. OpenAI’s scale makes it the obvious target. Public trust offers no buffer: only 5% of Americans say they “trust AI a lot,” according to YouGov in December 2025.

The deeper risk: What if OpenAI bet on the wrong architecture?

OpenAI’s financial projections assume that scaling LLMs, the approach that built ChatGPT, will continue delivering returns. But two of the field’s most important voices now argue otherwise.

Ilya Sutskever, OpenAI’s former chief scientist, declared at NeurIPS 2024 that “pre-training as we know it will unquestionably end.” The internet’s text, “the fossil fuel of AI”, has been exhausted. “We have but one internet,” he said. “We’ve achieved peak data.”

Yann LeCun, Meta’s chief AI scientist and a Turing Award laureate, goes further: LLMs are fundamentally incapable of the reasoning, planning, and world-modeling that real intelligence requires. “I think the shelf life of the current paradigm is fairly short, probably three to five years,” he said at Davos in January 2025.

This creates an innovator’s dilemma for OpenAI. The company has committed over $100 billion to scaling LLMs, building Stargate’s data centers, securing compute contracts, hiring thousands of engineers optimized for the current paradigm. If the next breakthrough comes from world models, embodied AI, or some architecture not yet imagined, OpenAI’s infrastructure could become a stranded asset.