OpenAI Burned $3.7 Billion in Q1 and Filed for a $1 Trillion IPO in the Same Week

OpenAI made $5.7 billion in the first quarter of 2026. It spent $3.7 billion of that. Both numbers tripled year on year. The company then filed confidentially for an IPO that a source told Reuters could value it at up to $1 trillion. There is a version of this story where the revenue growth curve eventually crosses the cost curve and the whole thing makes sense. OpenAI itself has told investors it does not expect to be profitable until the end of the decade. For now, it holds $73 billion in cash. The question the IPO will force onto the table is a simple one: does the factory model for frontier AI eventually produce the same margins as software, or does it keep burning money in proportion to its revenue for the next five years? Nobody knows. The IPO will be the first time public market investors get to price their answer.

OpenAI Filed for an IPO at up to $1 Trillion While Burning $3.7 Billion in a Single Quarter

OpenAI confidentially filed for a US IPO on June 8, 2026, with a source telling Reuters the listing could come as early as September and value the company at up to $1 trillion. The financial data that arrived around the same time from The Information, based on documents shared with shareholders, showed Q1 2026 revenue of $5.7 billion and cash burn of $3.7 billion, or roughly 65% of revenue consumed before profit. Both figures tripled from Q1 2025. The operating loss for the quarter was $9.3 billion. The net loss was $21.3 billion, though that figure includes a $12.4 billion non-cash accounting charge that did not represent a direct cash outflow.

The $73 billion in cash on hand at the end of Q1 provides the context that prevents those numbers from being immediately alarming. At $3.7 billion per quarter, that is roughly five years of runway without raising another dollar. But OpenAI is not sitting still. It is spending aggressively on compute, model development, engineering, and the infrastructure required to deliver inference at a scale that 900 million weekly active users demand. The spending grows with the revenue because the product itself, frontier-model inference, gets more expensive to deliver as more people use it. That is the opposite of the software economics that underpin most technology company valuations. Traditional software has near-zero marginal cost per additional user. OpenAI does not.

The IPO timing is instructive. OpenAI does not need the money. It has $73 billion. What the IPO provides is a public market valuation that a confidential S-1 filing, submitted without public disclosure to the SEC before the roadshow, allows companies to test investor appetite while keeping financials private until closer to the listing date. The competitive dynamic with Anthropic, which filed earlier and is also targeting a public listing, is part of the calculus. Both companies are racing to establish their public market narrative before the other sets the benchmark. The company that lists first defines the comparable for the one that follows.

For smaller AI companies and investors watching what AI infrastructure economics look like at scale, the OpenAI financials are the most transparent data point available from the frontier. The burn-to-revenue ratio has not improved as the company has scaled. Whether it ever does is the central question of the AI investment thesis. Everything else, the model releases, the partnership announcements, the government contracts, is ultimately a bet on whether that ratio eventually flips.

Anthropic Is in Discussions to Run Claude on Microsoft’s Custom Maia 200 AI Chips

Anthropic is in early-stage discussions with Microsoft to run Claude inference workloads on Microsoft’s custom Maia 200 AI chips via Azure, according to CNBC. The Maia 200, launched in January 2026 on TSMC’s 3nm process, is specifically designed for inference workloads and claims more than 30% better performance per dollar than competing silicon. A deal would add Microsoft’s homegrown chip to Anthropic’s existing compute stack, which already spans Nvidia GPUs, AWS Trainium, and Google TPUs. Anthropic would become the first frontier AI lab to run production workloads on all four major cloud providers’ AI silicon simultaneously.

The compute diversification story is not primarily a technology story. It is a cost story. Inference, the process of running a trained model to respond to a user query, is where the bulk of frontier AI operating costs sit at production scale. Each inference request consumes GPU time, which costs money. If the Maia 200 delivers 30% better performance per dollar than Nvidia’s equivalent, and Anthropic routes a meaningful portion of its inference volume through it, the effect on unit economics is material. The OpenAI financials published this week illustrate how much that matters. If inference costs fell 30%, the $3.7 billion quarterly burn does not fall 30%, because compute is not the only expense. But it falls meaningfully.

For smaller AI companies building on foundation models via API, the Anthropic-Microsoft chip discussion matters in a different way. The cost of running inference on frontier models, which flows through to API pricing, is the primary input cost for most AI application companies. Any structural reduction in inference cost at the frontier level eventually flows downstream through competitive pricing pressure. The companies that have built thin margins on top of current API pricing are the most exposed when that happens. The ones with genuine value-add above the model layer are the ones that survive the compression.

Nvidia and SK Hynix Announced a Multiyear Memory Partnership and the Next AI Rack Will Cost $7.8 Million

Nvidia (Nasdaq: NVDA) and SK Hynix announced a multiyear technology partnership to co-develop next-generation memory aligned with Nvidia’s AI infrastructure roadmap, spanning Vera Rubin AI supercomputers, Vera CPUs, RTX Spark PCs, and Jetson Thor robotics platforms. SK Hynix will also use Nvidia tools to accelerate semiconductor simulation and build digital twins for autonomous fab operations. The deal addresses what Nvidia’s own roadmap now identifies as one of its primary supply constraints: extended development cycles for memory components at the scale and performance the next generation of AI infrastructure requires.

The Morgan Stanley figures published this week put the cost of the next Nvidia rack in context. The Vera Rubin-based VR200 NVL72 rack will cost hyperscale cloud providers approximately $7.8 million per unit, up from roughly $4 million for the prior GB300 generation. Memory now accounts for approximately 25% of total system cost, or about $2 million per rack, driven by a threefold increase in LPDDR5X content. Each Rubin GPU is priced at approximately $55,000 for volume hyperscaler purchases. This is the cost escalation that makes compute diversification, Anthropic running on Microsoft’s chips, and the Blackstone-Google TPU joint venture rational rather than experimental. At $7.8 million per rack, the total cost of building an AI training cluster is not something that keeps getting cheaper just by waiting.

The memory supply chain story runs through the same logic that makes the Nvidia-SK Hynix deal significant for smaller players. AI infrastructure spending is concentrating around a small number of suppliers at each layer of the stack: Nvidia for GPUs, SK Hynix and Micron for high-bandwidth memory, TSMC for advanced packaging and process nodes. Companies trying to build AI infrastructure at scale without the purchasing power of a hyperscaler are paying the same unit prices while absorbing the same supply constraints. The democratization narrative for AI compute is real at the model API layer. At the infrastructure layer, concentration is accelerating.

The US Government Put $500 Million Behind SandboxAQ to Find Alternatives to PFAS and Rare Earths in Chip Manufacturing

The US government committed $500 million to SandboxAQ, a company backed by Nvidia, to boost domestic chip supply chains. The specific focus is finding alternative materials for semiconductor manufacturing, particularly replacements for PFAS, the class of persistent chemical compounds used extensively in chip fabrication, and rare earths used in manufacturing equipment. SandboxAQ CEO Jack Hidary told Reuters that the company’s work could let manufacturers pick alternative chemicals and destroy PFAS at their own plants.

The PFAS problem in semiconductor manufacturing is not widely understood outside the industry. The production of advanced logic chips requires dozens of fluorinated compounds that are persistent in the environment and human tissue, and increasingly subject to regulatory restriction in both the US and EU. The cost of managing PFAS disposal and the regulatory risk of non-compliance is a growing line item for TSMC, Samsung, Intel, and every foundry trying to scale. A domestically developed alternative that performs comparably at scale could reduce both the regulatory and supply chain exposure for advanced semiconductor manufacturing.

For smaller companies working on materials science, chemistry, and quantum sensing applications relevant to semiconductor manufacturing, the SandboxAQ investment is a signal about where the government is directing capital at the intersection of national security and chip supply chain independence. The AI infrastructure buildout is creating demand for solutions up and down the manufacturing stack, not just at the chip design layer. The companies building those solutions are operating in a market where the customer is federal government procurement, which has different timelines and different qualification requirements than commercial markets, but which also does not disappear in a down market cycle.

OpenAI September IPO Timeline, Nvidia June 24 Annual Meeting, EU AI Act August 2, and Anthropic S-1 to Watch

OpenAI IPO: confidential filing submitted June 8. If the September timeline holds, the public S-1 would become available in July or August, roughly 21 days before the roadshow begins. That document will be the most scrutinized financial disclosure in the history of the AI industry. Watch for any announcement of underwriter selection or S-1 public filing date, which would start the clock.

Nvidia June 24 annual meeting: Nvidia’s 2026 Annual Meeting of Stockholders is online on June 24. No significant product announcements are expected but management commentary on the Vera Rubin demand pipeline, the SK Hynix memory partnership timeline, and any guidance update following the Broadcom-driven semiconductor selloff will be watched closely.

EU AI Act August 2 enforcement: 44 days away. High-risk AI system requirements come into force with fines up to 35 million euros or 7% of global revenue. Watch for any major AI company announcing compliance status changes, product modifications for EU markets, or the first enforcement action from the European AI Office after the deadline passes.

Anthropic S-1: Anthropic filed confidentially before OpenAI and is presumably further along in the process. Any announcement of a public filing date would reset the AI IPO race narrative. The company that lists first defines the comparable for the second. The Anthropic S-1 will need to address the SpaceX compute deal, the $65 billion Series H valuation, and the safety warning it issued this month in the context of a $965 billion pre-IPO valuation. Those are not easy things to reconcile in a prospectus.

Sources

Editorial Disclosure

This analysis is based entirely on publicly available information including press releases, SEC filings, and verified news sources. Securities discussed include NVIDIA Corporation (Nasdaq: NVDA) and Microsoft Corporation (Nasdaq: MSFT). OpenAI is a private company referenced for its IPO filing and Q1 2026 financial data sourced to The Information via named published reports, not independently verified. Anthropic PBC is a private company referenced for its compute discussions. SandboxAQ is a private company referenced for its government funding announcement. SK Hynix (KRX: 000660) is referenced for its Nvidia partnership announcement. Gilead Sciences Inc. (Nasdaq: GILD) is referenced in the inference cost context only. aktiego.com has not received any compensation from any company mentioned, their management, investor relations representatives, or any third party. No staff member or principal of aktiego.com holds a position in any security mentioned at the time of publication. OpenAI financial data sourced to named published reports from The Information and cited accordingly. Forward-looking commentary regarding IPO timelines, regulatory enforcement, product announcements, and market developments is opinion only. References to individual companies are for market context and analytical purposes only and do not constitute investment recommendations. All securities carry significant investment risk including total loss of capital. Coverage on aktiego.com is provided for informational and educational purposes only. aktiego.com is not a registered investment advisor. Nothing in this article constitutes financial, investment, or professional advice. Readers are encouraged to conduct their own due diligence and consult a qualified financial advisor before making any investment decisions. For more information please see our full DISCLAIMER.

OpenAI Burned $3.7 Billion in Q1 and Filed for a $1 Trillion IPO in the Same Week

Join our Mailing List