OpenAI raised $122B. My reflections on when capital becomes a “strategic weapon”

OpenAI raised $122B this week at an $852B post-money valuation. The round included $50B from Amazon, $30B from Nvidia, $30B from SoftBank, an undisclosed amount from Microsoft, and an additional $12B from a mix of institutional and, for the first time, retail investors through bank channels. The company is generating $2B in monthly revenue. It has 900M weekly active users. It is not profitable, does not expect to be until 2029 at the earliest, and projects $665B in cumulative cash burn through the end of the decade.

$122B raised. $665B still to burn. That gap is worth sitting with. My reflections below…

Travis Kalanick (TK) described capital as a strategic weapon during Uber’s growth years. I’ve been enamored by this concept; especially after I heard it in his latest set of podcast interviews post Atoms announcement. His framing was specific. In markets with network effects and winner-take-most dynamics, the ability to raise and deploy capital faster than your competitor is itself a form of competitive advantage. Capital funds the subsidies that acquire users. Users create network density. Network density creates defensibility. The capital is not incidental to the strategy… It is the strategy.

TK systematized this at Uber. 4 rooms running in parallel in the NYC office, 12 hours a day, 90 min slots, tiered by check size. The $250M-and-over room was his. The fundraise was run like a sales ops. Capital was the product being manufactured, and deploying it faster than Lyft, Didi, or anyone else was the mechanism through which market position was created.

OpenAI’s $122B round is the TK doctrine operating at a whole new scale. The company that can deploy the most compute wins the model quality race. The company that wins the model quality race captures the most users. The most users generate the most data and revenue. The revenue justifies the next raise. The raise funds the next round of compute, and therefore optimizes for terminal value maximization in perpetuity. This is not product-market fit in the traditional sense. It is capital-market fit: the ability to raise faster, deploy faster, and convert capital into competitive position before anyone else can respond.

In most private market contexts, capital efficiency is the discipline. Revenue per dollar of invested capital, the relationship between gross margin and operating margin, the gap between EBITDA and free cash flow. The goal is to demonstrate that the business generates more value than the capital it consumes.

In the “capital-as-weapon model,” the goal inverts. The business consumes capital as its primary competitive activity. Profitability is deferred because the act of spending itself can become the moat. The company that stops spending loses position. The company that keeps spending maintains it only if the next round arrives. The fundraise is an operating function.

If you manage a fund with a 10-year life and you’re evaluating participation in this round, the math works differently than anything your LPs signed up for.

You are buying at $852B post-money. For the valuation to produce a 3x return on invested capital (ROIC), the exit needs to happen at roughly $2.5T. For a 5x, you need $4.3T. These are numbers that currently belong to Apple, Microsoft, and Nvidia in the public markets. The implicit assumption is that OpenAI will become one of the 5 most valuable companies on earth, and it needs to do so within a timeframe that produces a net IRR above 20% for the fund to justify the allocation.

IRR is a time-weighted metric. A 3x over 3 years produces an IRR of ~44%. A 3x over 7 years produces an IRR of ~17%. The same multiple, delivered on a different timeline, tells a completely different story to an LP evaluating fund performance. The path to an attractive IRR requires not just a massive terminal valuation but a relatively fast one. The IPO (reportedly targeted for Q4 2026) is not incidental to the investment thesis. It is the investment thesis. Without a near-term liquidity event, the IRR math deteriorates rapidly regardless of TVPI.

TVPI, total value to paid-in capital, is the metric that flatters. It captures both realized distributions and unrealized marks. A fund that invested in OpenAI at $300B in early 2025 and marks it at $852B today can report a TVPI north of 2.5x on that position, even though no cash has been returned to LPs. The position looks excellent on paper. The GP walks into the next fundraise with a compelling TVPI and a marquee portfolio company. The LP, meanwhile, has received zero distributions.

DPI, distributions to paid-in capital, is the metric that ultimately matters. It measures actual cash returned. For every dollar the LP committed, how many dollars have come back? LPs across private markets are increasingly anchoring on DPI as the primary standard of fund performance. Median US buyout DPI for 2012-2015 vintages sits around 1.4-1.7x. Funds that arrive at re-up conversations with strong TVPI and weak DPI face harder questions than they used to.

The OpenAI round, for traditional venture and growth investors, is almost entirely a TVPI event. No plans to distribute capital. The exit pathway is the IPO, which may or may not price at or above the round’s implied valuation, on a timeline the company does not fully control. A public market correction, a competitive setback, or a slower-than-expected revenue ramp could push the IPO into 2027 or beyond, compressing IRR even if eventual TVPI is attractive. The fund that reports a 2.5x TVPI but returns capital in year 9 at a 12% net IRR will have a different conversation with its LPs than the fund that returns 1.8x in year 5 at a 25% net IRR.

The same LPs who allocate to venture and growth also allocate across the rest of private markets. When capital is concentrated in a small number of mega-positions like OpenAI, the LP’s portfolio becomes top-heavy in unrealized, long-duration bets. The demand for shorter-duration, cash-generative strategies that produce real DPI increases. The gravitational pull of AI capital is paradoxically creating a more disciplined LP appetite for everything else.

The structural feature of this round that deserves the most scrutiny is the circularity (going beyond the size of the round).

Amazon invested $50B. Amazon also operates AWS, the cloud infra provider that OpenAI has contracted to use as part of an 8-year, $100B usage agreement. A meaningful share of the capital Amazon deploys as an equity investment will return to Amazon as a cloud services customer. $35B of Amazon’s commitment is contingent on OpenAI going public or reaching AGI by end of 2028. That is not a standard or typical equity position… this is a structured instrument with conversion triggers.

Nvidia invested $30B, largely in computing capacity rather than cash. Nvidia manufactures the GPUs that OpenAI purchases for training and inference. The investment flows back to Nvidia through hardware procurement.

Microsoft has invested more than $13B historically and participated again in this round. Microsoft operates Azure, the primary cloud infra on which OpenAI runs. Microsoft holds profit-sharing rights in OpenAI. Revenue generated by OpenAI’s API flows partially through Azure, and a portion returns to Microsoft as hosting fees.

SoftBank invested $30B and is the financial lead of Stargate, the $500B data center joint venture. SoftBank reportedly took a $40B bridge loan to fund its commitment.

In each case, the investor is simultaneously a supplier, a customer, or a counterparty. The capital does not arrive on OpenAI’s balance sheet as unencumbered equity in the way a typical venture check does. It arrives within a web of commercial relationships where a significant portion of the invested capital will cycle back to the entity that deployed it.

This has a specific analogue in operating businesses: vendor financing. When a supplier funds a company’s purchase of the supplier’s own product, the economics look different from a pure equity investment even if the legal structure is identical. The capital is real. The commercial relationship is real. The circularity does not make either party’s position fraudulent. It does mean that evaluating the investment on a standalone basis, without accounting for the offsetting commercial flows, will overstate the amount of freely deployable capital the recipient actually has.

OpenAI’s CFO, Sarah Friar, described the round as providing “a lot of flexibility.” That may be true relative to the alternative of not raising. It is less obviously true when a substantial portion of the $122B is pre-committed to spending with the entities that provided it.

There is a point in a company’s life where capital stops enabling the strategy and starts becoming the strategy. You can usually identify the moment in retrospect, because it is the point at which the fundraise begins consuming as much organizational energy and narrative bandwidth as the product itself.

Uber crossed that threshold somewhere around 2015. Rider subsidies, driver incentives, geographic expansion into markets that could not yet sustain unit-level profitability. Each decision was rational in the “capital-as-weapon” framework. Each also moved the company further from a business that could stand on its own economics. The question was always whether ride volume would persist at full price. In many markets, it did. Uber eventually reached profitability. The thesis was validated, though it took a decade and the destruction of several competitors to get there.

OpenAI crossed that threshold before it ever generated a dollar of revenue. The compute required to train frontier models is so capital-intensive that no revenue base, no matter how fast-growing, has been sufficient to fund the next generation of investment. The company projects $14B in operating losses in 2026. Internal projections show total cash burn of $665B through 2030, revised upward by $112B from earlier estimates. The $122B raise provides ~24 months of runway before the company needs to raise again, either through the IPO or additional private capital.

The 48% gross margin on inference is where the structural story lives. For every dollar of revenue, $0.52 goes to the direct cost of serving the query. R&D and infra spending push operating margins deeply negative. This is not a margin profile that improves purely through scale, because inference costs are variable. Every additional user consumes real compute. Efficiency gains from hardware and algorithmic improvements are real, and they do reduce the cost per unit of intelligence over time. They have not been sufficient to outpace the growth in absolute cost as usage scales. Inference costs quadrupled in 2025 while revenue tripled.

A business where variable costs grow as fast as revenue does not generate incremental operating leverage. The path to profitability requires either a structural change in the cost base, pricing power sufficient to expand gross margins, or a reduction in below-the-line spending as a percentage of revenue. OpenAI is betting on all 3 happening simultaneously by the end of the decade. That bet may prove correct. It is also a bet, not a demonstrated trajectory, and the difference between a projected margin expansion and a realized one is the difference between TVPI and DPI.

The first observation is about capital and conviction. The investors in this round are not making mistakes. Amazon, Nvidia, SoftBank, and Microsoft each have strategic reasons to deploy this capital that extend beyond financial returns. The investment is partially a cost of doing business in the AI ecosystem: a way to secure commercial relationships, infra access, and positioning in a market they believe will define the next 20 years. ROIC is only one dimension of the return they are seeking. When you understand the investment through that lens, the structure makes sense. It also means the round should not be evaluated as a pure financial signal about OpenAI’s standalone value.

The second is about TVPI and DPI as a philosophy (beyond the metric). There is a meaningful distinction between a business that creates value on paper and a business that generates cash. A company with a rising implied multiple and declining FCF is telling you something important. The AI sector, at this stage, is a TVPI story. The marks are rising. The multiples are expanding. The cash is not coming back (yet). Realized returns are the only returns that count, ultimately.

The last is about the question that every capital-as-weapon story eventually confronts: what happens when the weapon stops working? Uber’s answer was to cut costs, restructure operations, reduce driver incentives, and find profitability on the other side. That took a decade. OpenAI’s projected timeline to profitability is similar. Whether the company gets there depends on continued access to capital, continued technical leadership, and a competitive environment that permits margin expansion. Each of those conditions is plausible. None of them guaranteed.

Capital can be a weapon. It can also be a dependency. The businesses that compound over a full cycle are the ones that know, with precision, which one they are holding.

Ref link to OpenAI announcement.