NextFin News - OpenAI's head of data centers, Chris Malone, is leaving the ChatGPT maker, the latest senior departure in a year of executive turnover as the company pushes its initial public offering into 2027 and wrestles with a costly pivot in how it builds AI compute capacity. Malone, who joined OpenAI from Meta in March 2025 to run the design, construction, and operation of the company's data center fleet, exits as OpenAI scales back its own build-out plans in favor of renting more servers from cloud providers.
The Departure and the Wider Exodus
Malone's exit is not an isolated personnel change. It lands in the middle of a broader reshuffling of OpenAI's infrastructure leadership and a string of high-profile departures that have thinned the company's senior ranks through 2026. In April, three senior executives who helped launch the original Stargate data center initiative — Peter Hoeschele, Shamez Hemani, and Anuj Saharan — left or were preparing to depart to pursue a new venture together. Chief revenue officer Denise Dresser, chief operating officer Brad Lightcap, and applications CEO Fidji Simo have also exited the company this year, while marketing chief Kate Rouch stepped down to focus on cancer recovery and head of ethics Chloé Bakalar left in July, less than a year after joining.
The timing matters. OpenAI had been preparing for a public listing as early as the third or fourth quarter of 2026, with bankers and lawyers already hired. Now, according to people involved in the deliberations, the company is leaning toward holding off until 2027. Chief executive Sam Altman has reportedly told advisers that a valuation below $1 trillion is a "nonstarter" — a steep climb from the company's last private mark of $730 billion and the $852 billion post-money valuation of the $122 billion funding round that closed on March 31, 2026.
That round was anchored by Amazon at $50 billion, NVIDIA at $30 billion, and SoftBank at $30 billion, with continued participation from long-term partner Microsoft. OpenAI described the capital as committed funding to accelerate the next phase of AI, and it came with a warning shot for anyone expecting a near-term listing: the company's own messaging emphasized compute build-out over profitability timelines. For SoftBank, the delay carries a direct mark-to-market cost. The Japanese conglomerate, a $30 billion anchor investor in the March round, fell 13 percent in a single day on reports of the IPO delay — a reminder that the AI boom is underwritten by balance sheets that need exits, not just road maps.
Stargate's Strategy Pivot: From Self-Build to Cloud Rental
The infrastructure team Malone helped build is being reorganized around a fundamental change in how OpenAI plans to secure compute. After months of planning to construct its own data centers, the company is shifting toward renting more AI servers from cloud providers. In a leadership reshuffle reported in recent weeks, OpenAI divided its compute and infrastructure teams into three groups: those designing data centers, those managing commercial partnerships with cloud and chip companies, and those running the data centers OpenAI rents.
Sachin Katti, the former Intel executive who joined OpenAI in November 2025, now oversees all three Stargate groups.
"AI demand is moving fast, and our compute team has built a powerful ecosystem across clouds, chips, and infrastructure to meet it. The progress has been incredible, and I'm excited about the work ahead as we keep building the systems that bring AI breakthroughs to people everywhere."
Katti's statement accompanied a leadership reshuffle that moved the people who built Stargate into new reporting lines. Before his arrival, Malone and Peter Hoeschele oversaw data center groups and reported to president Greg Brockman. Under the new structure, Hoeschele took commercial relationships with cloud and chip companies, while Malone co-led the technical engineering and design team alongside Adrian Caulfield, a former Microsoft engineering leader. The reorganization is a direct response to the strategy readjustment: OpenAI recently dropped plans to expand its Abilene, Texas, Stargate site to 2 gigawatts, capping the facility at 1.2 gigawatts instead. "We considered expanding it further, but ultimately chose to put that additional capacity in other locations," the company said, pointing to more than half a dozen sites under development, including its Oracle-backed project in Wisconsin.
The shift also reflects a rapidly changing contract book. In late February 2026, OpenAI agreed to use 2 gigawatts of Amazon Web Services' Trainium chips, expanding its cloud contract with Amazon by $100 billion. Around the same time, the company said it planned to spend roughly $600 billion on compute over the next four years — a reduction from an earlier plan to spend $1.4 trillion over eight years. OpenAI remains a customer of Microsoft, Oracle, and CoreWeave, among others.
Malone had been one of the most visible advocates for the self-build approach. Speaking at the Yotta 2025 conference in Las Vegas, he said:
"We have lots of channels by which we deliver capacity. We think it's important for us to also self-build. We have a very strong conviction about scaling, and also have direct buy-in from our leadership."
That conviction is now being tested as the company recalibrates toward rented capacity.
The economics explain the reversal. Building your own data centers concentrates capital expenditure and execution risk on OpenAI's balance sheet, but it lowers the marginal cost of compute over the asset's life. Renting from cloud providers does the opposite: it preserves cash today while locking in higher per-unit costs tomorrow. For a company projecting losses of roughly $14 billion in 2026 alone, the choice is not theoretical. The $600 billion four-year compute plan is still enormous by any standard, but the shift from self-build to rental is the difference between owning the factory and leasing the machines — and it changes who captures the margin on the next generation of AI infrastructure.
The IPO Math Behind the Delay
The personnel churn and strategy pivot are unfolding against a backdrop of financial pressure that makes the 2027 delay easier to understand. OpenAI's annualized revenue rate exited 2025 above $20 billion, chief financial officer Sarah Friar confirmed in a January 2026 blog post, up from $6 billion in 2024 and $2 billion in 2023. But growth has come at a steep price. Gross margins fell to 33 percent in 2025 as inference costs roughly quadrupled year over year to about $8.4 billion, and inference spending is projected to reach $14.1 billion in 2026. Internal forecasts cited in reporting show a net loss of roughly $14 billion for 2026, with cumulative losses through 2028 projected around $44 billion.
There is also the matter of existing obligations. Under the renegotiated Microsoft partnership finalized in April 2026, OpenAI continues to pay Microsoft a revenue share through 2030, now subject to an overall cap of $38 billion — down from a prior projected trajectory of roughly $135 billion. Even with the cap, the arrangement requires approximately $6 billion in payments in 2026. Independent estimates put OpenAI's cash burn at roughly $27 billion in 2026 and $63 billion in 2027, with the company not expected to turn cash-flow positive until 2030.
The valuation target compounds the challenge. Altman's $1 trillion floor would value OpenAI at roughly 40 times its 2025 revenue run-rate — a multiple that public-market investors will scrutinize against losses that show no sign of narrowing soon. The caution is not unfounded. SpaceX's initial public offering in July 2026 was the largest ever, raising more than $85 billion and reaching a debut valuation of $1.77 trillion, but its shares have since slumped to $153 from a high of $202. That post-IPO slide has made advisers to OpenAI more cautious about timing.
There is another wrinkle in the IPO math. The $122 billion raised in March was not all traditional equity. A substantial portion is structured as committed capital tied to infrastructure and vendor arrangements — Amazon's $50 billion anchor, for example, is closely linked to AWS capacity commitments, and SoftBank's $30 billion is tranched across 2026 rather than wired in a single payment. That structure buys runway, but it does not buy the same flexibility as unrestricted equity. When a company's largest investors are also its suppliers, the incentives around pricing, capacity, and timing become entangled in ways that a public shareholder base may not welcome.
What It Means: Cyclical Churn or Structural Stress?
The question investors should ask is whether this is ordinary Silicon Valley turnover or a signal of deeper strain. Some of the exits are clearly personal: Rouch left for cancer recovery, and Simo cited a severe exacerbation of a chronic illness. Others follow the familiar pattern of senior engineers spinning out to start new companies, as the Stargate trio did.
But the concentration of departures in infrastructure and go-to-market roles — the very functions needed to execute a Stargate-style build-out and to prepare a company for public markets — points to something more structural. OpenAI is attempting to do three difficult things at once: transition its compute strategy from self-build to cloud rental, manage a capital stack that includes tens of billions in annual obligations, and reposition for a public listing at a valuation that assumes continued hypergrowth. Each of those tasks requires experienced lieutenants, and the company is losing them in the middle of the transition.
The strongest counter-thesis is that OpenAI has more than enough talent and capital to absorb these losses. The company still commands the industry's most coveted AI models, it has deep relationships with every major cloud provider, and a $122 billion funding round at an $852 billion post-money valuation proves that sophisticated investors are willing to back the strategy for years. Under this view, Malone's departure is a personnel event, not a strategic warning — and the IPO delay is prudent timing rather than a retreat.
That case has merit, but it depends on one assumption: that OpenAI can keep scaling revenue fast enough to outrun its cost structure before public-market discipline sets in. The falsifying signal is concrete. If OpenAI's annualized revenue run-rate does not accelerate meaningfully above the $20 billion exit-2025 level through 2026 while compute spending remains near the $600 billion four-year plan, the growth-at-any-cost model comes under pressure that no private funding round can fully insulate. Reports of a five-month plateau in the run-rate near $25 billion through mid-2026 would, if sustained, be the first data point pointing in that direction.
The second-order implication cuts across the AI industry, not just OpenAI. If the leading lab concludes that renting cloud capacity is cheaper and less risky than building its own data centers, the signal to the market is that the hyperscalers — Amazon, Microsoft, Oracle, and CoreWeave — retain the structural advantage in the AI value chain. OpenAI's models may be the product consumers see, but the economics of the pivot say the infrastructure owners still hold the pricing power. That is the real story behind the headline: the company that once planned to own the factory is now choosing to rent the machines, and the landlords are collecting.
What Comes Next
In the short term, expect more organizational churn as the new infrastructure structure under Katti beds in and as OpenAI decides whether to file confidentially for an IPO before the end of 2026 or wait until 2027. The company's data center partners — Oracle, Microsoft, CoreWeave, and now Amazon with its Trainium deployment — are the near-term beneficiaries of the strategy pivot, since rented capacity shifts capital expenditure from OpenAI's balance sheet to theirs.
Over the medium term, the key data points to watch are the 2026 revenue trajectory, the gross-margin path as inference costs scale, and the pace of cash burn against the roughly $27 billion projected for the year. A successful GPT-6 launch in late 2026 could re-accelerate demand and support the trillion-dollar narrative; a stumble would make the 2027 delay look like the beginning of a longer wait.
In the long run, the structural question is whether AI infrastructure becomes a rented utility or a self-built moat. OpenAI's answer to that question will shape not just its own margins but the bargaining power of the cloud providers that supply it.
Scenarios for the IPO
Three paths are now in play. In the base case, OpenAI files confidentially in late 2026 and lists in 2027 at a valuation between $852 billion and $1 trillion, contingent on revenue re-acceleration and a stable post-SpaceX market for mega-cap tech debuts. In the upside case, a strong GPT-6 cycle and improving gross margins let the company price above $1 trillion, vindicating Altman's "nonstarter" stance. In the downside case, the run-rate plateau persists, losses widen toward the $63 billion 2027 projection, and the company is forced to choose between a lower valuation and a longer wait — with SoftBank and other tranched investors applying pressure for an exit.
The central judgment: Malone's departure is less about one executive than about a company trying to change its compute strategy, its capital structure, and its public-market timing all at once — and losing key people while doing it. OpenAI can survive the turnover if revenue keeps compounding. If it doesn't, the 2027 IPO delay will look less like patience and more like a delay of the inevitable.
Explore more exclusive insights at nextfin.ai.
