Intel at the Crossroads: Equity Capital, Foundry Recovery, and the AI Infrastructure Test
Intel's historic $20 billion equity raise and early signs of packaging gains position the company for a role in the AI buildout, but its foundry model remains unproven at scale and its data center roadmap trails competitors.
When Microsoft formally accepted IREN's first AI data center facility this week, triggering the opening drawdown of a $9.7 billion contract, it sent a clear signal: hyperscaler commitment to physical AI infrastructure remains structurally in place, even as financial markets periodically revisit the sustainability of that spending. That commitment flows back up the supply chain — to power, cooling, real estate, and ultimately to chips and the fabs that make them. Intel finds itself at an unusually complex position in this moment: simultaneously a chip designer trying to reclaim competitive relevance, a foundry looking to win business that TSMC has long dominated, and a company whose balance sheet is being fortified — reportedly for the first time since 1971 — through a major public equity raise.
That offering priced at $95 per share and grew from an initial $15 billion to a final $20 billion, according to reporting from August 17. The scale matters: it gives Intel runway to fund both its manufacturing buildout and the research cycles needed to compete in AI accelerator design. The market has absorbed a related data point with notable interest — Nvidia's strategic stake in Intel, which cost approximately $5 billion, is reported by TechSpot to now be worth close to $25 billion. If accurate, that figure captures the gap between what Nvidia paid for foundry optionality and what the market now assigns to Intel's manufacturing future. That Druckenmiller's fund has moved into Intel alongside Broadcom and Arm, per a Pluang report from August 17, suggests that institutional capital is positioning for an AI semiconductor landscape broader than any single company — and that Intel's foundry story carries credibility with at least some sophisticated allocators. Goldman Sachs, according to a Vinanet report, is facilitating compute financing for both Nvidia and Intel, another signal of Intel's presence in the large-scale AI infrastructure conversation.
Intel's own capital expenditure history is the most concrete data point for evaluating that story. Annual capex peaked at $25.75 billion in fiscal 2023 before contracting to $23.94 billion in 2024, then declining more sharply to $14.65 billion in 2025 — a reduction of roughly 43% from the peak. Through the first half of 2026, Intel has spent $6.19 billion, annualizing to approximately $12.4 billion. The pullback reflects a deliberate recalibration after foundry customer wins came more slowly than initially projected. Whether the investment is beginning to pay off is visible at the margins. TechPowerUp reported this week that HBM shipments to Intel's Malaysia facility — internally designated Project Pelican — are surging, consistent with a meaningful production ramp. Wccftech separately reported that Intel may be gaining ground on TSMC's advanced packaging business, noting that surging AI demand has overwhelmed CoWoS capacity at the world's dominant contract chipmaker — a structural opening that Intel appears to be exploiting.
The headwinds are real and proximate. Intel shares fell alongside SanDisk and SK Hynix in a mid-August selloff — shares in the group plunged as much as 8%, per NDTV Profit — as investors questioned whether hyperscaler AI capital expenditure would moderate. When Google's Marvell custom chip deal emerged on August 21, Intel slipped again, as 24/7 Wall St. noted, underscoring the structural threat from ASIC customization: for inference workloads, hyperscalers increasingly prefer bespoke silicon they can control, a dynamic that favors designers like Marvell, Broadcom, and in-house teams over general-purpose CPU manufacturers. Intel's confirmation this week that Nova Lake — its next-generation architecture — will launch on consumer desktop before data center configurations compounds that concern: it signals the data center roadmap remains sequenced behind a consumer refresh cycle, an ordering that won't reassure enterprise customers weighing x86 inference options. A Forbes analysis from August 19 offered perhaps the most uncomfortable framing: export controls on Nvidia, AMD, and Intel cause damage far exceeding direct revenue losses in restricted markets, pointing to relationship erosion and R&D distortions that accumulate quietly over time.
The honest read on Intel at this juncture is that the company has capital, visible strategic alignment — the Nvidia stake is a concrete signal — and early evidence of manufacturing recovery, but has not yet demonstrated that these inputs translate into the foundry wins and data center design cycles that would make the equity raise appear prescient in retrospect. The IREN-Microsoft deal is a reminder of how large the addressable market for AI infrastructure has become; Intel's challenge is converting proximity to that opportunity into revenue. Three signals warrant close attention: whether Project Pelican's HBM ramp produces confirmed, named customer announcements beyond early sampling; whether Intel pulls forward the Nova Lake data center timeline to establish an x86 presence in the AI inference generation; and whether Intel's 2027 capital expenditure guidance — expected in early 2027 — signals renewed expansion or further moderation, which would imply the foundry model is still not covering its costs. Until those data points arrive, Intel's role in the AI infrastructure cycle is one of positioning, not yet of performance.