Revisiting Intel After the TD Cowen Lift and CPU Demand Signal

Server CPU demand has been the quiet surprise of 2026 so far. The AI build-out has driven GPU headlines, but the CPU attach rate to AI server configurations remains meaningful, and the refresh cycle from the late-2020 and 2021 Xeon deployments is now overdue. Enterprises that held through the pandemic capex are starting to budget for the replacement, and Intel is the largest beneficiary of that refresh by installed base.

The AMD EPYC platform continues to gain share in the hyperscaler segment, and that trend is not reversing. But the enterprise segment that Intel has historically dominated is still a $45 billion market, and the share defence there has been better than feared. That is the nuance behind the TD Cowen lift, and it is the data point that the sell-side is starting to coalesce around.

The TD Cowen upgrade is constructive on demand, but the firm kept a Hold rating on the stock. That is the correct posture. The CPU demand recovery is now a reasonably confident base case. The foundry story is still binary.

Intel 18A is the node that matters. First production capacity is coming online this year, and the external customer pipeline has been building slowly. Microsoft has been the anchor customer. The question is whether the second major customer arrives on time, and whether the yield curve on 18A is close enough to TSMC's equivalent node to be competitive on cost.

If 18A delivers, the foundry business is worth $30 to $50 billion of enterprise value on its own. If it stalls, the current share price is close to fair value with no additional upside. That is the asymmetry that makes Intel interesting, and it is why the signal on foundry execution matters far more than the signal on CPU demand.

Our revised 2026 revenue estimate moves to $58 billion, a 4% lift from the prior model and in line with the consensus that has been drifting higher since the March results. Gross margin forecasts are tighter, with the base case landing at 38% and the bull case at 42%. Operating income comes in between $3.5 billion and $5.5 billion depending on the foundry trajectory.

The balance sheet carries roughly $40 billion of net debt, down from the peak but still meaningful. The dividend has been suspended, which removed a constituency of long-term holders but also unlocked roughly $3 billion of annual cash flow for reinvestment. That optionality has value in a capex-heavy business.

Earnings per share in the base case is $0.55 to $0.70. The stock at $24 is trading at roughly 35x on the base case and 45x on the bear case. That multiple is defensible only if the 18A story creates a path to materially higher earnings by 2028.

AMD continues to drive share gains in server and desktop. Nvidia is the generational winner in AI compute. TSMC is the benchmark for foundry execution. Intel is trying to compete across all three fronts with a balance sheet that is tighter than its competitors and a product roadmap that has slipped twice in three years.

The CPU demand recovery helps Intel more than AMD in absolute dollar terms because the enterprise segment is more installed-base driven than the hyperscaler segment. AMD's share gains are real but they are happening in the segment that is growing faster, while Intel is holding serve in the segment that is growing slower. That is a survivable dynamic but not a thrilling one.

By comparison, TSMC's 2nm node is scheduled for high-volume production in the second half of 2026. Intel 18A needs to be competitive with TSMC N2 on both performance and cost. The window for that comparison is now 12 to 15 months away.