The lazy question after a 9% day is whether the analyst is right about the target. You cannot answer that, and trying is how this kind of piece goes wrong: a price target is an opinion about a multiple five years out, and arguing with it is just swapping one opinion for another.
The question that can be answered is narrower and more useful. Northland's upgrade names a specific mechanism — Intel's role in Terafab will benefit Intel Foundry — and a mechanism is checkable. There is a line in Intel's accounts where a foundry customer shows up. So: how big is that line today, what is in it, and has Intel itself said anything in a filing about the customer everyone is pricing?
That turns out to be answerable in full, and the answer is more interesting than the rating.
What Northland actually said, and what moved
Northland's Gus Richard upgraded Intel to Outperform from Market Perform on September 8 with a $120 price target, citing three things: "material progress" on the turnaround, a server-CPU shortage working in Intel's favour, and Terafab — the partnership with SpaceX and Tesla will, in his words, "materially benefit Intel's foundry business and drive the needed scale of its process technology footprint."
That second clause is the more careful half of the claim, and it is worth separating from the first. Scale and utilisation for a leading-edge node is a different argument from foundry revenue, and it is the better one.
Intel closed at $104.47 on September 8, up 9.05% from the $95.80 close on September 4 — the prior session, with Labor Day between them. The target sits about 15% above that close.
Two things ran on the same tape, and the 9% belongs to both. The other was a report that Intel will raise PC CPU prices around 10% in October, which we wrote about separately on the day: Intel's CFO said in July it would still be behind in the fourth quarter, and a price rise into that is rationing, not margin defence. That is a client-CPU story and this is a foundry story. Do not let anyone hand the whole 9% to Terafab.
The line a Terafab contract would land in is $293 million, and Intel says most of the growth is Altera
Intel Foundry booked $5,765M of revenue in the June 2026 quarter against a segment operating loss of $(2,089)M. Almost none of that revenue is outside money. Intel's quarterly filing is explicit:
Intel Foundry also includes certain third-party foundry and assembly and test revenues from external customers that totaled $293 million in the three months ended June 27, 2026 … ($22 million in the three months ended June 28, 2025).
So the external business is $293M of a $5,765M segment — about 5%. The other 95% is internal wafer transfer to Intel's own product groups, which is why the segment barely moves consolidated revenue while carrying the whole capital programme and all of the depreciation.
Now the part that matters most, and it is Intel's own explanation rather than ours. From the same filing's discussion:
External revenue was $293 million in Q2 2026, up $271 million from Q2 2025 … primarily due to Altera's transition to an external customer following the deconsolidation of Altera in Q3 2025.
Read that carefully. Of the $293M, $271M of the year-over-year increase is Altera — a business Intel used to consolidate, now buying wafers from Intel as a third party. The same quarter a year earlier, before that reclassification, external customers paid Intel $22 million.
That $22M is the honest picture of Intel's third-party foundry business as a going concern a year ago. It is the base a Terafab contract would be added to.
Intel's own filing names no Terafab, and calls the 14A customers potential
Intel announced it was joining Terafab on April 7, 2026 — publicly, with a post on X carrying a photograph of chief executive Lip-Bu Tan and Elon Musk, and a statement that Intel's ability to "design, fabricate, and package ultra-high-performance chips at scale will help accelerate Terafab's aim to produce 1 TW/year of compute." The stock rose about 3% that day. Anyone describing Intel's role as new or unannounced this week is five months late.
What Intel has not done is put it in a filing. Terafab appears zero times in Intel's June-quarter report, filed on July 24 — three and a half months after the announcement. What the filing does say about the node Terafab would use is this:
During Q2 2026, we committed to completing development of Intel 14A … We also made continued progress towards meeting performance and design milestones for potential significant customers to evaluate Intel 14A for their future products. We intend to accelerate various of our manufacturing expansion projects, though the scale and pace of our manufacturing expansion projects will ultimately be dictated by the amount of committed demand for Intel 14A that we are able to obtain from our Intel products roadmap and design wins with potential significant external customers.
Every load-bearing word there is conditional. Customers are "potential" and they "evaluate." Expansion is "dictated by the amount of committed demand … that we are able to obtain." Intel is telling you that committed 14A demand is the thing that has not happened yet, in the same document that never says Terafab.
Musk said on Tesla's April earnings call that "we plan to use Intel's 14A process," and Intel declined to comment at the time. A plan stated on a call is not a wafer commitment in a filing, and the filing is where a foundry customer eventually has to appear.
The $16.8 billion everyone is quoting is not a plant cost, and it is not comparable to Northland's $55 billion
One number is being passed around this week attached to the wrong thing, and we are close enough to it to be precise.
Texas published the eight Terafab tax agreements, and we went through all of them on September 7. Two separate figures come out of those filings and they are constantly swapped:
- $1,660,434,060 — the school maintenance-and-operations tax forgone across all eight agreements, running to 2046. That is our headline figure from that piece, and it is a tax the districts will not collect, not money anyone spends.
- $16,801,000,000 — phase one's eligible property, being Anderson-Shiro's $10,366,217,000 plus Iola's $6,434,783,000. This is the "$16.8 billion" the Governor's own August announcement used.
The $16.8 billion is the one in circulation, and it is not a capital budget. Eligible property is a tax category: it excludes land and land acquisition, inventory, and anything already standing on the site. The real phase-one outlay is $16.8 billion plus those carve-outs, and nobody outside the consortium has that number.
Which means the $16.8 billion and Northland's phase-one estimate of $55 billion are not a contradiction to be resolved, and not the same quantity measured twice. One is a floor drawn from two tax filings; the other is an analyst's estimate of total construction spend. Northland's full-build figure of $120 billion into the late 2030s, its 22.4 million advanced-logic wafers a year, and its first product in mid-2028 or 2029 are all Northland's estimates and appear in no filing. Label them that way or do not use them.
The filings are also still silent on the thing we flagged two days ago: the applicant on all eight agreements is TeraFab AI, LLC, Intel is named nowhere in them, and who owns TeraFab AI, LLC has not appeared in any filing we can find. That question is open, and Tesla's next annual report remains the likeliest place it gets answered.
What this does to our model: the timing already agrees, the size is the argument
Here is the part worth the reader's attention, because it is the one place where the upgrade and our published numbers touch directly.
Our Intel model is not sceptical about 14A's timing. Its foundry vertical grows at +1.8% a quarter in the terminal period, and the stated reason is "internal wafer demand tracks Products, plus a slow external ramp after the 2028 Intel 14A run." Northland dates first Terafab product to mid-2028 or 2029. The two agree on when. They disagree on how much, and our model declines to size it — because Intel discloses external foundry revenue only from early 2025 and skipped two quarters of it, so an external line would be more gap than data.
So the useful question is not whether Terafab is real. It is how much foundry would have to do to carry a $120 target on its own. That is arithmetic, and we ran it rather than guessing.
On our published assumptions the base case is $36.62 a share. Removing Intel Foundry from the model entirely takes it to $18.85 — so the whole foundry segment, internal transfer and external customers together, is worth about $17.77 a share today. Doubling the entire segment's value still leaves you around $54.
To reach $120 by growing foundry alone, the vertical's terminal growth has to go from 1.8% a quarter to about 18.5% a quarter, which produces $120.97. Compounded across the model's twenty-quarter horizon, that is Intel Foundry at roughly $687 billion of annualised revenue by 2031 — about 10.7 times Intel's entire current company revenue, and several times TSMC's whole business.
That figure is not a forecast. It is a reductio, and it makes the point cleanly: Terafab cannot be what a $120 target is made of. The gap between our $36.62 and $120 is not a foundry-sized gap. Our model's own valuation note says where it actually lives — the exit multiple and the terminal margin, with the market already paying about 15 times the model's 2031 EBITDA. Northland is buying a turnaround and a server shortage and a re-rating, with Terafab as the reason the fabs eventually fill. Read the upgrade that way and it is coherent. Read it as a Terafab revenue call and the arithmetic will not carry you.
No model input changes on this news. A rating is not a disclosure.
What to watch
- A committed-demand statement for Intel 14A in a filing or on a call. Intel's own words make this the gate: expansion is "dictated by the amount of committed demand" it can obtain. A named external customer, a wafer commitment or a prepayment is the disclosure that would move this from evaluation to revenue.
- The external foundry line next quarter. It was $293M in the June quarter against $22M a year earlier, and Intel attributed the increase primarily to Altera. The number to watch is external revenue that Intel does not attribute to Altera — that is the third-party business actually growing.
- Whether Intel names Terafab in a filing at all. It has been a participant since April and has not. The third-quarter report is the next opportunity.
- Who owns TeraFab AI, LLC. Still unanswered, still absent from the eight Texas agreements, and Tesla's next annual report is the most likely place a consolidation or equity-method treatment surfaces.
- Northland's note itself, if it is ever published in full. Nothing public decomposes the $120 between the turnaround, the server shortage and Terafab. Without that split, anyone attributing a share of the target to Terafab is inventing the weights.
Northland's rating, $120 target and quoted reasoning are as reported from its September 8 note, as is the May 2026 downgrade to Market Perform on valuation with the target suspended. Everything here about Intel Foundry — June-quarter revenue, segment operating loss, external revenue of $293 million against $22 million a year earlier, the Altera attribution and the Intel 14A passage — is quoted or taken from Intel's June-quarter report filed July 24; the count of Terafab mentions in it is ours. The April 7 announcement and Intel's statement are as Intel published; Musk's 14A remark is from Tesla's April earnings call as reported. The $1,660,434,060 forgone and $16,801,000,000 of phase-one eligible property are from the eight published Texas agreements, covered here on September 7. Northland's $55bn, $120bn, 22.4 million wafers and 2028–29 dates are its estimates and appear in no filing. The $104.47 and $95.80 closes are September 8 and September 4 and will differ from today's quote. Fair values, the foundry drop test and the 18.5% requirement are our model's arithmetic on our own assumptions.
