Silicon SidelineSILICON SIDELINE
AI Economy Index −8.7%
NASDAQ −3.6%
Gap −5.1 pts
Today +2.8%

AI Economy Index is behind the NASDAQ since 5 Oct

Since 5 Oct. Closed 27 Oct. See the scoreboard

Weekly findingStatus: Reported

Five reported AI chip deals add up to more than $160 billion, and no company has said any of them is done. Five reported AI chip deals are still open, and no company has said any of them is done.

Think of five new stadiums that have been reported but not signed. Nobody has said the deals are done.

Read the briefing See the sources

Finding dated 21 Oct 2026. Sources linked below.

Latest briefings

Keep score of the AI economy.

AI Power Rankings

As of 27 Oct 2026

Private companies are placed on their latest sourced valuation, and where that valuation is about a year old or more, the line says when it was set.

* Rankings show where a company stands in the AI economy. They say nothing about its shares.

The divisions · as of 21 Oct 2026

Eight divisions

From the power plant to your phone.

Cloud and compute

Healthy

Latest play NiSource says Microsoft signed a 20-year energy agreement that adds about 1,100 MW of new generation for its northwest Indiana data center campus from 2028

More on this division

The companies that rent computing power to AI labs and businesses.

· NiSource: NiSource announces long-term agreement with Microsoft expected to increase customer savings and support Indiana's economic growth

Cloud & computeInfra software

See the cloud and compute companies

Healthy Day-to-day Injured list Off-season

Samsung guides to a record quarter

Samsung expects about 107.4 trillion won of third quarter operating profit on about 195 trillion won of sales, up from 12.17 trillion won of profit a year earlier. Healthy.

Quote, 8 Oct 2026: "Consolidated Operating Profit: Approximately 107.40 trillion Korean won"

Source: Samsung Electronics, 8 Oct 2026

ASML says AI demand outruns its tool supply

ASML says its "supply as currently planned for both 2027 and 2028 will not fully meet the demand of our customers", after what it called record third quarter sales of 11.7 billion euros. Healthy.

Source: ASML, Form 6-K, 14 Oct 2026

Microsoft goes big on its own Maia chips

Microsoft says its Maia chips will scale to multiple gigawatts from 2027 and become the main system running AI answers across its products. Healthy.

Quote, 12 Oct 2026: "We will scale our Azure Maia family to multi-gigawatt deployments starting in 2027, making Maia the primary inference system for Microsoft products and services."

Source: Official Microsoft Blog, 12 Oct 2026

Camtek sees over $800 million in 2027

Camtek says AI chip packaging demand gives it an initial 2027 revenue outlook of over $800 million, after what it called record third quarter revenue. Healthy.

Quote, 14 Oct 2026: "2027 initial revenue outlook at over $800 million"

Source: Camtek, Form 6-K, 14 Oct 2026

BUBBLE WATCH · EVIDENCE AS OF 21 OCT 2026

Bubble Watch

Nine conditions that need to stay true for the AI build to keep getting paid for. They come from people close to the money and the building. Unlike opinions, conditions can be checked, so each one is tracked against dated, sourced events. Every row names who said it and links to where. The call moves when the evidence does.

  1. Can they actually plug it all in?Toward NiSource says Microsoft signed a 20-year energy agreement that adds about 1,100 MW of new generation for its northwest Indiana data center campus from 2028 16 Oct 2026

    A chip is useless without power, a building and a grid connection. This row tracks whether the physical build-out lands anywhere near the pace the spending assumes.

    About 25 GW of new AI compute actually comes online in 2027, against SemiAnalysis’s 43 GW forecast. In his words: “I would suggest Dylan’s forecast to 43 gigawatt next year is too aggressive … I think the total amount we’re actually going to stand up is somewhere closer to 25 gigawatt.” The limits he named: permitting and local opposition, grid interconnection delays, skilled labor shortages and sold-out power equipment.

    Stated by Brad Gerstner (Altimeter Capital), All-In Summit, 15 Sep 2026 · source ↗ · by 31 Dec 2027

    Up next: State permitting actions (Texas audit outcome, Virginia 2027 legislation); utility interconnection announcements.

    Open this question in full

  2. Is borrowing still cheap enough to build?Against US 10-year Treasury yield crosses 5% for the first time since 2007 23 Sep 2026

    Data centers are built with borrowed money. When long-term interest rates rise, every project gets more expensive to finance, whatever the demand looks like.

    The 10-year Treasury yield stays below about 5.5%, and oil retreats. In his words: “If rates were to go to five and a half on the 10-year, that’s going to be a big burden … on the equity market,” and “rates are following oil prices to a certain extent.” He gave no numeric level for oil. The 5% warning level on this row is AI Map’s own, set in advance.

    Stated by Brad Gerstner (Altimeter Capital), All-In Summit, 15 Sep 2026 · source ↗

    Up next: Next FOMC meeting (late October 2026); weekly 10-year yield and oil price.

    Open this question in full

  3. Are governments slowing it down?Against FERC rejects the core of PJM's large-load plan, keeping only a 50 MW definition and a registry 9 Oct 2026

    States and countries are writing the rules for data centers now. Those rules cover power, water, permits and tax. The condition is not that nothing happens, but that what happens adds cost and process without stopping the build out.

    Rules on AI and data centers add cost and process without halting the build-out. In his words: “Regulation’s a threat. It’s a risk.” He called for “common sense, pragmatic solutions,” pointed to the 67 fission reactors shut down after activist pressure, and said: “Can’t allow this to occur to AI.”

    Stated by Brad Gerstner (Altimeter Capital), All-In Summit, 15 Sep 2026 · source ↗

    Up next: Texas audit completion (possibly late 2026); Virginia 2027 legislative session; other states following.

    Open this question in full

  4. Are the AI labs actually making money?Against OpenAI's annualized revenue is roughly $50bn, not the ~$68-70bn reported in September, per FT, Bloomberg and CNBC 8 Oct 2026

    Almost everything else rests on this. The data centers, the chips and the power deals are all being financed against an expectation that the companies making AI models grow revenue very fast. This row tracks whether they do.

    The top three AI labs lift combined annualized revenue from about $100B (July 2026) to at least $180B by the end of 2026. In his words: “I think they need to collectively get to at least $180 billion by the end of the year … just to keep the AI trade intact.” He put the takeoff case at about $8B a month for a leading lab, against $4B.

    Stated by Brad Gerstner (Altimeter Capital), All-In Summit, 15 Sep 2026 · source ↗ · by 31 Dec 2026

    Up next: Next disclosed or credibly reported monthly run-rate from OpenAI and Anthropic.

    Open this question in full

  5. Can anyone pay for all these data centers?Toward SpaceX's AI business brings in $7.6 billion in the third quarter as computing power reaches 1.7 GW 20 Oct 2026

    The largest technology companies are spending on a scale normally seen in railways and oil. The question is whether the money coming back from AI arrives quickly enough to cover it.

    Revenue from the labs renting AI capacity grows fast enough to pay for the build-out. In his words: “If you’re going to build a trillion and a half dollars a year in capex, somebody has to pay for it … Microsoft’s not paying for it. They’re building it to rent it.” He said that revenue has to go from about $200B at the end of 2026 “to 450 to 800 or a trillion dollars just to keep up.”

    Stated by Brad Gerstner (Altimeter Capital), All-In Summit, 15 Sep 2026 · source ↗

    Up next: Hyperscaler Q3 2026 earnings and capex guidance (late October); lab financing disclosures.

    Open this question in full

  6. Will one of the big AI labs go public?Against OpenAI's CEO rules out a 2026 IPO, calling now an ill-advised moment to go public 12 Sep 2026

    A listing would force a frontier lab to publish real numbers on a schedule. Whatever it did for the share price, it would replace a lot of guesswork with filings.

    A frontier-lab IPO goes ahead this year without a halt or postponement. In his words: “I think we are going to see an IPO this year,” and a halt or postponement of the Anthropic IPO “would obviously be a major issue. I don’t think that’s going to happen.”

    Stated by Brad Gerstner (Altimeter Capital), All-In Summit, 15 Sep 2026 · source ↗

    Up next: Anthropic IPO filing/pricing news; OpenAI listing plans (Altman has said no IPO in 2026).

    Open this question in full

  7. Is the rally still tied to earnings?Toward TSMC raises its 2026 sales growth outlook to about 44% after third-quarter revenue of $46.72 billion topped its guidance, and says demand stays strong 15 Oct 2026

    The claim being tested is that this does not look like 2000, because the gains come from rising profits rather than rising valuations. That is checkable every quarter.

    Market gains stay driven by earnings rather than rising valuations. In his words: “This is not about multiple expansion. This is an earnings driven market expansion,” with NVIDIA at “14 times next year’s fully taxed GAAP earnings. This is no bubble like it was in 2000.” He also flagged that semiconductors were 70% of the Nasdaq’s return, “both good and bad.”

    Stated by Brad Gerstner (Altimeter Capital), All-In Summit, 15 Sep 2026 · source ↗

    Up next: Q3 2026 earnings season; index concentration data.

    Open this question in full

  8. Is AI making ordinary companies more productive?Nothing has yet cleared the bar this tracker requires. That is a real answer, not a gap.

    Not whether people use it, but whether it shows up in what companies actually report. This is the difference between AI as a cost and AI as a return.

    AI lifts companies’ margin expansion from the historical ~38 basis points a year toward ~100. In his words: “Can we turn the 38 bips to 100 bips of margin expansion because of AI? The answer is obviously yes.” His examples were companies growing revenue without growing headcount.

    Stated by Brad Gerstner (Altimeter Capital), All-In Summit, 15 Sep 2026 · source ↗

    No signal has cleared the bar for this row yet. The tracker counts published, dated events from the last 90 days, and analyst estimates and internal projections that are not in a filing never count toward a verdict.

    Up next: Q3 2026 earnings commentary on AI-driven margins.

    Open this question in full

  9. Is anyone actually using AI agents?Nothing has yet cleared the bar this tracker requires. That is a real answer, not a gap.

    Launches and download estimates are easy. This row waits for something a company discloses itself: sustained use, paid conversion, or revenue broken out.

    Consumer AI agents become a real, token-hungry category. In his words: “We’re going to have consumer agents in everybody’s pocket … This could be another trillion dollar category, but it’s definitely going to consume massive tokens.” AI Map counts only company-disclosed usage, paid conversion or transactions toward this row, not downloads.

    Stated by Brad Gerstner (Altimeter Capital), All-In Summit, 15 Sep 2026 · source ↗

    No signal has cleared the bar for this row yet. The tracker counts published, dated events from the last 90 days, and analyst estimates and internal projections that are not in a filing never count toward a verdict.

    Up next: A company-disclosed usage, subscriber or transaction figure for a consumer agent (Meta, OpenAI or Anthropic), or a quarter in which agent revenue is broken out.

    Open this question in full

These are conditions AI Map tracks, not predictions and not our opinion of whether any of it is a good idea. Each row names whoever stated the threshold and links to where they said it. Where the wording is our own paraphrase, the row says so. We are not affiliated with, endorsed by or reviewed by anyone cited. A row moves against the condition when a major dated event points against it, and with the condition when the qualifying evidence only points toward it. When nothing has cleared the bar, the row stays open. Analyst estimates and internal projections that are not in a filing never move a verdict, unconfirmed reports are never published, and a signal stops counting after 90 days. It is not a rating, a forecast, or a recommendation about any company or security. How a verdict is set, and what it is not.

Get the PulseSee them on the mapRead the briefings

Around the league

  • All-In recap: CoreWeave's debt selloff

    All-In recap: CoreWeave's debt selloff

    Recap, no quotes. On episode 293 the hosts opened on CoreWeave's debt selloff and asked whether the AI bubble just popped. Their views, not ours.

    Source: All-In Podcast, episode 293, 17 Oct 2026

  • Anthropic researchers on self copying AI

    Anthropic researchers on self copying AI

    Anthropic researchers say small AI models may be months away from being able to copy themselves onto cheap connected devices like off brand TV boxes. In their tests a 9B model got inference running on 8 of 9 hardware types. Research and opinion, not a company move.

    Quote, 12 Oct 2026: "In our evaluations, a 9B model got inference running on 8 of the 9 hardware types we tested."

    Source: Anthropic research, 12 Oct 2026

Every source, linked

CEO Huddle

01 / MISSION

MAKE THE AI ECONOMY VISIBLE

Most people see products.
We map the system beneath them.

AI is not one industry. It is a chain of power generation, fabrication equipment, chips, data centers, cloud platforms, models, software and physical machines. This database, and the relationship map behind it, show how those layers depend on one another, and the evidence behind every connection.

Go deeper

The map286 companies, one picture of who pays whom. CompaniesSearch any team and read its sourced file.
JUMP TO A SECTOR
01 Energy & grid 02 Data centers 03 Chip equipment 04 Fabrication & memory 05 AI chips & networking 06 Servers & AI factories 07 Cloud & compute 08 Foundation models 09 Infra software 10 Applications 11 Robotics & physical AI
01

Map relationships

Follow suppliers, customers, partners, spending, ownership and competition across the full AI stack.

02

Show the evidence

Every published connection is tied to a readable source, date, classification and statement of its limits.

03

Separate signal from hype

Current operations, dated announcements and competitive comparisons stay visibly distinct, and every record carries the date its evidence was published and the date it was last reviewed.

04

Explain it plainly

Built for intelligent non-specialists who want the system without reading hundreds of filings.

Find a company

Every story here comes with the Pulse. Sign up for the Pulse

This is general information, not investment advice.

FOUNDER'S NOTE

Silicon Sideline is an independent educational research project founded by Steven Dudley. It begins with a simple question: what is actually being built, why does it matter, and how are the companies building it connected?

We cover opportunity and constraint with equal seriousness. Inclusion in this database is not an endorsement of a company or its securities.