November 5, 2026
One Week, Two Verdicts on the Same Buildout
On Nov. 5, NEPC's Q3 2026 private markets report put AI at $426.7 billion, or 82.7%, of US venture deal value through September. On the same session the Nasdaq Composite fell 5.5% and the S&P 500 4.4%; the Magnificent Seven lost more than $1 trillion in market value, Micron fell 17%, and NVIDIA fell 9%, surrendering more than $440 billion in a day. Since Oct. 6, the largest AI and chip names have lost roughly $5 trillion.
AI share of US venture deal value
82.7%
$426.7B of $515.8B through the third quarter of 2026 (NEPC)
North America venture deal value, Q3 2026
$98.4B
Q3 spending fell 61% from the Q1 peak of $253.5B
Nasdaq Composite, Nov. 5 close
-5.5%
Third correction of 2026 and more than 10% below its high
AI and chip market value lost since Oct. 6
$5T
Yahoo Finance calculation, Nov. 1, 2026
Anthropic committed compute purchases
$895.4B
Revised S-1 filed Nov. 2, 2026
These are not contradictory prints. The private mark prices the scarce inputs of the buildout — accelerators, high-bandwidth memory, power and shells — while the public tape prices the return on those inputs, and above all the cash burn and the cost of the capital that funds them. One market is underwriting a physical buildout; the other is underwriting an income statement. Which one is right decides who funds the next dollar of AI capacity.
The private side
82.7% Is Mostly a Shrinking Denominator
AI's share of US venture deal value hit 82.7% through the third quarter, and it is tempting to read that as a demand signal. Look at the denominator instead. North America venture deal value fell from $253.5 billion in Q1 to $163.9 billion in Q2 to $98.4 billion in Q3 — down 61% from the Q1 peak — so AI's share rose while the market contracted by two-thirds.
North American venture deal value fell every quarter of 2026 while AI's share of it kept rising
North America venture deal value by quarter, 2026 ($B). Source: NEPC, Quarterly Private Markets Report, Nov. 5, 2026.
Unit: $B
Q1 2026
AI and non-AI rounds combined
253.5
Q2 2026
Down 35% sequentially
163.9
Q3 2026
Down 40% sequentially
98.4
- AI took 82.7% of deal value on roughly 45% of deal count, so the implied average AI round is about $80 million against roughly $14 million for everything else — the average AI check runs nearly six times the average non-AI check.
- The buyer base is thinning. First-time funds raised $4.9 billion across 81 funds through Q3 versus $9.4 billion across 195 funds in 2025, and growth equity raised $19.6 billion across 73 funds versus $50.7 billion across 105.
- Concentration is severe: 26 funds of $1 billion or more raised $73.7 billion, roughly 68% of all North American venture capital raised through Q3.
- Carta's platform data corroborate the shape rather than contradicting it: $49.3 billion across 1,179 rounds in Q3 — about half NEPC's market-wide $98.4 billion — with deal count down 21% year over year.
- The marks are doing the work. NEPC's rolling one-year venture return reached 32.02% as of Q2 2026, from 24.54% in Q1 and 12.93% a year earlier.
Where the private bid went
The Private AI Bid Migrated Up the Cap Table and Into the Buildout
Carta's Q3 data show where the private AI money actually sits. AI took about 57% of all capital invested at Series B and beyond this year, the highest share ever recorded at that stage, while its share of seed and Series A cash fell to 44% from 48% in 2025. The private AI premium has migrated up the cap table, out of applications and into physical capacity.
| Carta platform metric | Reading |
|---|---|
| AI share of capital, Series B and beyond, year to date | About 57%, the highest ever recorded at that stage |
| AI share of capital, seed and Series A, year to date | 44%, down from 48% in 2025 |
| Total venture funding in Q3 | $49.3B across 1,179 rounds |
| Deal count, year over year | Down 21% |
| Median valuations by stage | Above 2025 at every stage and above 2021 at most stages |
| Hardware startup funding, 2026 through Q3 | Nearly $34B, more than double the first three quarters of 2025 |
- Supply-side evidence matches the shift. SK hynix began shipping HBM4 in earnest in the September quarter, expects HBM4 to surpass HBM3E on both volume and revenue in Q4, and plans investment to rise significantly next year as Yongin Phase 1 ramps.
- The exit channel stayed open while the tape fell: North American venture exit value reached $2.3 trillion across 1,115 exits in Q3, with IPOs accounting for roughly $1.8 trillion across 58 listings.
- That is why the pipeline into public markets is not blocked — it is simply much narrower than the private market's $2.3 trillion of quarterly exits suggests.
The price of the mark
Anthropic's Own S-1 Shows How the Mark Works, and How It Breaks
Anthropic's S-1, filed Oct. 26 and revised Nov. 2, is the cleanest window into the private mark. The company reported roughly $36.7 billion of trailing revenue — $4.585 billion in 2025, $16.286 billion in the first half of 2026 and about $17.0 billion in the September quarter alone — against a first-half net loss of $80.7 billion, of which $72.4 billion was the non-cash remeasurement of convertible notes, derivative liabilities and SAFEs. A rising private valuation turns into a reported loss through those instruments, which is how a record private mark and a broken public tape can sit in one set of accounts.
The revised filing also discloses about $895.4 billion of committed compute and infrastructure purchases, substantially all of it compute, while Reuters reports the prospectus plans $518 billion of cloud, compute and infrastructure spend over coming years against $131.7 billion of cash and short-term investments at June 30. The obligations survive whatever the IPO prints, because they are take-or-pay contracts written at private-mark-era terms.
| Private AI asset | Most recent mark | Status | Trailing revenue |
|---|---|---|---|
| Anthropic | $965B (May 28, 2026) | IPO sale could value it above $2T; press reports put the discussed range at $1.5T–$2.0T | $36.7B |
| OpenAI | $852B (Mar. 31, 2026), on $122B of committed capital | Reportedly raising about $30B at a $1.25T valuation | Not disclosed |
One AI label, two price levels: private at 41x–55x trailing revenue, public at 6x–16x
Price to trailing revenue. Listed market capitalisations at the Nov. 5, 2026 close over trailing-twelve-month revenue; Anthropic's $36.7B is computed from its S-1 and revised S-1. A trailing basis for Anthropic rather than an annualised quarter is the more conservative choice for the private mark.
Unit: x trailing revenue
Anthropic at $1.5T
Lower end of the discussed IPO range
40.9
Anthropic at $2.0T
Reuters: the sale could value it above $2T
54.6
$1.39T market cap, $89.1B revenue
15.6
$4.44T market cap, $303.0B revenue
14.7
$3.48T market cap, $471.1B revenue
7.4
$1.35T market cap, $241.6B revenue
5.6
$743B market cap, $133.2B revenue
5.6
The public side
The Public Tape Broke on Cash Burn, Not on Demand
The sell-off had a checkable cause. Alphabet and Meta spent a combined $78.9 billion of capital in the September quarter and together produced negative $10.2 billion of free cash flow, while depreciation on those assets remains a fraction of what is being spent. The market is not arguing about AI demand; it is arguing about who pays for it.
| Q3 2026 | Alphabet | Meta |
|---|---|---|
| Revenue | $127.577B, up 25% year over year | $64.601B |
| Operating income | $40.576B, 32% margin | $12.988B |
| Depreciation and amortisation | $8.369B | $7.154B |
| Capital expenditure | $44.924B | $33.942B |
| Free cash flow | -$9.065B | -$1.156B |
| Funding response | 2026 capex held; significant increase flagged for 2027 | Record $97.5B equity raise; $25B of common at $625 a share |
| Share price reaction | -10% on Oct. 28, its biggest one-day percentage drop on record | -3.3% on Nov. 5 |
- Meta is the sharper signal. It answered the burn with the largest equity offering in its history — about $37.5 billion of common and convertible preferred stock, plus a program for up to $60 billion more from Q1 2027 — while raising the floor on2026 capex to $135 billion. Equity issued at $625 a share is now below the Nov. 5 close of $609.40.
- J.P. Morgan expects global AI capex of about $900 billion in 2026 and $5.5 trillion by 2030; Amazon, Alphabet, Meta, Oracle and Microsoft are on course to spend as much this year alone as they did cumulatively since ChatGPT launched in November 2022.
- The index damage compounded: the Nasdaq-100 fell more than 5.8% in October with more than $2 trillion of value erased, then entered its third correction of 2026 on Nov. 2.
The funding layer
The Marginal AI Dollar Now Costs 15% to 16%
So where does the marginal AI dollar go when public equity withdraws? Into private credit, where AI-linked paper now prices at 15% to 16%. Direct-lending funds raised $129.15 billion in the first nine months of 2026, up 35% year over year, even as third-quarter lending volume fell 28% to $49.77 billion.
- CoreWeave is the stress test. It carries about $50 billion of debt; the cost of insuring that debt against default climbed to roughly 1,200 basis points and yields on some of its bonds reached 15% to 16%.
- The lender group is concentrated and the terms are tightening. Carlyle said private credit could be called on to finance roughly $1 trillion of the AI-compute buildout, more than half the industry's assets under management; CoreWeave's facilities grew from $2.3 billion in 2023 to $7.5 billion in 2024 and $8.5 billion in March 2026.
- New direct-lending issuance clustered at 450–499 basis points (54% of 2026 volume) and 500–549 basis points (32%); non-software loans are pricing below SOFR+500 because competition for AI-linked paper is so intense, while middle-market first-lien spreads widened slightly to 515.75 basis points with the Fed having added 25 basis points on Sept. 16.
- Money is already migrating back out of private credit: $16.39 billion of direct-lending loans were refinanced into the broadly syndicated market in Q3, against just $2.60 billion moving the other way.
- The private market carries two marks and they are moving in opposite directions. Ares Capital ended Q3 at 0.98 times book with quarterly earnings down 22% year over year, while NEPC's rolling one-year venture return rose to 32.02%. Private equity marks are rising as private credit marks fall.
Where the marginal AI dollar is going
Public equity
$97.5B
Meta record raise, $25B of common at $625 a share
Hyperscaler capex
~$900B
Global AI capex in 2026, per J.P. Morgan
Direct lending
$129.15B
Raised by direct-lending funds in the first nine months of 2026, up 35% year over year
Private equity
$108.5B
North American venture fundraising through Q3 2026, across 699 funds
AI infrastructure debt
15%–16%
Yields on CoreWeave bonds after the October repricing
Private AI commitments
$895.4B
Anthropic's committed compute purchases per its revised S-1
The supply chain
The Suppliers Are Still Getting Paid, and Their Multiples Were Cut Anyway
The drawdown is a statement about capex, not about demand. In the same weeks the tape collapsed, the parts of the buildout that sell for cash posted records: suppliers were marked down on their customers' capex decisions, not on their own order books.
| Company | Latest reported quarter | Revenue | Profitability | Nov. 5 close vs. 52-week high |
|---|---|---|---|---|
| NVIDIA | Q2 FY2027, ended Jul. 26, 2026 | $96.221B, up 106% year over year | 75.0% gross margin; Data Center revenue $89.0B, up 117% | $184.35, 24.3% below $243.37 |
| SK hynix | 3Q26 | ₩104.6T, up 32% sequentially | ₩83.3T operating profit, an 80% margin; net cash ₩98.7T | ₩1,264,000, 57.7% below ₩2,987,000 |
| Micron | FQ4 2026, ended Sept. 3, 2026 | $54.229B, up 31% sequentially | 86.8% gross margin; Core Data Center $18.002B at a 90% margin | $656.39, 47.7% below $1,254.61 |
| Broadcom | Q3 FY2026, ended Aug. 2, 2026 | $29.6B; AI semiconductors $16.7B, up 221% year over year | $13.665B free cash flow, 46% of revenue | $291.53, 40.9% below $493.28 |
| Arista Networks | Q3 2026 | $3.66B, up 58.5% year over year | 49.1% operating margin | $159.16, 26.8% below $217.41 |
| Equinix | Q3 2026 | $2.571B, up 11% year over year; record $533M annualised gross bookings | Third consecutive guidance raise, to $10.33B–$10.37B of 2026 revenue | $997.35, 10.8% below $1,118.65 |
- Micron guided fiscal Q1 2027 to $61.5 billion of revenue at roughly an 86% gross margin, while Arista Networks grew 58.5% and Equinix lifted guidance for a third straight quarter on record bookings of $533 million, up 35% year over year.
- SK hynix expects the memory shortage to persist through 2030 and plans to spend significantly more next year —the supply chain is investing into its own downcycle, which is what a shortage call looks like.
- NVIDIA guided fiscal Q3 revenue to $108.0B, up from $96.221B, while the stock fell 9% on Nov. 5 and more than $440 billion of value disappeared in that session.
- On earnings the suppliers are the cheapest way to own the buildout: roughly 4.3x trailing for SK hynix and 4.9x forward for Micron, against 14.6x for Alphabet and 21.5x for Meta.
So the public tape is repricing the owners of the buildout, not its suppliers — and the suppliers' multiples were cut anyway because they carry the owners' capex risk in their share prices. That asymmetry is the trade. Own the contracted revenue, and let the narrative be free.
Horizons
Two Horizons, One Catalyst
| Catalyst | Timing | What it decides |
|---|---|---|
| Anthropic IPO pricing | Mid-to-late November 2026 | Whether a private mark above $965B can clear in a market three corrections deep |
| NVIDIA fiscal Q3 FY2027 results | November 2026 | Whether $108.0B of guided revenue survives the capex scare |
| Meta at-the-market equity sales | From Q1 2027 | Whether up to $60B of stock can be placed without further dilution |
| Alphabet 2027 capex | With fourth-quarter results in January 2027 | Whether $44.9B of quarterly capex becomes a 2027 step-up |
| CoreWeave credit pricing | Ongoing | Whether 15%–16% yields and 1,200 basis points of default insurance hold |
| Private-credit redemptions | Each quarter | Whether gated withdrawal requests stay below new inflows |
Over the next two quarters, the single most informative print is Anthropic's IPO price. A deal near the reported range would say the private mark was right all along and the public tape is a discount; a deal at or below the $965 billion May mark would say the opposite, and would reset the private credit that funded the buildout. One auction now arbitrages a $5 trillion gap.
Over one to three years, the binding variable is not the demand curve but the depreciation clock. Alphabet is spending $44.9 billion a quarter against $8.4 billion of quarterly depreciation, and Meta is spending $33.9 billion against $7.2 billion; as those assets age into the income statement, reported returns fall even if utilisation holds. Private credit becomes the systemically important funding channel, because the public equity bid has gone and the private-equity bid is concentrated in a shrinking pool of funds.
Falsifiers
What Would Break This View
- If Anthropic prices at or below $965 billion, the private mark was a liquidity artefact and the reset propagates into private credit, then into supplier multiples. A soft IPO breaks both prices, not one.
- If hyperscaler capex guidance is cut — Alphabet's 2027 step-up is the first test — the suppliers' contracted volumes become the buffer, which is why long-term agreements covering more than 50% of 2027–2028 memory capacity matter so much.
- If private credit stops lending, the buildout stalls: third-quarter direct-lending volume already fell 28% year over year and net new issuance is migrating back to the syndicated market.
- If memory supply normalises before hyperscaler capex does, the suppliers' record margins become the liability, and the public tape's discount proves to have been early rather than late.
The honest read is that the public tape and the private mark are not both wrong. The private mark prices inputs that are genuinely scarce, and the suppliers' reported margins confirm it. The public tape prices something else: the cost of capital for an industry spending about $900 billion a year, and the depreciation that spending will eventually produce. Own the contracts, rent the narrative — and watch one auction.
Where the two-price problem shows up in listed shares
- Data Center revenue of $89.0B in Q2 FY2027 grew 117% year over year, and management guided Q3 revenue to $108.0B.
- The stock trades at roughly 13x forward earnings after shedding $440 billion of value in a single session.
- Days to quarters: fiscal Q3 results in November test the $108.0B guide; one to three years: the risk is a hyperscaler capex cut, not chip demand.
- Q3 operating profit of ₩83.3T on ₩104.6T of revenue is an 80% operating margin, with net cash of ₩98.7T.
- Management expects the shortage to persist through 2030 and has locked long-term agreements covering more than 50% of 2027–2028 capacity.
- Days to quarters: HBM4 becomes the mainstream product from Q4; one to three years: the de-rate persists unless AI capex guidance stops falling.
- FQ4 revenue of $54.229B at an 86.8% gross margin, with Core Data Center at $18.002B and a 90% gross margin.
- The stock trades at roughly 4.9x forward earnings after a 17% one-day fall.
- Days to quarters: fiscal Q1 2027 guidance of $61.5B at roughly 86% gross margin is the checkpoint; one to three years: customer capex budgets, not HBM supply, set the cycle.
- Q3 FY2026 AI semiconductor revenue of $16.7B grew 221% year over year, and Q4 revenue is guided to about $34.8B, up 93%.
- Free cash flow of $13.665B covered 46% of revenue while the shares closed 41% below their 52-week high.
- Days to quarters: Q4 FY2026 results in December; one to three years: AI revenue depends on the same hyperscaler budgets now funded with equity and private credit.
- Revenue rose 25% to $127.577B with a 32% operating margin, and the stock trades at about 14.6x trailing earnings.
- Against that: free cash flow was negative $9.065B on $44.924B of quarterly capex, with $8.369B of depreciation and a flagged 2027 capex step-up.
- Days to quarters: the fourth-quarter report in January 2027 sets 2027 capex; one to three years: depreciation on a fast-growing asset base compresses reported returns.
- Q3 free cash flow was negative $1.156B on $33.942B of capex, cash fell to $24.354B, and 2026 capex is now guided to $135B–$145B.
- The record $97.5B raise was struck at $625 a share, above the Nov. 5 close of $609.40.
- Days to quarters: the at-the-market program for up to $60B starts with Class C trading in Q1 2027; one to three years: dilution and depreciation compound.
- About $50B of debt, with yields on some bonds at 15%–16% and default insurance near 1,200 basis points.
- The equity trades at 3.2x trailing sales with net debt at 11.5x EBITDA and capex running at 271% of revenue.
- Days to quarters: the price of the next facility; one to three years: the model only works if lab demand converts into contracted, take-or-pay revenue.
- Quarterly earnings fell 22% year over year and revenue slipped 2.4%, yet the shares still yield10.1%.
- The stock trades at 0.98x book with a payout above reported earnings.
- Days to quarters: quarterly results and redemption gates; one to three years: private credit becomes the system's funding channel while its own marks reset.
