SharonAI Holdings
This is the archived full research. Read the story
An Australian company that rents NVIDIA's AI chips to AI labs and technology firms on five-year contracts, with billions of dollars signed and very little billed yet.
The scores
Growth3/4 ↑
Revenue went from $0.51M to $9.93M a quarter in a year and $3.8B is contracted, but billing has run behind estimates twice.
Business quality2/4 →
Five-year take-or-pay contracts in a shortage, but no owned power or buildings and much larger rivals selling the same thing.
Financial position2/4 ↓
$1.34B of cash roughly matches what it owes suppliers, and about $5.0B of commitments are unfunded; the 10-Q says more financing is required.
Execution record2/4 →
Contracts signed fast, but revenue missed estimates by 65% and 56% in successive quarters and both top finance roles changed.
Per share capture2/4 ↓
Shares nearly quadrupled in nine months, though mostly sold at prices far above today's; the founders' votes outweigh their stake.
Potential3/4 →
A specialised AI cloud market above $25B growing fast; capacity is contracted. Not a 4: funding, not demand, limits it.
Risk profile2/4 →
Funding, delivery and a few large anonymous customers carry the case, and the GPUs lose value as new chips arrive.
Valuation3/4 ↑
About 1.3 times 2027's forecast revenue against CoreWeave's 2.7, but only if the unfunded commitments do not dilute it away.
Evidence quality3/4 →
Statements, obligations and the funding facility are filed; contract values, prices per megawatt and the NVIDIA figure are management's.
If it works 3.0× · if it fails you keep up to 35%.
The Verdict
We rate SharonAI Neutral. It rents clusters of NVIDIA chips, mostly on five-year contracts that customers pay whether they use them or not. The business is mixed: billions are signed, but it owns neither power nor buildings, and sales have come in late twice. The price is low against what the signed contracts would bill, but about $5.0 billion of equipment is still to be paid for. If January's billing arrives and debt fills the gap, it is cheap; if new shares fill it, it is not. As a bet it is speculative: about three times its value if everything bills, about a third kept if not.
The long caseAI needs more computing than can be switched on, so whoever has GPUs running near the buyer earns rent on long prepaid contracts; SharonAI has signed that demand onshore.
How we weighed it
The price against our value
Cuts both ways
No fair value is drawn: no lasting profit yet. On forecast revenue the price is low against CoreWeave's, but only before the unfunded equipment is paid for.
The business
Cuts both ways
Mixed
Long prepaid contracts in a shortage, but it owns neither the power nor the buildings, and larger rivals offer the same.
Growth against what the price assumes
In its favour
Revenue is multiplying from a tiny base and contracts already signed would carry it much further.
What could go wrong
Against it
Most of the equipment is unfunded, billing has slipped twice and a few anonymous customers hold most contracts.
Against competitors
Against it
CoreWeave, Nebius and the big clouds offer the same compute at far greater scale.
How its group is behaving
Cuts both ways
AI cloud names move together on AI spending and funding news, and the group fell hard in October.
The latest news and report
Cuts both ways
The third quarter brought more contracts and a first GPU loan, but revenue again fell well short of estimates.
The evidence under this - 10 claims
Customer deposits were $331.5M in nine months against $5.3B of commitments, so prepayments fund only part of the build.
The NVIDIA arrangement, the largest contract by management's count, sits mostly outside this figure because its minimum depends on use.
CoreWeave is about forty times SharonAI's size and investment-grade lenders treat it differently, so its multiple flatters a small rival.
Roughly $5.0B of unfunded commitments (C32) will add debt or shares, so the value per share needs far more than a third of capacity billing.
Customer prepayments of up to 30% of contract value (C03) and further GPU-backed loans planned for 2027 (C10) will fund part of the gap without new shares.
- Q3 revenue was 56% below the analysts' estimate and Q2's 65% below, so the company's own timing has run late twice.
- The January target needs about 12,000 GPUs commissioned in December and January.
Management says newer contracts price above $19M a megawatt, which would raise the full figure to about $2.5B.
New GPU debt ranks ahead of shareholders and is not yet on the balance sheet, so the figure is a ceiling, not an estimate of recovery.
The company
Training and running AI takes thousands of NVIDIA chips working together. SharonAI buys them, installs them in data centres it rents from landlords such as NEXTDC, and sells time on them, mostly reserved for five years, sometimes by the hour.
Buyers could rent from Amazon, Microsoft or Google, from bigger specialists such as CoreWeave, or buy their own chips. They come to SharonAI because chips switched on nearby are scarce: outside price indices show rents for these chips rising through 2026.
The evidence under this - 5 claims
The NEXTDC order was amended and phased from 2027, so most of the contracted megawatts are not yet built or energised.
CoreWeave is about forty times SharonAI's size and investment-grade lenders treat it differently, so its multiple flatters a small rival.
The same index fell roughly three-quarters from early 2024 to late 2025, which shows how fast a chip generation loses its rent once supply catches up.
How it makes money
Customers pay for reserved chip time. On a typical contract they prepay part of it, SharonAI orders the chips against that money, and monthly billing starts about five months later. Management says new contracts bring in more than $19 million a year for each megawatt of power the chips use.
Little of it shows yet: third-quarter 2026 sales were $9.93 million, against $0.51 million a year earlier, and running costs were more than two and a half times that.
The evidence under this - 4 claims
Customer deposits were $331.5M in nine months against $5.3B of commitments, so prepayments fund only part of the build.
No customer or outside index confirms prices per megawatt; outside indices quote per GPU-hour on older chips.
Most of the cost is staff and set-up for capacity not yet billing, so the ratio will fall sharply if January's billing pace arrives.
Where the money comes from, stream by stream
| Stream | Share of sales (nine months to September 30, 2026) | How it earns it |
|---|---|---|
| GPU cloud | 100% | Rented NVIDIA GPU clusters, mostly reserved for five years under take-or-pay contracts with prepayments, some on demand; ESDS's capacity began billing in Q3. |
Who runs it
James Manning, a co-founder who previously ran a listed bitcoin miner, became chief executive in January 2026. A new finance chief from Macquarie and a new operating chief from Telstra joined in August and September. The three founders' special shares carry about 36% of the votes, and one fund owns 21% of the ordinary shares. Pay has been small, and no insider has bought or sold on the market in the past year.
The evidence under this - 5 claims
Why it wins, and who could beat it
Why customers come back
What SharonAI has is chips it can switch on soon: it is an NVIDIA cloud partner, has a six-year arrangement with NVIDIA for up to 40,000 of its newest chips, and has power booked with data-centre landlords in Australia and New Zealand.
That buys a place in a queue, not a lasting edge. It shows in pricing: management says prices on new contracts rose from about $15 million to over $19 million a megawatt a year, and outside indices show chip rents up since late 2025.
The evidence under this - 4 claims
- The 10-Q states no fixed aggregate value, so the $4.9B is management's reading of a minimum that depends on utilisation.
- NVIDIA's own filings do not name SharonAI.
The NEXTDC order was amended and phased from 2027, so most of the contracted megawatts are not yet built or energised.
No customer or outside index confirms prices per megawatt; outside indices quote per GPU-hour on older chips.
The same index fell roughly three-quarters from early 2024 to late 2025, which shows how fast a chip generation loses its rent once supply catches up.
The threat
CoreWeave sells the same thing at about forty times the size, and Nebius, IREN, Amazon, Microsoft and Google all rent the same chips. A broker's bear case on the whole group is that rents fall as new chips arrive, the biggest buyers build their own, and borrowing gets dearer. SharonAI owns no buildings or power, so its edge is speed, not cost.
The evidence under this - 3 claims
CoreWeave is about forty times SharonAI's size and investment-grade lenders treat it differently, so its multiple flatters a small rival.
The rivals: CRWV, NBIS, IREN, AMZN, MSFT, GOOGL, and 1 more that is private, in-house or listed abroad.
Who else does this, rival by rival
| Ticker | Rival | Where it competes | How far along | Priced against it |
|---|---|---|---|---|
| CRWV | CoreWeave, Inc.Priced against it: Yes | SharonAI GPU cloud | On sale | Yes |
| NBIS · IREN | Nebius Group N.V. and IREN LtdPriced against it: Yes | SharonAI GPU cloud | On sale | Yes |
| AMZN · MSFT · GOOGL | Amazon Web Services, Microsoft Azure and Google CloudPriced against it: Shown, not counted | SharonAI GPU cloud | On sale | Shown, not counted |
| In-house | Large buyers building their own GPU fleetsPriced against it: In-house | SharonAI GPU cloud | On sale | In-house |
How big it could get
How big this could get
Specialist AI clouds like SharonAI sold more than $25 billion of computing in 2025, by an outside count, and the power AI servers use is expected to grow about 30% a year to 2030. Even January's target billing would be under a fiftieth of that market, so demand is not what limits SharonAI; money to buy chips is.
The evidence under this - 3 claims
Synergy's forecast of close to $400B by 2031 is a projection, not a measurement, and is not used here.
The IEA's own range runs from a lower efficiency case to a lift-off case; the base case is one scenario of several.
Management says newer contracts price above $19M a megawatt, which would raise the full figure to about $2.5B.
Who has signed
ESDS, an Indian cloud company, signed for $1.25 billion over five years and started paying in the third quarter. An AI lab signed $1.32 billion in New Zealand, a technology company $950 million, and two more about $1.2 billion after the quarter, all unnamed. NVIDIA's six-year arrangement has no fixed total: its minimum depends on use.
The evidence under this - 3 claims
Only ESDS is named; the other counterparties are anonymous and none has disclosed the contract in its own filing.
Neither counterparty is named.
- The 10-Q states no fixed aggregate value, so the $4.9B is management's reading of a minimum that depends on utilisation.
- NVIDIA's own filings do not name SharonAI.
The results
The numbers
Operating margin is −587%.
Running costs take 667¢ of each sales dollar, down from 855¢ in the year to Dec 2024.
Shareholders' money lost 95% a year, down from −13% in the year to Dec 2024.
It used $75.1M of cash over the last twelve months, so the $1.34B it holds lasts about 17.9 years at that pace.
It holds $1.34B of cash against $1.01B of debt, $333M more cash than it owes.
All the numbers
Filed figures from the company's 10-K and 10-Q, read 2026-11-17. Fiscal years end in December.
| Dec 2024 | Dec 2025 | 12 mo to Sep 2026 | |
|---|---|---|---|
| Size and growth | |||
| Sales | $438K | $1.57M | $12.5M |
| Sales growth | - | +257% | - |
| Profit | |||
| Gross margin | −64% | 6.4% | 80% |
| Operating margin | −919% | −880% | −587% |
| Net margin | −891% | −25× | −42× |
| What growth costs, per sales dollar | |||
| Selling, general and admin | 540% | 773% | 385% |
| Pay in shares | 58% | 112% | 65% |
| Capital spending | 672% | 699% | 255% |
| Cash | |||
| Cash from operations | −$2.21M | −$2.64M | −$43.1M |
| Free cash flow | −$5.15M | −$13.6M | −$75.1M |
| Balance sheet | |||
| Cash and short-term investments | $4.42M | $71.1M | $1.34B |
| Debt | - | $258M | $1.01B |
| Net cash | - | −$187M | $333M |
| Returns and shares | |||
| Shares (diluted) | - | - | 38.5M |
What growth costs
It is too early to say whether growth will make SharonAI more profitable. Running costs were more than two and a half times third-quarter sales, and the commitments for chips are hundreds of times them. No quarter yet shows what a fully billing cluster earns.
The evidence under this - 3 claims
Most of the cost is staff and set-up for capacity not yet billing, so the ratio will fall sharply if January's billing pace arrives.
Some commitments are matched by customer prepayments still to arrive, which the 10-Q does not total.
The last report (16 Nov)
Quarter ended September 2026, filed figures.
The third-quarter 2026 report, out on 16 November, showed sales of $9.93 million, up from $0.51 million a year earlier but 56% below what analysts expected. The business lost $37.46 million before interest. Over nine months it raised about $2.0 billion and took $331.5 million of customer prepayments.
It also showed the size of the bill: $5.3 billion of commitments for chips and space, and a first $356 million loan secured on chips. The filing says more money must be found.
What it did to the thesisIt added signed demand and weakened confidence in timing and funding, so the case holds without getting stronger.
The quarter in detail
Sales growth from the previous quarter, filed figures. The company has restated its own history, so a quarter either side of the restatement may describe a different business; n/c marks one that does not compare.
Earnings per share $-1.14 against $-3.19 expected; $-1.14 as filed, with no adjustments.
The evidence under this - 8 claims
Some commitments are matched by customer prepayments still to arrive, which the 10-Q does not total.
At 9.95% and with a loan-to-value test on chips that lose value, the debt is costly and can force early repayment if GPU prices fall.
Operating cash flow was positive $208.2M over nine months because customers prepaid, and the company was in a net current asset position of $359.7M.
The price, and where we disagree
In short. Fair. Low against what the signed contracts would bill, but about $5.0 billion of equipment is unfunded, so how it is paid for decides what a share is worth.
The shares fell about 52% in October on reports of delays and a funding gap, and closed on 16 November 79.3% below their high of the past year. The third-quarter report came out after that close.
The next test is the fourth-quarter report, due about 24 February 2027: whether billing reached $400 to $450 million a year in January, and whether the chips for 2027 are paid for with loans rather than new shares.
The evidence under this - 5 claims
- Q3 revenue was 56% below the analysts' estimate and Q2's 65% below, so the company's own timing has run late twice.
- The January target needs about 12,000 GPUs commissioned in December and January.
At 9.95% and with a loan-to-value test on chips that lose value, the debt is costly and can force early repayment if GPU prices fall.
Customer prepayments of up to 30% of contract value (C03) and further GPU-backed loans planned for 2027 (C10) will fund part of the gap without new shares.
How we value it
No fair value: the analysts' forecasts show no lasting operating profit after stock pay, so a value built on earnings would rest on a guess; the bet prices it instead.
How we read the multiple
Count what a buyer of the whole company takes on: shares, convertible bonds, leases and what it owes for equipment, less cash. About $1.75 billion.
That is about 1.3 times what analysts expect it to sell in 2027, against about 2.7 times for CoreWeave. Valued as CoreWeave is, the price needs about a third of the contracted power billing.
But about $5.0 billion is unfunded. Borrowed against the chips, the shares are cheap; raised in shares near today's price, it could nearly double their number.
Against its own history. It has traded for less than a year, so there is no normal for it to be measured against.
The evidence under this - 6 claims
Leaving the payables out, as the vendor does, gives about $0.69B; they are counted because they are owed for GPUs already on the balance sheet.
CoreWeave is about forty times SharonAI's size and investment-grade lenders treat it differently, so its multiple flatters a small rival.
Roughly $5.0B of unfunded commitments (C32) will add debt or shares, so the value per share needs far more than a third of capacity billing.
Customer prepayments of up to 30% of contract value (C03) and further GPU-backed loans planned for 2027 (C10) will fund part of the gap without new shares.
Priced against its rivals
The competitors its own annual report names, each confirmed against the rival's filing. Measured on company value for each $1 of sales, cheapest first:
| Ticker | Company | company value for each $1 of sales |
|---|---|---|
| CRWV | CoreWeave | 8.9× |
| NBIS | Nebius Group | 15.0× |
| IREN | IREN Limited | 19.6× |
| SHAZ | SharonAI Holdings | 27.4× |
| AMZN | Amazon.com not in the median | 3.7× |
| GOOGL | Alphabet not in the median | 6.8× |
| MSFT | Microsoft not in the median | 11.2× |
The middle of the other 3 is 15.0. Amazon.com, Inc., Alphabet Inc., Microsoft Corporation compete with it in one product line inside a much bigger company, so their price says little about this business and they are not counted in the middle. SharonAI Holdings, Inc. is dearest of 4.
Rivals' figures come from a data provider, so read the spread of the group rather than any single pair.
On past sales it is the dearest of its rivals, but those sales came before its chips switched on; on next year's forecasts it is the cheapest.
Also checked and set aside: Applied Digital (a substitute, not the same product); Core Scientific (a substitute, not the same product); Hut 8 (a substitute, not the same product); Bitdeer Technologies Group (a substitute, not the same product); HIVE Digital Technologies (a substitute, not the same product).
The bet: what it pays if it works
This is a bet that all 129 contracted megawatts bill at about $15 million each a year. Valued as CoreWeave is, the company would then be worth about three times today's value, before whatever new debt pays for the chips.
If the build stalls for want of money, lenders and suppliers are paid first and the chips sell below cost: shareholders would keep at most about a third of today's value.
| Revenue in two to three years | Contracted megawatts billing needed | At the stated multiple |
|---|---|---|
| $425.00M | about 28 at 100% kept | 0.7× |
| $1.94B (the big win) | about 129 at 100% kept | 3.0× |
| $3.07B | about 205 at 100% kept | 4.7× |
To pay 2× (a Fair bet) it would need about $1.29B of revenue (about 86 contracted megawatts billing); to pay 5× (a Cheap bet), about $3.23B of revenue (about 216 contracted megawatts billing), at the same multiple.
The steps it has to take
- Done Take-or-pay contracts signed with ESDS, a global AI lab and a technology company, plus NVIDIA
- Done About $2.0B raised in nine months of 2026
- Done First GPU-backed facility, $356M
- Slipped Revenue on the analysts' schedule due 2026-09
- In progress About 12,000 GPUs commissioned due 2027-01
- In progress January billing of $400M to $450M a year due 2027-02
- Not started Further GPU loans for the 2027 GPUs, without new shares due 2027-06
- Not started NEXTDC's 72 MW phases switched on due 2027-12
The detail behind the label
The evidence under this - 5 claims
Management says newer contracts price above $19M a megawatt, which would raise the full figure to about $2.5B.
The NEXTDC order was amended and phased from 2027, so most of the contracted megawatts are not yet built or energised.
CoreWeave is about forty times SharonAI's size and investment-grade lenders treat it differently, so its multiple flatters a small rival.
New GPU debt ranks ahead of shareholders and is not yet on the balance sheet, so the figure is a ceiling, not an estimate of recovery.
Customer prepayments of up to 30% of contract value (C03) and further GPU-backed loans planned for 2027 (C10) will fund part of the gap without new shares.
Where we disagree with the market
Analysts forecast fast growth from the contracts; sellers doubt the company can pay for and deliver them.
The contracts would bill far more than the price needs; how the chips are paid for decides what a share gets.
Both camps are reading filed facts: $3.8 billion of obligations on one side, $5.3 billion of commitments and two missed quarters on the other. Neither alone settles it.
We would turn positive if January bills at $400 million or more and the 2027 chips are paid for with loans, and negative on a share sale or another miss.
What analysts say, and the evidence
The evidence under this - 7 claims
The NVIDIA arrangement, the largest contract by management's count, sits mostly outside this figure because its minimum depends on use.
Some commitments are matched by customer prepayments still to arrive, which the 10-Q does not total.
- Q3 revenue was 56% below the analysts' estimate and Q2's 65% below, so the company's own timing has run late twice.
- The January target needs about 12,000 GPUs commissioned in December and January.
Customer prepayments of up to 30% of contract value (C03) and further GPU-backed loans planned for 2027 (C10) will fund part of the gap without new shares.
What could go wrong, and how we would know
The risks
- The money runs short. About $5.0 billion of commitments are unfunded, and the filing says shares are among the ways to fill the gap.
- Late again. Sales missed forecasts by more than half in each of the last two quarters.
- Few, unnamed customers. Three customers and NVIDIA hold most of the contracts, and three of them are not named.
- Chips lose value. Rents for a chip fell about three-quarters once supply caught up before, and loans are secured on the chips.
- Bigger rivals. CoreWeave and the big clouds rent the same chips at far greater scale.
The evidence under this - 7 claims
Customer prepayments of up to 30% of contract value (C03) and further GPU-backed loans planned for 2027 (C10) will fund part of the gap without new shares.
Operating cash flow was positive $208.2M over nine months because customers prepaid, and the company was in a net current asset position of $359.7M.
Only ESDS is named; the other counterparties are anonymous and none has disclosed the contract in its own filing.
The same index fell roughly three-quarters from early 2024 to late 2025, which shows how fast a chip generation loses its rent once supply catches up.
At 9.95% and with a loan-to-value test on chips that lose value, the debt is costly and can force early repayment if GPU prices fall.
CoreWeave is about forty times SharonAI's size and investment-grade lenders treat it differently, so its multiple flatters a small rival.
What we still do not know
- What NVIDIA's arrangement actually guarantees; the agreement is not public.
- Who the unnamed AI lab and technology company are, and how strong their finances are.
- What the remaining chip loans will cost, and how much of the gap they cover.
- What management said on the third-quarter call; it was not yet published.
The evidence under this - 4 claims
- The 10-Q states no fixed aggregate value, so the $4.9B is management's reading of a minimum that depends on utilisation.
- NVIDIA's own filings do not name SharonAI.
Only ESDS is named; the other counterparties are anonymous and none has disclosed the contract in its own filing.
At 9.95% and with a loan-to-value test on chips that lose value, the debt is costly and can force early repayment if GPU prices fall.
Customer prepayments of up to 30% of contract value (C03) and further GPU-backed loans planned for 2027 (C10) will fund part of the gap without new shares.
What would change our view
- Neutral→AttractiveJanuary bills and the gap is borrowed
It needs all of:
- ○Billing of $400 million a year or more in January 2027.checked at the next results, due by 24 Feb
- ○The 2027 chips paid for with chip-backed loans and prepayments, with no new shares or convertible bonds.watched as it happens
- Neutral→UnattractiveBilling slips again or the build takes new shares
It needs any of:
- ○Billing below $300 million a year in January 2027.checked at the next results, due by 24 Feb
- ○New shares or a new convertible bond sold to pay for the 2027 chips.watched as it happens
- ○A contracted customer cancels, or NVIDIA's minimum goes unmet.watched as it happens
We would also be wrong if
- The 12,000 powered chips not switched on by March 2027.
What is coming
- 31 January 2027 About 12,000 GPUs commissioned: Whether energised capacity becomes billing capacity on management's schedule.
Every dated event ahead
- 31 January 2027 About 12,000 GPUs commissioned: Whether energised capacity becomes billing capacity on management's schedule.
- 24 February 2027 Q4 and full-year 2026 results and the 2026 annual report: January's billing pace against $400M to $450M, the funding raised for 2027 and any new contracts.
- 30 June 2027 Further GPU-backed loans for the 2027 GPUs: Whether the gap is funded with debt against GPUs and contracts or with new shares.
- 30 June 2027 The New Zealand AI lab contract begins billing: Whether the $1.32B contract starts on time.
Since we wrote this
- 17 NovThesis writtenFirst thesis: billions signed, the build unfundedNeutral with medium conviction: the price is low against contracted revenue, but about $5.0B of commitments are unfunded and revenue has slipped twice.
Sources 16
Sources - 16
Every claim above cites one of these, or states the reasoning that derived it. The quote is frozen at the date it was read: a link can rot, and an agent fetch returns a reading of a page rather than the page. Backs counts the whole claim ledger, so it can name a claim this page does not surface.
- S01primarycompany filing2026-03-31Item 1 business and competition (thirteen named rivals), the share classes, executive officers and pay, and the 2025 statementsbacksC01C13C24C37C40C42C43
- S02primarycompany filing2026-11-16the balance sheet, cash flows, commitments of $5.3B, the funding language, convertibles, the June financing, remaining performance obligations, the NVIDIA arrangement note and the share countbacksC01C02C03C04C06C07C08C09C11C12C13C15C16C21C22C25C41
- S03primarycompany release2026-11-16revenue, adjusted EBITDA, contracted capacity, total contract value, the ARR target, pricing per megawatt, energisation and contracts signed after the quarterbacksC02C04C05C10C17C18C19C20C21C22C23C26
- S04primarytranscript2026-08-06SharonAI second-quarter 2026 earnings call transcriptthe NVIDIA collaboration described as $4.9B, the contracting and prepayment model, and pricing on the NVIDIA resale programme
For take-or-pay contract, we are paid irrespective of whether they use the compute 100% of the time or 40% of the time
backsC03C15C20C21C43 - S05primarycompany filing2026-10-01the $356M facility in two tranches, its rates, 42-month term and loan-to-value covenantbacksC10
- S06primarycompany filing2026-09-17Situational Awareness LP, Schedule 13D/A on SharonAI Holdings Situational Awareness LP8,070,950 Class A shares, 21.1%backsC14
- S07primaryvendor data2026-11-17Research pack, frozen at content/SHAZ/research-pack.jsonthe 16 November close of $20.13, the 252-day range and returns, earnings history against estimates, FINRA short interest, analysts' revenue path, the insider window and the filed cohort's multiplesbacksC26C27C28C29C42
- S08primaryindustry data2026-04-02Synergy Research Group, neocloud market forecast Synergy Research Groupthe measured size of the specialised AI cloud market in 2025
neocloud revenues are scaling at an unprecedented pace, reaching $9 billion in Q4-up 223% year-over-year-and exceeding $25 billion for full-year 2025.
backsC33 - S09primarygovernment data2025-04-10data-centre electricity more than doubling to about 945 TWh by 2030 and accelerated servers growing 30% a year
Electricity consumption in accelerated servers, which is mainly driven by AI technology adoption (Box 2.1), is projected to grow by 30% annually in the Base Case
backsC34 - S10primaryindustry data2026-04-02SemiAnalysis, The Great GPU Shortage SemiAnalysisthe one-year H100 contract price series
H100 1-year GPU rental contract pricing has shot up almost 40% from a low of $1.70/hr/GPU in October 2025 to $2.35/hr/GPU by March 2026.
backsC35 - S11primaryindustry data2026-10-01$2.77 per GPU-hour on October 1, 2026 against $2.00 in December 2025
H100 cloud rates plummeted from around $8/hr in early 2024 to the $1.50-$3.00/hr range by late 2025 across specialist providers.
backsC35 - S12secondarynews2026-11-04the October fall and the reasons reporting gave for it
SharonAI now faces potential delays on its facilities along with a funding gap as it tries to scale what it calls AI factories.
backsC39 - S13secondarynews2026-06-12Converge Digest, 12 June 2026: Australia's Sharon AI signs NVIDIA deal for 40,000 GB300 GPUs Converge Digestthe 72 MW footprint and up to 40,000 GB300 GPUs under the NVIDIA collaboration
support the deployment of a 72MW AI factory infrastructure footprint in Australia
backsC15 - S14primarycompany filing2026-02-26a rival's own description of GPU cloud capacity sold to AI labs and enterprises on committed contractsbacksC36
- S15secondarynews2026-09-21the bear case as an analyst stated it; the note itself was not read
The firm flagged three converging pressures: falling GPU rental prices, hyperscalers scaling captive fleets, and rising financing costs on multi-billion-dollar buildouts.
backsC38 - S16primaryvendor data2026-11-09CoreWeave research pack, frozen at content/CRWV/research-pack.json CRWVCoreWeave's company value at the 9 November close and analysts' 2027 revenue of $26.47B from 19 analystsbacksC36