Forty-year horizon

Ten choices,forty years.

In Bali, spare capital goes into land, a villa, or a cafe. Outside Bali it goes into mining. Lately there is a new option, a data centre for someone else's AI. The easiest option is the one nobody brings up: a stock index or gold, bought from a phone. We build AI software instead. This page opens with twenty years that actually happened, then puts all ten choices on one axis and cuts each of them by its own odds of failure.

Every curve here is modelled from assumptions we wrote ourselves, not measured and not our own results. The figures they are anchored to come from official sources and are linked below.

The steepest curveis not why we chose.

01 · The question

Why not justbuy land too?

This is the question we get most, and it is a fair one. In Bali, the story of wealth is the story of land. Someone bought in Canggu a decade and a half ago and the price is now a multiple of what they paid. Everyone here knows at least one of those people.

The other versions come up just as often: open a cafe, build a villa, or with serious capital, go into mining. In the last two years a new option appeared that sounds closest to what we do, which is to build a data centre and rent it to whoever is running the AI.

We are not saying any of them is wrong. What this page shows is not which asset wins, but what you are actually buying when you pick one, and what changes when the horizon is pulled out to forty years.

Forty years is not chosen for drama. It is the distance at which compounding shows its real shape, and it is long enough to show that the final result is not decided by how fast something climbs.

02 · What actually happened

Twenty measured yearsbefore forty modelled ones.

Every other curve on this page is a model. This one is not. These are real year-end closing prices from 2005 to 2025, drawn twice: once in rupiah and once in US dollars, on the same scale, so the gap between the two panels is the currency and nothing else.

FIG. 05 The IDX Composite, the S&P 500 and gold, twenty years, in rupiah (index, year-end 2005 = 100, measured in rupiah) DATA

+Show the numbers

Year	IDX Composite	S&P 500	Gold (US$)	Rate (Rp/US$)
2005	1,162.64	US$1,248.29	US$517.10	Rp9,806
2006	1,805.52	US$1,418.30	US$635.20	Rp8,981
2007	2,745.83	US$1,468.36	US$834.90	Rp9,460
2008	1,355.41	US$903.25	US$883.60	Rp11,160
2009	2,534.36	US$1,115.10	US$1,095.20	Rp9,427
2010	3,703.51	US$1,257.64	US$1,421.10	Rp8,886
2011	3,821.99	US$1,257.60	US$1,565.80	Rp8,855
2012	4,316.69	US$1,426.19	US$1,674.80	Rp9,612
2013	4,274.18	US$1,848.36	US$1,201.90	Rp12,217
2014	5,226.95	US$2,058.90	US$1,183.90	Rp12,420
2015	4,593.01	US$2,043.94	US$1,060.30	Rp13,845
2016	5,296.71	US$2,238.83	US$1,150.00	Rp13,466
2017	6,355.65	US$2,673.61	US$1,306.30	Rp13,558
2018	6,194.50	US$2,506.85	US$1,278.30	Rp14,553
2019	6,299.54	US$3,230.78	US$1,519.50	Rp13,919
2020	5,979.07	US$3,756.07	US$1,893.10	Rp13,833
2021	6,581.48	US$4,766.18	US$1,827.50	Rp14,285
2022	6,850.62	US$3,839.50	US$1,819.70	Rp15,620
2023	7,272.80	US$4,769.83	US$2,062.40	Rp15,425
2024	7,079.90	US$5,881.63	US$2,629.20	Rp16,086
2025	8,646.94	US$6,845.50	US$4,325.60	Rp16,709

FIG. 06 The same three lines, twenty years, in US dollars (index, year-end 2005 = 100, measured in US dollars) DATA

+Show the numbers

Year	IDX Composite	S&P 500	Gold
2005	100	100	100
2006	170	114	123
2007	245	118	161
2008	102	72	171
2009	227	89	212
2010	352	101	275
2011	364	101	303
2012	379	114	324
2013	295	148	232
2014	355	165	229
2015	280	164	205
2016	332	179	222
2017	395	214	253
2018	359	201	247
2019	382	259	294
2020	365	301	366
2021	389	382	353
2022	370	308	352
2023	398	382	399
2024	371	471	508
2025	436	548	837

All three of these can be bought from a phone today, with no permit and no staff. Over the last twenty years, measured in rupiah, the IDX Composite returned 7.4 times, the S&P 500 9.3 times and gold 14.3 times.

In dollars all three are smaller: 4.4 times, 5.5 times and 8.4 times. Not one figure in the data changed, only the currency it is measured in. The rupiah panel leads because rupiah is what a reader here spends, and the dollar panel sits beside it because a page showing only the first would be booking a falling currency as performance.

The easiest thing to buy produced the largest number, in both currencies. That is not a fault in the chart. It is the result, and this page would be considerably easier to write if it said something else.

The gap between the panels is the currency

A dollar cost Rp9,806 at the end of 2005 and Rp16,709 at the end of 2025. The rupiah lost an average of 2.7 percent a year for twenty consecutive years. That is exactly the distance between the two figures above: the S&P 500 returned 8.9 percent a year in dollars and 11.8 percent in rupiah, and the 2.7 percent between them is not something the asset did. For an Indonesian holding a dollar asset it is a lift every year. For anyone holding only rupiah, the same number runs the other way.

Asset	Close 2005	Close 2025	Multiple, rupiah	A year, rupiah	Multiple, dollars	A year, dollars
IDX Composite	1,162.64	8,646.94	7.4x	10.6%	4.4x	7.6%
S&P 500	US$1,248.29	US$6,845.50	9.3x	11.8%	5.5x	8.9%
Gold	US$517.10	US$4,325.60	14.3x	14.2%	8.4x	11.2%

Neither stock line includes dividends

Both are price indices here, so both are understated by the same kind of amount and neither is handed an advantage the other is denied. With dividends reinvested the S&P 500 made 11.0 percent a year in dollars over this window rather than 8.9. The IDX Composite has generally yielded more than the S&P 500 across the period, so the Indonesian line moves up on that adjustment too, not down. Gold pays nothing, so its line needs no adjustment at all.

The thing everyone was buying instead

Over exactly the same twenty years, Bank Indonesia's residential property price index rose an average of 3.57 percent a year. That is the national index, and tourist-zone land in Bali has run hotter than it. But the national figure is the one that gets published, and the national figure sits below all three lines above.

After the window closed

The chart stops at year-end 2025 so that twenty years means twenty years. As of 8 September 2026 the IDX Composite stood at 6,686, down 22.7 percent from that close and 26.8 percent below its peak of 9,134 on 20 January 2026. At its low on 8 June it was 41.5 percent below that peak. The S&P 500 was at 7,674 and gold at US$4,394. The Indonesian line is in its deepest fall since 2008 while you read this.

This is the hardest part of the page to write, and that is precisely why it is here. The three lines above ask nothing of you. No staff, no permits, no Monday mornings. Over twenty measured years they beat almost anything you could have built yourself in Bali. What follows is not an argument against that. It is what those three lines do not show.

03 · The raw curves

Now the fortyyears ahead.

Ten assets, one index, a logarithmic vertical scale. Equal distance means equal multiple rather than equal rupiah. From here on everything is modelled rather than measured.

FIG. 07 Ten assets, one axis, forty years (index, starting at 100) MODEL

Scale

Through year

Move across the chart to read every asset at that year.

Ten lines at once is a full plot. There are only three grey steps, so identity is carried by the stroke style, the name at the end of each line, and the table below. Narrow it with Software only, or read any single year off the pointer.

+Show the numbers

Year	AI software	Bitcoin	IDX Composite	S&P 500	Gold	Data centres	Cafes & restaurants	Mining	Villas & hotels	Bali land
0	100	100	100	100	100	100	100	100	100	100
5	448	318	169	161	147	193	176	161	154	140
10	2,011	1,009	284	259	216	371	311	259	237	197
15	4,600	2,029	437	390	296	570	456	364	332	257
20	10,523	4,080	672	586	405	878	671	510	466	336
25	15,462	6,278	965	823	530	1,174	897	651	609	419
30	22,719	9,659	1,385	1,154	692	1,572	1,201	831	795	522
35	28,996	12,926	1,854	1,544	883	2,006	1,533	1,011	991	635
40	37,007	17,298	2,481	2,066	1,128	2,560	1,956	1,230	1,235	772

Stop here and the conclusion is easy: build software, done. The problem is that this chart assumes every asset survives to year forty. For land that is a reasonable assumption. For a cafe, and for a young software company, it is badly wrong.

04 · Cut by the odds of failure

Now multiply bythe odds of surviving.

Each curve is multiplied by the chance you still hold a working version of it at year forty. For a business an acquisition counts as surviving, because the owner still got paid.

FIG. 08 Value at year forty, before and after the cut (multiple of the starting stake) MODEL

Raw
After the cut

+Show the numbers

Asset	Raw	After the cut	Odds of surviving	A year, after the cut
AI software	370x	55.5x	15%	10.6%
Bitcoin	173x	104x	60%	12.3%
IDX Composite	24.8x	21.1x	85%	7.9%
S&P 500	20.7x	18.6x	90%	7.6%
Gold	11.3x	9.6x	85%	5.8%
Data centres	25.6x	14.1x	55%	6.8%
Cafes & restaurants	19.6x	1.0x	5%	-0.1%
Mining	12.3x	5.5x	45%	4.4%
Villas & hotels	12.4x	8.6x	70%	5.5%
Bali land	7.7x	7.0x	90%	5.0%

Two things change

The order flips. Bitcoin passes software, the cafe drops from sixth to last and is worth roughly the money that went in, and the two stock indices, which ask for no work at all, come third and fourth, above data centres, villas, land and mining. None of it is a typo and none of it is false modesty. All of it falls straight out of the assumptions we wrote ourselves, and we show it because this is the half that would change a reader's mind.

About the cafe

The cafe figure is the most important number on this page and the least comfortable one. It is not our opinion: it rests on the business survival tables, where only about half of restaurants are still trading past their fifth year, and on a net margin that averages a few percent of revenue. Forty years is eight of those five-year windows in a row.

05 · The worst year

Not onlyhow high.

An average annual rate is a comfortable number to read. What makes people sell at the worst possible moment is not the average, it is the worst year. Three rows here are measured rather than modelled: the IDX Composite, the S&P 500 and gold.

FIG. 09 Deepest fall, and how long it takes to get back (percent from peak. Price where a price exists, revenue where it does not) MODEL

AI software

Back to even: 4 years

Customers leave together, usually in the same bad economy that closed their shops first.

Bitcoin

Back to even: 3 years

An eighty-five percent fall from a peak has already happened, and each time it looked like the end.

IDX Composite

Back to even: 2 years

Down 61 percent in 2008 and back inside two years, the fastest recovery on this page. It is down 41.5 percent again from its January 2026 peak, and that one is not finished.

S&P 500

Back to even: 6 years

Down 57 percent between October 2007 and March 2009. It took five and a half years just to see the same number again.

Gold

Back to even: 9 years

The shallowest fall on this page and the slowest recovery. The August 2011 peak was not matched until July 2020, and nothing was paid out across those nine years of waiting.

Data centres

Back to even: 3 years

Empty racks when a large contract moves out, and hardware that ages whether or not it is being used.

Cafes & restaurants

Back to even: 5 years

The same number as the villa, because it was the same month. The difference is that a cafe has nothing to sell when it closes.

Mining

Back to even: 5 years

Thermal coal rose almost ninefold in twenty months and then gave back two thirds. You do not set the price.

Villas & hotels

Back to even: 5 years

2020 taught this to everyone in Bali. Empty rooms, and the loan still due every month.

Bali land

Back to even: 7 years

It looks calm because nobody quotes a price daily. The discount only shows up when you need to sell fast.

+Show the numbers

Asset	Deepest fall	Back to even	What causes it
AI software	-55%	4 years	Customers leave together, usually in the same bad economy that closed their shops first.
Bitcoin	-85%	3 years	An eighty-five percent fall from a peak has already happened, and each time it looked like the end.
IDX Composite	-61%	2 years	Down 61 percent in 2008 and back inside two years, the fastest recovery on this page. It is down 41.5 percent again from its January 2026 peak, and that one is not finished.
S&P 500	-57%	6 years	Down 57 percent between October 2007 and March 2009. It took five and a half years just to see the same number again.
Gold	-44%	9 years	The shallowest fall on this page and the slowest recovery. The August 2011 peak was not matched until July 2020, and nothing was paid out across those nine years of waiting.
Data centres	-45%	3 years	Empty racks when a large contract moves out, and hardware that ages whether or not it is being used.
Cafes & restaurants	-83%	5 years	The same number as the villa, because it was the same month. The difference is that a cafe has nothing to sell when it closes.
Mining	-65%	5 years	Thermal coal rose almost ninefold in twenty months and then gave back two thirds. You do not set the price.
Villas & hotels	-83%	5 years	2020 taught this to everyone in Bali. Empty rooms, and the loan still due every month.
Bali land	-35%	7 years	It looks calm because nobody quotes a price daily. The discount only shows up when you need to sell fast.

Land looks safest on this chart precisely because nobody prices it every day. Quiet is not the same as stable.

06 · What you can steer

Does your workchange it?

Everything above this point is arithmetic. What follows is the part that actually decided it for us, and it does not appear on any of the curves.

FIG. 09 Seven things the curves do not show (scored 0 to 4) MODEL

Measure	AI software	Bitcoin	IDX Composite	S&P 500	Gold	Data centres	Cafes & restaurants	Mining	Villas & hotels	Bali land
Capital to start Zero means billions of rupiah up front	3 / 4 3	4 / 4 4	4 / 4 4	4 / 4 4	4 / 4 4	0 / 4 0	1 / 4 1	0 / 4 0	0 / 4 0	0 / 4 0
Speed of exit From deciding to sell to holding the money	1 / 4 1	4 / 4 4	4 / 4 4	4 / 4 4	3 / 4 3	1 / 4 1	1 / 4 1	2 / 4 2	1 / 4 1	1 / 4 1
Your work changes the result Whether working harder moves the line at all	4 / 4 4	0 / 4 0	0 / 4 0	0 / 4 0	0 / 4 0	1 / 4 1	3 / 4 3	1 / 4 1	2 / 4 2	0 / 4 0
Grows past a room count Whether extra demand stops turning into revenue	4 / 4 4	4 / 4 4	4 / 4 4	4 / 4 4	3 / 4 3	1 / 4 1	1 / 4 1	0 / 4 0	1 / 4 1	1 / 4 1
Cost of one more customer Four means close to nothing	3 / 4 3	4 / 4 4	4 / 4 4	4 / 4 4	3 / 4 3	1 / 4 1	1 / 4 1	0 / 4 0	1 / 4 1	0 / 4 0
Proprietary data accumulates An asset nobody sells, only collects	4 / 4 4	0 / 4 0	0 / 4 0	0 / 4 0	0 / 4 0	1 / 4 1	1 / 4 1	0 / 4 0	1 / 4 1	0 / 4 0
Work for people in Bali Not a financial reason, and we do not dress it up as one	4 / 4 4	0 / 4 0	1 / 4 1	0 / 4 0	0 / 4 0	2 / 4 2	4 / 4 4	1 / 4 1	3 / 4 3	0 / 4 0

Scored 0 to 4 from the point of view of someone starting in Bali without outside capital. Four is best for that person, not best in general. A cafe scores badly on almost every row and top marks on local jobs, and both are true at the same time.

07 · The answer

Only one lineanswers to work.

Land does not care what you did on Monday

Land in Canggu appreciates because someone else is willing to pay more for it. You cannot make it climb faster by working harder. Bitcoin is exactly the same, and so is the coal price. All of them are assets you wait on, not assets you work.

Software climbs because something got built

Every feature that lands, every customer who decides not to leave, every price rise nobody objects to, changes the slope directly. That does not make it easier. It makes it earnable.

A data centre is AI, and still a building

This is the option closest to what we do, and we still passed on it. A data centre rents racks to the people building the AI. It has the same wall a hotel has, only measured in megawatts instead of rooms, and the data moving through it belongs to somebody else.

The index is the honest benchmark, not the enemy

If all you want is for money to grow while you do nothing, buy the index. We mean that as advice. Over the last twenty years the IDX Composite returned 7.4 times and gold 14.3 times, with no staff and no permits. We did not start a company to beat those numbers. We started one because an index cannot be worked on, and nothing we do on a Monday changes a single figure inside it.

A cafe is a job, not an asset

A cafe employs more people than anything else on this page, and that is genuinely worth something. But as a place to leave money for forty years, the survival tables say otherwise, and we are not going to hide the number.

Money is not the only thing that accumulates

Every correction a shop owner makes while using our product comes back as data only we hold. That asset appears on none of the charts on this page, and nobody sells it.

The money circulates here

A villa owned from abroad sends its rent out of Bali. A mine sends its output off the island. Software built here pays engineers who also live here. That is not a financial reason, and we will not pretend it is one.

How to read it

So the answer is not that software beats Bitcoin. Look at the third chart, it does not. The answer is that software is the engine that produces the money and Bitcoin is the tank the remainder sits in. Which one gets paid first is set out on the capital page.

Where the capital goes

08 · Assumptions & sources

Disagree withthe inputs, not the picture.

The first table holds every input behind the ten curves above. The second holds the published figures they are anchored to, with links. Change one input and the picture changes.

Asset	Rate per decade (%)	Odds of surviving 40y	Deepest fall	A year, over 40y
AI software	35 / 18 / 8 / 5	15%	-55%	15.9%
Bitcoin	26 / 15 / 9 / 6	60%	-85%	13.7%
IDX Composite	11 / 9 / 7.5 / 6	85%	-61%	8.4%
S&P 500	10 / 8.5 / 7 / 6	90%	-57%	7.9%
Gold	8 / 6.5 / 5.5 / 5	85%	-44%	6.2%
Data centres	14 / 9 / 6 / 5	55%	-45%	8.4%
Cafes & restaurants	12 / 8 / 6 / 5	5%	-83%	7.7%
Mining	10 / 7 / 5 / 4	45%	-65%	6.5%
Villas & hotels	9 / 7 / 5.5 / 4.5	70%	-83%	6.5%
Bali land	7 / 5.5 / 4.5 / 4	90%	-35%	5.2%

Read the decade rates in order: first decade, second, third, fourth. All are nominal, include income reinvested, and are before tax. Indonesian inflation keeps running underneath them, so the real figures are smaller than they look here.

The figures they are anchored to

Figure	What it measures	Source
3.57%	Average annual rise in Indonesia's residential property price index, 2003 to 2025	Bank Indonesia · BPS
0.83%	The same index in the fourth quarter of 2025, the slowest since the series began	Bank Indonesia
1.91%	Indonesian inflation averaged over 2025, to read the nominal figures above against	Bank Indonesia · BPS
-82.96%	Fall in foreign arrivals to Bali in 2020, from 6.3 million people in 2019	BPS Provinsi Bali
67.29%	Star-hotel occupancy in Bali, July 2026, the highest of any province	BPS
51%	Restaurants still trading past their fifth year	US Bureau of Labor Statistics
5%	Restaurant net margin, in a range of three to five percent of revenue	Datassential
US$425	Newcastle thermal coal at its May 2022 peak, up from US$45 a tonne in September 2020	IEA
-65%	The fall in that price from the peak through 2026	IEA · Trading Economics
58%	Bitcoin's annualised return over the last ten years	PortfoliosLab
-85.3%	Bitcoin's deepest fall from a peak, January 2015	PortfoliosLab
13.7%	Forecast growth in Indonesia's data centre market to 2031. Other houses publish 8 to 20 percent	Arizton
19.9%	Forecast growth in installed data centre power capacity to 2030	Mordor Intelligence
46.8%	Information-sector businesses still trading after five years	US Bureau of Labor Statistics
34.7%	All private-sector businesses still trading after ten years	US Bureau of Labor Statistics
10.55%	The IDX Composite annualised over twenty years, year-end 2005 to year-end 2025, before dividends	IDX · Trading Economics
-60.7%	The IDX Composite's deepest fall, from its January 2008 peak to its October 2008 low. It regained the peak in April 2010	IDX · Trading Economics
8.88%	The S&P 500 price index annualised over the same twenty years, in dollars, before dividends. With dividends reinvested the figure is 11.0 percent	S&P Dow Jones Indices
11.2%	Gold annualised over the same twenty years, in dollars per troy ounce	World Gold Council
-44.4%	Gold's deepest fall, from its August 2011 peak to its December 2015 low. It did not regain that peak until July 2020	World Gold Council
2.7%	Average annual fall in the rupiah against the dollar over the same twenty years, from Rp9,806 to Rp16,709	Bank Indonesia · JISDOR

An anchor is not a forecast. It fixes a defensible starting scale for each curve; the rate at which each one decays per decade is still our assumption. One asset here has no anchor for its rate at all, AI software, because the company does not exist yet. Only its odds of surviving are anchored. The last six rows are a different kind of thing: they are measurements taken from the twenty-year window in section 02, not assumptions.

Legal note

This page compares the economic shape of different asset classes. It is not an investment offer, not a solicitation to buy securities, and not investment advice.
Every curve here is computed from assumptions we wrote on this page. The figures they are anchored to are published by official sources; the curves are not. They are not measurements, not historical data, and not a promise of results.
Past performance of any asset class does not guarantee future results. Land, villas, cafes, mines, data centres, Bitcoin, gold, stock indices, and shares in a private company can all fall sharply, and several of them can fall to zero.
The twenty-year figures in section 02 are real year-end closing prices, not a model. They come from public market data: the IDX Composite, the S&P 500 price index, the dollar gold price per troy ounce, and the dollar to rupiah rate. Neither stock index there includes dividends, and the page states the size of that gap.
Stock indices and gold appear here as a benchmark, not as a product we sell. We are not an investment manager, not a broker, and not a gold dealer.
The Bitcoin referred to here is our own company treasury asset. We are not a crypto trading operator, we sell no crypto product, and we give no buy or sell advice to anyone.
Crypto assets carry high risk and can fall sharply. Supervision sits with OJK, and tax is calculated under the rules in force at the time of the transaction.
The business survival figures used here come from United States statistics, because no equivalent published series exists for Indonesia. We use them as they are rather than quietly adjusting them.

Want to argue with the numbers?

We open the assumptions, the model, and the measured figures that are not on this page.

Book 15 minutes

Next: where the capital goes