The Data Desk

Run the numbers yourself.

Every chart here is built from figures reported in The Buildout, and five of them run on live market and economic data. Move the sliders, change the yardstick, pick your entry point. The arithmetic is ours and labeled as ours, and every source is named.

The live macro tape

Five charts that test The Buildout against today’s Treasury, inflation, energy and profit data.

Data Desk · Live · Part 07Rates

The curve, live

Treasury yields from one month to 30 years, today against the day the Fed hiked. Pick a date and see which end of the curve moved.

Live data, updated Oct 8, 2026, 6:30 a.m. ET

The DeepStack read. Since the hike, the 30-year has moved +32 basis points and the 2-year +3. The Fed sets the short rate. The long end is setting the price of the buildout’s money.

Source: U.S. Treasury; Federal Reserve H.15 via FRED (as of Oct 8, 2026, 6:30 a.m. ET). Comparison dates use daily closes; changes are DeepStack arithmetic.

Show the data table
TenorToday, Oct 7A week ago (Sep 30)Fed hike day (Sep 16)A month ago (Sep 4)A year ago (Oct 7, 2025)
1M4.07%————
3M4.22%4.20%4.14%3.91%4.01%
6M4.28%————
1Y4.42%————
2Y4.77%4.88%4.74%4.37%3.57%
3Y4.87%————
5Y5.03%5.09%4.86%4.54%3.71%
7Y5.15%————
10Y5.28%5.29%5.01%4.78%4.14%
20Y5.71%————
30Y5.67%5.64%5.35%5.24%4.73%
Read Part 07: The Fed Hiked. The Long End Still Climbed.

Today, Oct 7Fed hike day, Sep 16

Change since Sep 16, basis points. Rising yields shown in red: dearer money for borrowers.

  1. 3M+8 bp
  2. 2Y+3 bp
  3. 5Y+17 bp
  4. 10Y+27 bp
  5. 30Y+32 bp

Since Sep 16, the 2-year has moved +3 bp and the 30-year +32 bp. The curve has steepened by 29 basis points: the long end is leading.

Data Desk · Live · Part 08Infrastructure Finance

The real price of patience

Each Treasury yield split into the real return and the inflation the market expects. Set your own inflation assumption and see what a lender actually earns.

Live data, updated Oct 8, 2026, 6:30 a.m. ET

The DeepStack read. A real yield near 2.9% is the floor under every loan in Part 08. If inflation runs at what households expect (3.0% over five years) rather than the 2.36% bonds price, lenders are underpaid and borrowers are quietly subsidized.

Source: U.S. Treasury (as of Oct 8, 2026, 6:30 a.m. ET). Household expectations: Federal Reserve Bank of New York Survey of Consumer Expectations, September 2026, via Trading Economics. Breakevens and real yields at your assumption are DeepStack arithmetic.

Show the data table
MaturityNominal yieldReal yield (TIPS)Inflation priced
5-year5.03%2.66%2.37%
10-year5.28%2.92%2.36%
30-year5.67%3.36%2.31%
ExpectationInflation, % a year
Bond market, 10 years2.36%
Households, next year3.90%
Households, 3 years3.30%
Households, 5 years3.00%
Read Part 08: AI Lenders Are Financing the End of Scarcity

10-year Treasury5.28%

Real yield the market pays2.92%

Inflation the market prices2.36%

Real yield at your inflation2.92%

Expected inflation, % a year

  1. Bond market, 10 years2.4%
  2. Households, next year3.9%
  3. Households, 3 years3.3%
  4. Households, 5 years3.0%

At 10 years, Treasuries pay 5.28%, or 2.92% after inflation protection, so the market prices 2.36% inflation. At your 2.36% assumption, a lender earns 2.92% in real terms, the same as inflation-protected bonds pay.

Data Desk · Live · Part 07Rates

Inflation’s last mile

Headline and core consumer prices against the 2% the Fed aims for, with the forecast path. Change the window and see how long the last mile has lasted.

Live data, updated Oct 8, 2026, 6:30 a.m. ET

The DeepStack read. Long-dated money was priced for inflation back at 2%. Headline CPI has run above it every month since January 2023, and the forecast path is still at 2.7% in Q3 2027. That gap is why Note 04 calls rates the referee.

Source: BLS; Trading Economics (forecast paths) (as of Oct 8, 2026, 6:30 a.m. ET). BLS.gov cannot vouch for the data or analyses derived from these data after the data have been retrieved from BLS.gov. The Fed’s 2% goal is set for the PCE index; CPI is shown against it as markets read it.

Show the data table
MonthHeadline CPICore CPI
Aug 20263.4%2.4%
Jul 20263.4%2.5%
Jun 20263.5%2.6%
May 20264.2%2.9%
Apr 20263.8%2.8%
Mar 20263.3%2.6%
Feb 20262.4%2.5%
Jan 20262.4%2.5%
Dec 20252.7%2.6%
Nov 20252.7%2.6%
Sep 20253.0%3.0%
Aug 20252.9%3.1%
Jul 20252.7%3.1%
Jun 20252.7%2.9%
May 20252.4%2.8%
Apr 20252.3%2.8%
Mar 20252.4%2.8%
Feb 20252.8%3.1%
Jan 20253.0%3.3%
Dec 20242.9%3.2%
Nov 20242.7%3.3%
Oct 20242.6%3.3%
Sep 20242.4%3.3%
Aug 20242.5%3.2%
Jul 20242.9%3.2%
Jun 20243.0%3.3%
May 20243.3%3.4%
Apr 20243.4%3.6%
Mar 20243.5%3.8%
Feb 20243.2%3.8%
Jan 20243.1%3.9%
Dec 20233.4%3.9%
Nov 20233.1%4.0%
Oct 20233.2%4.0%
Sep 20233.7%4.1%
Aug 20233.7%4.3%
Jul 20233.2%4.7%
Jun 20233.0%4.8%
May 20234.0%5.3%
Apr 20234.9%5.5%
Mar 20235.0%5.6%
Feb 20236.0%5.5%
Jan 20236.4%5.6%
QuarterForecast
Q4 20263.4%
Q1 20273.1%
Q2 20272.9%
Q3 20272.7%
Read Part 07: The Fed Hiked. The Long End Still Climbed.

Headline CPICore CPIForecast

Headline inflation was 3.4% in Aug 2026, core 2.4%. Since Aug 2024, headline has been above 2% in 24 of 24 months. The forecast path reaches 2.7% by Q3 2027.

Data Desk · Live · Part 06Power

The energy bill behind the buildout

Fuel, wholesale power and the metals in the grid, live. Switch the window, then sort by the biggest move or by input.

Live data, updated Oct 8, 2026, 6:30 a.m. ET

The DeepStack read. The cheapest input to the AI buildout is American gas, −9% over the year, while European power is +68% on average and copper +39%. Physics sets the schedule; these prices say where it gets built.

Source: EIA via FRED; Trading Economics (European gas); Trading Economics (European power); IMF via FRED; Trading Economics (uranium); Trading Economics (lithium) (as of Oct 8, 2026, 6:30 a.m. ET). Changes are DeepStack arithmetic on the stored closes. Copper is a monthly average (latest Jul 2026); its changes compare monthly averages, not weeks.

Show the data table
Price change over 12 monthsValue
US natural gas−9%
Brent crude (spot)+87%
WTI crude (spot)+54%
European gas (TTF)+142%
German power+74%
French power+86%
Spanish power+8%
Italian power+91%
UK power+83%
Copper+39%
Uranium+13%
Lithium carbonate+70%
Price change over one monthValue
US natural gas+3.8%
Brent crude (spot)+22.7%
WTI crude (spot)+3.8%
European gas (TTF)+4.4%
German power+22.4%
French power+35.3%
Spanish power−29.5%
Italian power−2.5%
UK power+10.5%
Copper−0.1%
Uranium+0.4%
Lithium carbonate−14.9%
Read Part 06: Power Doesn’t Depreciate. Its Premium Moves.

Price change over 12 months

  1. European gas (TTF)€79.16/MWh+142%
  2. Italian power€218.84/MWh+91%
  3. Brent crude (spot)$125.44 a barrel+87%
  4. French power€111.07/MWh+86%
  5. UK power£156.65/MWh+83%
  6. German power€167.47/MWh+74%
  7. Lithium carbonateCN¥124,950/t+70%
  8. WTI crude (spot)$96.24 a barrel+54%
  9. Copper$6.14/lb+39%
  10. Uranium$89.85/lb+13%
  11. Spanish power€106.90/MWh+8%
  12. US natural gas$3.03 per MMBtu−9%

Over 12 months, the biggest move is European gas (TTF) at +142%. US natural gas is −9%.

Data Desk · Live · Part 10Markets

Profits ran ahead of the economy

US corporate profits against output per hour, both indexed to 100. Move the starting line and watch the gap Part 10 calls the denominator.

Live data, updated Oct 8, 2026, 6:30 a.m. ET

The DeepStack read. Since the quarter before ChatGPT, profits are +39% and output per hour +9.1%. The S&P 500’s 19 times earnings is a bet that productivity closes that gap. Until it does, more of the profit line depends on the spending boom continuing.

Source: BEA via FRED; BLS via FRED (as of Oct 8, 2026, 6:30 a.m. ET). Quarterly. Indexing is DeepStack arithmetic.

Show the data table
QuarterCorporate profits (USD billion, annual rate)Output per hour (index)
Q2 20263,877120.0
Q1 20263,601119.6
Q4 20253,581119.3
Q3 20253,402118.9
Q2 20253,282117.4
Q1 20253,279116.2
Q4 20243,211116.4
Q3 20243,025116.0
Q2 20243,069115.0
Q1 20242,995113.9
Q4 20233,051113.9
Q3 20232,992112.8
Q2 20232,896111.4
Q1 20232,898110.3
Q4 20222,792110.0
Q3 20222,887109.4
Q2 20222,775109.4
Q1 20222,605110.2
Q4 20212,689111.6
Q3 20212,682110.9
Q2 20212,683111.6
Q1 20212,463111.4
Q4 20202,198110.6
Q3 20202,414111.5
Q2 20201,806109.8
Q1 20201,983104.7
Q4 20192,198105.1
Q3 20192,205104.1
Q2 20192,161103.0
Q1 20192,133102.3
Read Part 10: The Bubble Is in the Denominator

Corporate profitsOutput per hour

Index, Q4 2022 = 100 (dashed line)

Since Q4 2022 (before ChatGPT), corporate profits are +38.9% and output per hour +9.1%. Profits have grown 4.3 times faster.

Markets and rates

What the market pays for AI profits, and what the bond market charges for patience.

Data Desk · Part 10Markets

Run the denominator

The S&P 500 trades at 19.0 times forward earnings, which sounds reasonable until you set it beside a 10-year Treasury paying 5.31%. Cut the earnings, move the yield, and watch the cushion disappear.

The DeepStack read. The multiple is not the problem. The profits under it are. A 10% earnings miss at today’s prices turns 19.0x into 21.1x and leaves stocks yielding 57 basis points less than a riskless bond.

Source: FactSet (forward multiple and averages, as of Oct 2, 2026); CNBC (10-year close, Oct 5, 2026). Scenarios are DeepStack arithmetic, not forecasts.

Show the data table
ScenarioEarnings change10-year yieldMultipleEarnings yieldCushion
Today0%5.31%19.0x5.26%−5 bp
Earnings 10% lower−10%5.31%21.1x4.74%−57 bp
Howell’s 5.5% line0%5.50%19.0x5.26%−24 bp
Katsenelson’s 6%0%6.00%19.0x5.26%−74 bp
Miss and 5.5%−10%5.50%21.1x4.74%−76 bp
Read Part 10: The Bubble Is in the Denominator

Multiple at today’s prices19.0x

Stocks’ earnings yield5.26%

Cushion over Treasuries−5 bp

  • 5-year average19.8xtoday 0.8x cheaper
  • 10-year average19.1xtoday 0.1x cheaper
  • End of June20.4xtoday 1.4x cheaper

At 19.0x, stocks yield 5.26% on forward profits against 5.31% on the 10-year. Treasuries pay 5 basis points more than stocks. The cushion is gone.

Data Desk · Part 07Rates

The 10-year ladder

The Fed raised its rate on September 16. The long end kept climbing anyway. Here is where the 10-year stands against the levels investors say will break something.

The DeepStack read. The move is real, not inflationary: 96 of the 104 basis points since February came from real yields. That is the market charging more for money, not panicking about prices, and it is the harder kind of rise to reverse.

Source: Federal Reserve (Sep 16 decision); FRED (10-year and real yields, Sep 16 and Sep 24, 2026); CNBC (Oct 5 close). Thresholds as stated by the investors named.

Show the data table
LevelWhat it isDateSource
3.75–4.00%Fed funds rangeAfter Sep 16Federal Reserve
5.01%10-year on hike daySep 16FRED
5.18%10-year, six trading days laterSep 24FRED
5.31%10-year closeOct 5CNBC
5.00%The RubiconThresholdSteve Eisman; Ren
5.50%Where cycles buckleThresholdMichael Howell
6.00%Not even at 6%ThresholdVitaliy Katsenelson; Michael Howell
Read Part 07: The Fed Hiked. The Long End Still Climbed.

Where it is

Where it breaks

10-year close

The highest close in roughly two decades.

Source: CNBC

The 104 bp climb from the late-February low, as of Sep 24

96 bp

Real yield: 96 bpExpected inflation: 8 bp

Real 10-year yield vs. the TSCS kill switch

2.85% on Sep 24Kill switch 3.15%: 30 bp away

10-year close, Oct 5: 5.31%. The highest close in roughly two decades.

Data Desk · Part 02Capital Markets

SpaceX: it depends when you bought

Eight weeks, four prices. Pick your entry point and see what the market paid you for believing.

The DeepStack read. The buyer at the offer price is up 8%. The believer who bought the peak is down 35%. The market punished the story long before the business missed a target.

Source: Market data through the August 12, 2026 close, as reported in Part 02. Returns are DeepStack arithmetic.

Show the data table
PointWhenPriceReturn to close
IPOJun 12$135.00+8.3%
PeakSummer high$225.64−35.2%
LowEarly August$104.83+39.4%
CloseAug 12$146.15—
Read Part 02: SpaceX Is Funding an AI Bet Before the Proof

If you bought at

Bought at the IPO ($135.00): up 8.3% at the Aug 12 close of $146.15.

Credit and collateral

The promises lenders are underwriting, measured against the cash that has to keep them.

Data Desk · Part 08Infrastructure Finance

Commitments versus cash

Lenders are underwriting a promise that hyperscaler cash flow will roughly triple by 2030. Pick a yardstick and see how far the commitments already run ahead of it.

The DeepStack read. Burry’s $2.7 trillion of leases and purchase commitments is 4.5 times what the hyperscalers generate in operating cash today, and more than six and a half times a year of earnings. The bonds treat the 2030 target as a formality. It is the whole bet.

Source: Michael Burry (commitments and earnings, five hyperscalers); Torsten Slok, Apollo (operating cash flow and 2030 path); Apollo via DeepValue Capital (backlog), September 2026.

Show the data table
MeasureValueSource
Hyperscaler earnings, past yearunder $0.4TMichael Burry
Operating cash flow today$0.6TTorsten Slok, Apollo
Cash flow needed by 2030$2.0TTorsten Slok, Apollo
Hyperscaler backlog$2.3TApollo via DeepValue Capital
Off-balance-sheet commitments$2.7TMichael Burry
Read Part 08: AI Lenders Are Financing the End of Scarcity

Measure everything against

  1. Hyperscaler earnings, past yearMichael Burryunder $0.4T0.7× cash flow today
  2. Operating cash flow todayTorsten Slok, Apollo$0.6Tyardstick
  3. Cash flow needed by 2030Torsten Slok, Apollo$2.0T3.3× cash flow today
  4. Hyperscaler backlogApollo via DeepValue Capital$2.3T3.8× cash flow today
  5. Off-balance-sheet commitmentsMichael Burry$2.7T4.5× cash flow today

Measured against operating cash flow today ($0.6T), off-balance-sheet commitments are 4.5 times, and cash flow needed by 2030 are 3.3 times that yardstick.

Data Desk · Part 03Credit

The cloud revenue hurdle

To justify next year’s spending, cloud revenue has to reach about $960 billion. It runs at about $380 billion today. Choose a growth rate and see how long the climb takes.

The DeepStack read. At the 40% growth the bulls cite, the climb takes almost three years. The spending it has to pay for arrives next year. That gap is what the debt is financing.

Source: Company-reported backlog and commitments; revenue hurdle and run rate as cited in Part 03, August 15, 2026. Growth paths are DeepStack arithmetic.

Show the data table
Growth rateTime to reach $960B from $380B
20% a year5.1 years
30% a year3.5 years
40% a year2.8 years
50% a year2.3 years
60% a year2.0 years
Contracted revenueValue
AWS backlog$496B
Google Cloud backlog$514B
Microsoft commercial commitments$678B
Read Part 03: The Machines Will Work. The Debt Might Not.

Years to the hurdle2.8

Run rate today$380B

Booked across three clouds$1.69T

At 40% a year, cloud revenue needs about 2.8 years to grow from $380 billion to $960 billion. The spending it has to justify lands next year.

Spending and power

How fast the money is going out, and what the physical bottleneck now costs.

Data Desk · Part 04Capex

The capex clock

Amazon, Google, Microsoft and Meta spent $165 billion on capital projects in three months. That is a pace you can watch.

The DeepStack read. Quarterly spending is up almost fivefold in three years, and new borrowing now equals about a third of it. When the cash machines start borrowing to feed the machines, the cycle has changed character.

Source: Prof G Markets (quarterly capital spending, Aug 4, 2026); Hayden Capital (incremental debt as a share of capital spending). Earlier quarters are implied by the reported growth rates. The clock is DeepStack arithmetic at last quarter’s pace.

Show the data table
QuarterCapital spending
Same quarter, three years earlier (implied)$33.5B
Same quarter, a year earlier (implied)$88.2B
Latest quarter$165.0B
PeriodNew debt as share of capital spending
20249%
Last 12 months32%
Read Part 04: AI Spending Has Met Its First Real Bill

Spent by the four since you opened this page

$0

  1. Same quarter, three years earlierimplied$33.5B
  2. Same quarter, a year earlierimplied$88.2B
  3. Latest quarter$165B

Per day$1.81B

Per second$20,928

New debt as share of capex9% → 32%2024 → Last 12 months

At last quarter’s pace the four spend about $1.8 billion a day, or roughly $20,900 every second.

Data Desk · Part 06Power

What power costs now

Two prices tell the story of the bottleneck: what the grid charges to guarantee capacity, and what a buyer pays to get compute now rather than later.

The DeepStack read. Grid capacity prices rose more than elevenfold in three auctions, and buyers pay double to skip the queue. Power is not a cost line in this cycle. It is the scarce asset, priced by whoever needs it soonest.

Source: PJM capacity auctions; IEEFA (63% of the increase, $9.3 billion, attributed to data centers); Nebius pricing as described by Michael Burry, August 12, 2026.

Show the data table
PJM capacity price, dollars per megawatt-dayValue
2024/25 auction$28.92
2025/26 auction$269.92
2026/27 auction$329.17
Nebius compute price, millions of dollars per megawattRange
One-to-three-year contract$20M to $25M
Delivery within six months$40M to $50M
Read Part 06: Power Doesn’t Depreciate. Its Premium Moves.

PJM capacity price, dollars per megawatt-day

  1. 2024/25 auction$28.92
  2. 2025/26 auction$269.92
  3. 2026/27 auction$329.17

PJM capacity cleared at $329.17 a megawatt-day in the latest auction, 11.4 times the $28.92 of two auctions earlier. IEEFA attributes 63% of the jump to data centers.

Data Desk · Part 01Chips

Orders outrun shipments

In a single quarter, Broadcom booked nearly three dollars of AI chip orders for every dollar it shipped. The money to pay for that demand is being raised in tranches.

The DeepStack read. A 2.8-to-1 book-to-bill is demand. Seventy-three billion dollars of private debt arranged in ten weeks to fund the hardware is something else: a sign that the buyers’ own cash is no longer enough.

Source: Invest In Assets (Broadcom orders and shipments, mid-2026); Michael Burry (Apollo transaction, late May); Apollo and Blackstone first tranche, as reported in Part 01.

Show the data table
Broadcom AI semiconductors, one quarter, billions of dollarsValue
AI chip orders booked$30.0B
AI chips shipped$10.8B
Private debt raised for AI compute, billions of dollarsValue
Apollo, late May$38B
Apollo and Blackstone$35B
Read Part 01: Nvidia’s Demand Is Real. Its Financing Is Fragile.

Broadcom AI semiconductors, one quarter, billions of dollars

  1. AI chip orders booked$30.0B
  2. AI chips shipped$10.8B

Broadcom booked $30 billion of AI semiconductor orders in one quarter against $10.8 billion shipped: 2.8 dollars of orders for every dollar delivered.

The AI business

Who is actually making money from the models, and who the agents are coming for.

Data Desk · Part 05AI Labs

The price of the same job

One coding task, three models. On a linear scale the cheapest one all but disappears. Switch to a log scale, then dial up the volume to see a buyer’s monthly bill.

The DeepStack read. Same task, a 64-fold price gap. A general-purpose model cannot charge frontier prices for work a router can send elsewhere. The moat has to be the work only you can do.

Source: Wyndo (cost of the same coding task, August 9, 2026), as reported in Part 05. Monthly bills are DeepStack arithmetic.

Show the data table
ModelCost per task1,000 tasks10,000 tasks100,000 tasks1,000,000 tasks10,000,000 tasks
Claude Opus 5$0.32$320.00$3,200$32,000$320,000$3.20M
Kimi$0.24$240.00$2,400$24,000$240,000$2.40M
DeepSeek$0.005$5.00$50.00$500.00$5,000$50,000
Read Part 05: OpenAI and Anthropic Are Not the Same Business
  1. Claude Opus 5$0.32 per task$320,000a month
  2. Kimi$0.24 per task$240,000a month
  3. DeepSeek$0.005 per task$5,000a month

Linear scale: bar length is proportional to price.

At 1,000,000 tasks a month: Claude Opus 5 $320,000, Kimi $240,000, DeepSeek $5,000. The most expensive model costs 64 times the cheapest.

Data Desk · Part 05AI Labs

Two labs, two businesses

OpenAI and Anthropic are usually traded as one bet on AI demand. Their second quarters say they are not the same business.

The DeepStack read. One lab grew 142% and reported adjusted operating income of $559 million. The other added about $3 billion of costs to win $1 billion of new revenue. Lending to them as a single credit is a category error.

Source: Michael Spencer (quarterly revenue; Anthropic adjusted operating income, preliminary); Steve Eisman (OpenAI costs). Anthropic’s first quarter is implied by its reported 142% growth.

Show the data table
Quarterly revenue, billions of dollarsQ1 2026Q2 2026
OpenAI$5.7B$6.7B
Anthropic$4.8B (implied)$11.6B
Revenue growth, second quarter over firstValue
Anthropic142%
OpenAI18%
OpenAI, change from the first quarter, billions of dollarsValue
New revenue$1.0B
Added costs$3.0B
Read Part 05: OpenAI and Anthropic Are Not the Same Business

Quarterly revenue, billions of dollars

Q1 2026Q2 2026

  1. OpenAI$5.7B$6.7B
  2. Anthropic$4.8Bimplied$11.6B

OpenAI went from $5.7 billion to $6.7 billion. Anthropic reached $11.6 billion, from an implied $4.8 billion.

Data Desk · Part 09Agents

Winners and losers at the first checkout

From the day Meta launched its shopping agent to October 2, the market sorted companies into two piles: the ones building agents, and the ones whose customers agents might reroute.

The DeepStack read. Every habit business in the sample fell between 11% and 19% while the index went nowhere. The market is not waiting for agents to steal customers. It is pricing the possibility now.

Source: Financial Modeling Prep, closing prices from September 8 to October 2, 2026.

Show the data table
Change in closing price, Sep 8 to Oct 2, 2026Value
Advanced Micro Devices+33%
Meta+18%
S&P 5000%
Expedia−11%
Netflix−14%
Bank of America−14%
Intuit−16%
Planet Fitness−17%
Booking Holdings−18%
Comcast−19%
Read Part 09: Habit Was the Moat. Agents Are Draining It.

Change in closing price, Sep 8 to Oct 2, 2026

  1. Advanced Micro Devices+33%
  2. Meta+18%
  3. S&P 5000%
  4. Expedia−11%
  5. Netflix−14%
  6. Bank of America−14%
  7. Intuit−16%
  8. Planet Fitness−17%
  9. Booking Holdings−18%
  10. Comcast−19%

Advanced Micro Devices rose 33% and Meta 18%. Booking fell 18%, Comcast 19% and Planet Fitness 17%. The S&P 500 was flat.

The tape, expanded

Every number in The Buildout.

40 figures from 10 stories, each with its date and context. Filter by beat, or search for a company, a rate or a name.

40 of 40 numbers

How to read these charts