Pinned Tweet
Bet The Decade: one company at a time, ten years ahead.
Our goal is simple. Tell the stories of the companies building the next decade (AI, chips, power, space, biotech, quantum, fintech) and show what has to go right, what could break it, and which numbers to watch. Built on official filings, not headlines.
Free tools, all running on your own assumptions:
📈 10 Years Forward: pick a company, load the BTD bear, base or bull scenario or set your own growth, margin and multiple, and see the math for 2031 or 2036 vs today's market cap.
📅 Earnings Calendar: announced results dates with a live countdown in your time zone and calendar files.
⚖️ Compare Companies: two to four companies side by side. Revenue, growth, margins, cash, debt and shares from the latest SEC filings.
🧮 Portfolio Growth Model: enter a portfolio and see illustrative bear, base and bull scenarios for 2032 and 2036.
Try them: betthedecade.com/tools/
🎬 Full company stories on YouTube: youtube.com/@BetTheDecadeTV
Follow for the next ten years. NFA.
$SPCX (SpaceX) gets another national-security launch extension: the Pentagon says SpaceX, ULA and Blue Origin received NSSL Phase 3 Lane 2 contract modifications with a combined ceiling of up to $985M.
Key facts:
• $985M is the shared maximum for all three providers, not SpaceX alone; no split disclosed
• No funds obligated at award; work runs through April 2033
• Contracting office: Space Systems Command
• Base Lane 2 awards (Apr 2025): SpaceX ~$5.92B, ULA ~$5.37B, Blue Origin ~$2.39B; SpaceX was slated for ~28 missions (~60%) in FY25–FY29
Why it matters: Lane 2 is the heavy, highest-assurance tier of US national-security launch. Each modification extends a multi-year government revenue line where reliability and cadence decide share, and reusable Falcon cadence is SpaceX's edge.
The headline number is a ceiling, not cash.
NFA
7
$SPCX: Nvidia added 7.7M SpaceX shares in Q3, while Ontario Teachers sold ~41% of its stake after the lockup.
Same filing day, two very different holders:
• Nvidia 13F (as of Sep 30): 130.5M SpaceX shares, up from 122.8M (+6.3%), worth ~$19.7B. $CRWV (CoreWeave) stake unchanged at 47.2M shares. Intel cut 25%, Nokia cut ~56%.
• Ontario Teachers' Pension Plan sold ~20.8M shares between August and late September (~51M to ~29.9M), per The Globe and Mail. Its reason: to "crystallize returns", while staying "confident about the company's future prospects".
The mechanism: the company selling the world's AI compute keeps adding exposure to the company that owns launch, Starlink and xAI. Early holders taking profit after a lockup is supply, not a verdict on the business. More lockup tranches run until Dec 8.
Note: a 13F is a Sep 30 snapshot and does not say how the shares were added.
h/t @SpaceInvestor_D | via SEC 13F, The Globe and Mail
NFA
15
$TSLA wants Europe to build robotaxis, not just cars.
Tesla plans a ~EUR 3B (~$3.3B) expansion at Giga Berlin in Grünheide, with a new hall able to produce up to 1 million Cybercabs. Construction is planned for 1H 2027 and production for 2029, with the permit application still due in 2026, via Reuters, Handelsblatt and dpa.
That would make Europe a second Cybercab hub. Robotaxi economics come from autonomy software margin per mile, not unit count, so the hall matters as distribution for the software. The AI is the product; the car is how it reaches the road.
NFA
16
$CRWV trained DeepSeek-V3 671B in 17.65 min on 256 NVIDIA Rubin GPUs in MLPerf Training v6.1, 2.7x faster than its own 256-GPU GB200 run (48.40 min), via MLCommons and CoreWeave.
- First MLPerf round with Rubin, published Nov 16; 4 Vera Rubin NVL72 racks running in CoreWeave's production environment (Spectrum-X Ethernet / ConnectX-9)
- Per CoreWeave, the only cloud provider with multi-rack Vera Rubin results; other cloud submitters included Azure, Oracle, Nebius and Lambda
- Single rack, 72 GPUs, Llama 3.1 8B: 7.16 min vs 13.70 min on GB200 (1.9x), which CoreWeave calls the fastest 72-GPU submission of the round
- The 2.7x is Rubin vs GB200, not GB300 (NVIDIA cites up to 2x vs GB300)
Same GPU count, roughly 2.7x less training time: customers get more training runs per rack-hour. As frontier models keep scaling, the labs that iterate fastest pull ahead, and they go where the newest generation already runs in production. Being first to operate a new generation at multi-rack scale is how a neocloud wins capacity contracts.
Watch: this is a benchmark, not a contract or revenue. The lead is temporary as hyperscalers get Rubin too, and the buildout is funded by capex and debt.
NFA
2
38
My $GRAB 2036 price target is $100.
Rides, food delivery, payments and lending for Southeast Asia in one app. Every trip and order feeds the data that powers the fintech side.
NFA.
How we get there (scenario model, not advice):
$GRAB 2036 base ~$88, bull ~$157
Revenue × net margin × P/E, with dilution included.
Run the numbers yourself: betthedecade.com/tools/growt…
Full $GRAB story on YouTube: youtu.be/fBSSTlPUeDM
NFA
3
18
Five quarters ago $IREN was a bitcoin miner. The mix has already flipped: AI Cloud is the large majority of revenue.
This is not a recap of last week's print. The decade test is whether that mix holds when the next megawatts have to be energized — and whether the company's own ARR plan, already moved to early 2027, still has a contracted, financed date.
New episode, full breakdown below. NFA
1
3
29
Bet The Decade retweeted
Could not make UBS and Jeffries conference but our team did a great job ..revealed that the Anktiva plus BCG resulted in a highly significant improvement of efficacy in patients who were BCG naive with bladder cancer compared to BCG alone..at p value less than 0.01. Immunotherapy 2.0 slowly but surely.
$NVDA (Nvidia): Washington just closed the AI-chip rental loophole, one day before Nvidia's first earnings without Jensen Huang as CEO.
What the Commerce Department's new BIS rule (interim final, issued today) does:
• Remote access now counts. Firms headquartered in countries of concern can no longer rent advanced US AI compute in third-country data centers to train models.
• Advanced chips shipped to Indonesia, Malaysia, Singapore, Thailand and Brazil now need a license. The rule calls them higher-than-normal diversion risk, and more countries can be added.
• Written at the direction of Treasury and the National Security Council.
• Per CNBC, it is the first US policy response to the scandal that led to Huang's resignation last week.
The mechanism: export controls used to follow the box. Now they follow the workload. A cloud contract in Singapore gets the same scrutiny as a server shipment.
Our read: more paperwork and time in Southeast Asia, while compute built in the US by neoclouds such as $CRWV and $IREN is not the target of this rule.
Question for tomorrow's call: how much data-center revenue sits in those five markets?
via BIS, CNBC. NFA
4
126
$TE (T1 Energy) just set up an at-the-market program to sell up to $100M of common stock, with Roth Capital as agent (424B5 + 8-K, Nov 16).
What the filing says:
• Proceeds: construction and equipment for Phase 1 of the G2_Austin cell fab, plus working capital
• The company says the ATM should at least partly satisfy the equity condition of the G2_Austin debt financing it is negotiating
• Agent commission up to 3.00%
• Last close cited: $1.91 (Nov 13)
The math: at $1.91, $100M would be roughly 52M new shares, about 17% of the ~299.5M shares outstanding on Sep 30. Real dilution depends on how much is sold and at what price; the company does not have to use all of it.
Our read: this is the price of the next step. Equity unlocks the debt package, the debt funds G2, and G2 is what turns @T1Energy from a US module maker into a US cell maker. The test now is whether that debt deal closes soon after.
NFA
About 90% of the silicon in a top AI chip package is memory, not the processor (Micron, Hot Chips 2026).
That is why memory is the quiet bottleneck of the AI build-out. SK hynix said on its Q3 call that its DRAM and HBM volumes are effectively sold out through 2029.
The mechanism: every jump in model size needs more bandwidth per GPU, so memory supply, not just GPU supply, sets how fast AI capacity can grow. Worth keeping in mind as $NVDA reports tomorrow.
Full breakdown in our episode below. NFA
$NVDA reports Tuesday, 17 November: the first print since the CEO change. The useful questions are about demand, not the org chart.
Listen for:
1. The Data Center split
2. Training vs. inference demand commentary
3. Inventories and supply commitments, the plumbing of the AI buildout
That print is a barometer for the whole stack: clouds, power, and everyone building around those chips. Guidance is the company's plan, not a forecast.
Our episode on how Nvidia got here is in the reply. NFA
1
3
25
$SPCX (SpaceX): Ship 42, the first Starship caught from orbit, "could probably be reflown with a few weeks of refurbishment," per Musk after inspecting it. Hours later he gave the first Starship V4 numbers.
What was said:
• Ship 42 heat shield and engines "look great"; airplane-like reuse with zero refurbishment still needs "many small fixes" (per Musk)
• Starship V4: ~70 m long (~40% longer), >60% more propellant, 9 engines incl. 3 extra Raptor Vacuums, higher-thrust V4 Raptors (per Musk; no official SpaceX spec yet)
Why it matters for the business:
1. Turnaround is the cost line. A ship that flies again in weeks instead of being rebuilt turns each launch from a hardware expense into refurbishment + propellant.
2. More propellant per ship means fewer tanker flights per Moon or Mars mission, so every tonne delivered beyond orbit needs fewer launches.
3. Together they set the cost of putting Starlink V3 and future orbital compute in space, the layer AI infrastructure off-planet would run on.
Our read: the decade thesis rests on cadence, not single flights. Flight 15 proved the catch; the next proof point is how fast a caught ship flies again. These are CEO statements, not a schedule.
NFA
ULA's Vulcan failed to reach orbit today with 40 Amazon Leo satellites on board: the first total mission failure in ULA's 20-year history.
What we know:
• Anomaly ~10 minutes into flight; reports point to the Centaur upper-stage burn, ULA says it has no specific data yet
• Vulcan's first flight in 9 months, after a solid-booster nozzle stand-down
• It was the first of 38 Vulcan launches Amazon bought for Leo
• Leo would have reached 436 satellites launched; the full network needs 3,300+
Why it matters for $SPCX:
Amazon's two heavy lifters are now both on the ground at once. Vulcan waits on a failure investigation, New Glenn is still returning to flight after its May pad explosion. Leo also has 13 Falcon 9 launches booked.
The mechanism: in LEO broadband the moat is launch cadence. Starlink rides a reusable rocket SpaceX owns and flies several times a week, and on Saturday Starship caught its upper stage returning from orbit. Competitors rent rides on rockets that still have to prove reliability.
Launch is the bottleneck of the space economy, and only one company controls its own.
NFA
$NVDA succession is now a Washington story.
Treasury Secretary Scott Bessent said Sunday that Nvidia's board should "go outside the company" for its next CEO and that it needs "a dramatic shakeup." He also said the board and outside counsel must answer questions about export-control compliance within the next 10 days.
Context: Jensen Huang stepped down Thursday and CFO Colette Kress is interim CEO. The FT reported the board approached Arm CEO Rene Haas; neither company has confirmed it.
Why it matters: Nvidia reports this week. The first call without Jensen now also carries a policy question: how much China-adjacent demand is still sellable, and how tight the next export rules get.
For AI-infra builders like $IREN, $NBIS and $CRWV, the read-through is GPU supply, delivery timing and pricing, not the name on the CEO door.
Bessent's words are pressure, not a board decision. via Reuters, Bloomberg
NFA
4
33
Thesis check $SPCX $ONDS $RKLB $TE $WULF: five portfolio theses against this week's data (Nov 9–15).
$SPCX: full reuse is the cost unlock. Status: on track.
Flight 15: Ship 42 reached orbit, deployed 26 Starlink V3, flew 14 orbits and was caught by the tower for the first time. Super Heavy cut its boostback short on a low fuel mass alarm and was diverted to a Gulf splashdown.
Open: the first V3 booster catch, and how fast a caught ship flies again.
$ONDS: backlog turns into revenue. Status: on track.
Q3: $167.5M revenue, $214M bookings (book-to-bill above 1), $835M backlog, FY27 guide of at least $1.1B.
Open: gross margin (21.8% in Q3) as that backlog converts.
$RKLB: cadence funds the next rocket. Status: on track.
Electron flew its 100th mission on Nov 14 (iQPS).
Open: Neutron's first flight.
$TE (T1 Energy): US solar build-out plus 45X credits. Status: still early.
A 13D/A shows Encompass committed to buy up to $15M of stock, amount at T1's choice, expiring Nov 16. The 10-Q discloses an SEC inquiry tied to 45X (an inquiry, not a charge).
Open: whether T1 uses the commitment, and how the inquiry ends.
$WULF: power is the AI bottleneck. Status: still early.
Kentucky Power filed for up to 1 GW for the Muskie campus by 2029. A filing is not an approval; a decision comes mid-2027 at the earliest.
Reader question of the week: "Is Encompass actually buying, or only committed?"
Only committed so far. The filing sets a ceiling, not a purchase.
Which thesis should we stress-test next week?
NFA
3
28

