InvestMoat

The next 5, 10 and 20 years

Every moat score on this site is a bet on what the world looks like a decade from now. This page writes those bets down: seven themes, where each one stands today, one call for each horizon, dated checkpoints we mark as they come due, and the observation that would prove us wrong. The names beside each theme carry their live moat score, so when a view changes, the scores show it first.

7 themes·15 checkpoints·51 names·23 of 25 IM25 holdings named
  1. 01AI compute moves from training to running
  2. 02Power becomes the bottleneck, not chips
  3. 03Agents do the work, the record keeps the moat
  4. 04Robots leave the cage
  5. 05The rails stay, the money changes
  6. 06Supply chains become strategy
  7. 07Medicine treats the chronic, earlier
2031
5 years out

What the current build-out turns into. Most of it is already contracted, permitted or in production.

2036
10 years out

Where the moats we rate today either compound or get competed away. The horizon our moat trajectory scores.

2046
20 years out

The direction of travel, not a forecast. The bets a 20-year holder is quietly making by owning the book.

IM25 holdingNumber is today's moat scoreWideningNarrowing

Every theme carries dated checkpoints, so this page can be marked like the IM25 is. We change a status when the evidence does and never delete a checkpoint that went against us.

1
Hit
4
On track
10
Watching
0
Off track
2026202720282029203020312032203320342035NOW2027 · Agents do the work, the record keeps the moat: The shake-out comes first: a large share of early agent projects is cancelled. (watching)2028 · Agents do the work, the record keeps the moat: Agents make at least 15% of day-to-day work decisions. (watching)2028 · Robots leave the cage: More than 700,000 industrial robots are installed in a year. (watching)2028 · Robots leave the cage: Home humanoids ship more than 100,000 units in a year. (watching)2028 · The rails stay, the money changes: Dollar stablecoins in circulation pass $2 trillion. (watching)2028 · Supply chains become strategy: MP Materials’ second US magnet plant starts commissioning. (watching)2028 · Supply chains become strategy: Starlink direct-to-cell runs on next-generation satellites at broadband speeds. (watching)2030 · AI compute moves from training to running: Inference is more than half of AI data-center compute. (on track)2030 · Power becomes the bottleneck, not chips: Ten large reactors are under construction in the US. (watching)2030 · Power becomes the bottleneck, not chips: Data centers use at least 9.5% of US electricity, the low end of the DOE lab range. (on track)2031 · AI compute moves from training to running: No foundry other than TSMC ships a top-five AI accelerator in volume at the leading node. (on track)2031 · Medicine treats the chronic, earlier: An oral GLP-1 is approved for obesity in the US. (hit)2031 · The rails stay, the money changes: Visa and Mastercard settle a meaningful share of their own volume in stablecoins. (on track)2026 · Robots leave the cage: Waymo reaches 1 million paid rides a week. (watching)2035 · Medicine treats the chronic, earlier: Nearly 30% of eligible US patients are on a GLP-1. (watching)

AI compute moves from training to running

Question: Will AI spending shift from training models to running them?

Spending shifts from building models to serving them, custom chips take share inside the hyperscalers, and the toll on every chip is paid at the foundry and the lithography tool.

Where we are today
  • Microsoft, Alphabet, Amazon and Meta guided to roughly $700B of capital spending for 2026 between them, most of it for AI data centers. Om Malik, Apr 2026
  • McKinsey expects inference to be more than half of AI compute by 2030, growing about 35% a year against about 22% for training. McKinsey, Feb 2026
  • TrendForce projects the US at about 22% of the world’s advanced chip capacity by 2030, with Taiwan falling from 71% in 2021 to about 58%. TrendForce, Mar 2025
  1. 2031
    5 years

    Most AI compute is spent on running models, not training them. A growing share of new inference racks runs on the hyperscalers’ own chips (TPU, Trainium and Broadcom or Marvell designs), so NVIDIA stops being the default for every rack while still leading at the frontier.

  2. 2036
    10 years

    Leading-edge logic is still made by one foundry using one lithography supplier, now in Arizona and Japan as well as Taiwan. Whoever designs the chip, TSMC and ASML are paid on it, and EDA software is paid on every tape-out.

  3. 2046
    20 years

    Compute is bought like electricity: a metered utility priced by the token. The moats sit where the physics is hardest to copy, not where the software is.

We are wrong if

A second foundry wins volume production of a top-five AI accelerator at the leading node, or hyperscaler chips stall and new inference builds land on NVIDIA by default again.

2026 capital spending guidance, $B
Amazon: $200B$200BAmazonMicrosoft: $190B$190BMicrosoftAlphabet: $185B (range $180B to $190B)$185BAlphabetMeta: $135B (range $125B to $145B)$135BMeta
ReportedOm Malik
Checkpoints
  • 2030

    Inference is more than half of AI data-center compute.

    On track

    McKinsey’s February 2026 model has inference growing faster than training and passing half by 2030.

  • 2031

    No foundry other than TSMC ships a top-five AI accelerator in volume at the leading node.

    On track

    Outside China, every merchant and custom AI accelerator shipping in volume today is made at TSMC.

Next · Theme 02Power becomes the bottleneck, not chips

Power becomes the bottleneck, not chips

Question: Will electricity limit the growth of AI data centers?

A data center can buy GPUs in months but waits years for a grid connection. Firm, carbon-free power under long contracts is the scarcest input in the AI build.

Where we are today
  • The IEA expects global data-center electricity use to more than double by 2030 to about 945 TWh, with data centers driving almost half of US demand growth. IEA, Apr 2025
  • A DOE lab report puts US data centers at about 4.5% of US electricity today and 9.5% to 15% by 2030. E&E News on the DOE lab data-center report, Jun 2026
  • The May 2025 executive orders target 400 GW of US nuclear by 2050, up from about 100 GW, and 10 large reactors under construction by 2030. World Nuclear News, May 2025
  • In October 2025 the US government signed an $80B deal with Westinghouse, owned by Cameco and Brookfield, to build AP1000 reactors. American Nuclear Society, Oct 2025
  1. 2031
    5 years

    Hyperscalers buy power the way they buy chips: 15 to 20 year contracts for firm supply. The existing nuclear fleet runs at higher output for longer, data centers use about a tenth of US electricity, and gas turbines and grid equipment stay sold out years ahead.

  2. 2036
    10 years

    The first AP1000s from the Westinghouse programme and the first small modular reactors are being built in the US. Utilities lock uranium and fuel services into long-term contracts rather than buying spot.

  3. 2046
    20 years

    Nuclear is a growing share of US generation again, and the companies that own the licences, the fuel chain and the service contracts look more like toll roads than commodity producers.

We are wrong if

Efficiency gains outrun usage and data-center load forecasts are cut, or no new US reactor reaches construction before 2031.

Share of US electricity used by data centers
Today: 4.5%4.5%Today2030: 11.8% (range 9.5% to 15%), forecast11.8%2030
Checkpoints
  • 2030

    Ten large reactors are under construction in the US.

    Watching

    The Westinghouse deal plans construction from 2030, so the target lands at the edge of the window.

  • 2030

    Data centers use at least 9.5% of US electricity, the low end of the DOE lab range.

    On track

    About 4.5% today. Hyperscaler capex and signed power contracts point to the range.

Next · Theme 03Agents do the work, the record keeps the moat

Agents do the work, the record keeps the moat

Question: How will AI agents change enterprise software?

AI agents take on a large share of routine office work. Software priced per seat gets repriced, and the moat moves to whoever holds the record the agent has to read and write.

Where we are today
  • Gartner expects agentic AI to make at least 15% of day-to-day work decisions by 2028, up from almost none in 2024. Gartner via WFTV, Sep 2026
  • Gartner also expects a third of enterprise software to include agentic AI by 2028, and more than 40% of agentic AI projects to be cancelled by the end of 2027 on cost and unclear value. Gartner via SD Times, Jun 2025
  1. 2031
    5 years

    Per-seat pricing gives way to usage or outcome pricing across enterprise software. Tools that were only an interface lose ground to cheaper AI-native rivals, while systems of record, identity and payments hold.

  2. 2036
    10 years

    An agent buying, booking or filing on your behalf is normal. The winners are the platforms an agent has to call to get anything done: the ledger, the marketplace, the checkout, the court record.

  3. 2046
    20 years

    Most software is written and run by machines. The durable businesses own data, permissions and trust that cannot be regenerated by a model.

We are wrong if

Seat counts at the large suites keep rising and agent revenue stays a rounding error through 2031.

Day-to-day work decisions made by AI agents
2024: 0%0%20242028: 15%, forecast15%2028
Reported ForecastGartner via WFTV
Checkpoints
  • 2027

    The shake-out comes first: a large share of early agent projects is cancelled.

    Watching

    Gartner expects more than 40% to go. A shake-out does not break the theme; the survivors set the pricing.

  • 2028

    Agents make at least 15% of day-to-day work decisions.

    Watching

    Gartner’s marker. Too early to measure.

Next · Theme 04Robots leave the cage

Robots leave the cage

Question: When will robotaxis and humanoid robots go mainstream?

Self-driving cars arrive city by city and factory automation keeps compounding. Humanoids start in plants and warehouses, then move into homes faster than the consensus expects: by the mid-2030s a home robot is a household purchase, like a car.

Where we are today
  1. 2031
    5 years

    Robotaxis run commercially in dozens of US cities. Fleet operators with their own apps take the densest routes, so ride-hailing networks have to earn the rest. In factories, the money is still in CNC, servos and arms. Home humanoids ship in the hundreds of thousands, rented by the month and still helped by remote operators for chores they have not learned.

  2. 2036
    10 years

    Driverless rides are an ordinary choice in most large US metros and on trucking corridors. Humanoids ship in the millions, and homes become a market as big as the factory: priced like a car and financed like one, a home robot does laundry, tidying and elder care in millions of households in rich countries.

  3. 2046
    20 years

    A humanoid is a normal appliance in many homes and on most factory floors. Labor scarcity and aging populations drive demand. Owning the actuators, gears, sensors and chips every robot needs beats picking the robot brand.

We are wrong if

Robotaxi launches stall at a handful of cities by 2031, or home humanoids still sell fewer than 100,000 a year by 2030 because they need a remote operator for most chores.

Industrial robots installed a year, thousands
2024: 542k542k20242025: 575k, forecast575k20252028: 700k, forecast700k2028
Reported ForecastIFR
Checkpoints
  • 2026

    Waymo reaches 1 million paid rides a week.

    Watching

    Doubling from the March run rate in nine months; the year-end figure settles it.

  • 2028

    More than 700,000 industrial robots are installed in a year.

    Watching

    IFR’s forecast. The 2025 figure, due next autumn, is the first read.

  • 2028

    Home humanoids ship more than 100,000 units in a year.

    Watching

    Our bullish home call needs this. 1X alone targets that capacity by the end of 2027; Tesla and Figure are aiming at homes too.

Next · Theme 05The rails stay, the money changes

The rails stay, the money changes

Question: Will stablecoins replace Visa and Mastercard?

Card networks, exchanges and rating agencies keep their seats, because rules and habits move slower than technology. New forms of money settle on top of them rather than around them.

Where we are today
  • Treasury Secretary Bessent said dollar stablecoins could pass $2 trillion by 2028. They were about $247B when he said it in June 2025. The Block, Jun 2025
  • Visa’s stablecoin settlement passed a $20B annual run rate in September 2026, up more than 15 times in a year, across more than 160 card programmes. The Block, Sep 2026
  1. 2031
    5 years

    Visa and Mastercard still carry most consumer card spending. Stablecoins are a multi-trillion dollar settlement layer that the networks and exchanges run themselves, and more institutions hold Bitcoin through regulated products.

  2. 2036
    10 years

    The ratings duopoly, index benchmarks and energy futures still sit inside regulation and contracts. Bitcoin sits beside gold as a reserve asset that some treasuries hold, not as everyday money.

  3. 2046
    20 years

    Most money is digital and programmable. The firms that set the rules, the benchmarks and the trust layer keep earning on all of it.

We are wrong if

Account-to-account payments take a meaningful share of US card volume, or a regulator opens credit ratings or index licensing to new entrants at scale.

Dollar stablecoins in circulation, $B
Jun 2025: $247B$247BJun 20252028: $2,000B, forecast$2,000B2028
Reported ForecastThe Block
Checkpoints
  • 2028

    Dollar stablecoins in circulation pass $2 trillion.

    Watching

    Roughly an eightfold rise from mid-2025. The networks adopting them is the part already happening.

  • 2031

    Visa and Mastercard settle a meaningful share of their own volume in stablecoins.

    On track

    Visa is already doing it at a $20B run rate; small next to its total, but growing fast.

Next · Theme 06Supply chains become strategy

Supply chains become strategy

Question: Why are rare earths, chips, copper and launch becoming strategic?

Governments now treat chips, magnets, launch and power as security. The suppliers that are written into those plans earn longer contracts and price floors.

Where we are today
  • The Department of Defense set a 10-year $110/kg price floor for MP Materials’ NdPr, backs a second US magnet plant due to start commissioning in 2028, and guarantees buyers for all of its output. MP Materials, Jul 2025
  • SpaceX agreed to pay $17B for EchoStar spectrum so Starlink can reach ordinary phones at up to 150 Mbps, with testing from late 2027 on satellites only Starship can launch. SatNews, Feb 2026
  • S&P Global scenarios put the copper shortfall at 1.5 to 9.9 million tonnes a year by 2035 (2.1 million by 2030 in the slow case), against a largest-ever deficit just under 1 million. Skillings on the S&P Global copper outlook, Apr 2026
  1. 2031
    5 years

    A Western rare-earth magnet chain runs at scale outside China, with US-made magnets inside car and phone motors. Leading-edge chips are made on US and Japanese soil as well as in Taiwan.

  2. 2036
    10 years

    Launch is routine infrastructure: satellite internet reaches phones directly, and one or two providers fly most of the mass to orbit. Copper is the metal the grid and the data center both run short of.

  3. 2046
    20 years

    Industrial policy is permanent. Owners of qualified, licensed capacity in contested supply chains are paid like utilities with growth.

We are wrong if

A lasting US-China thaw lifts export controls and Chinese supply undercuts Western magnet and processing prices, or launch cadence stalls.

Copper shortfall, million tonnes a year
2030: 2.1Mt, forecast2.1Mt20302035: 9.9Mt (range 1.5Mt to 9.9Mt), forecast9.9Mt2035
Checkpoints
  • 2028

    MP Materials’ second US magnet plant starts commissioning.

    Watching

    The DoD agreement sets the date; financing was committed at signing.

  • 2028

    Starlink direct-to-cell runs on next-generation satellites at broadband speeds.

    Watching

    Hinges on Starship cadence and the EchoStar transfer closing.

Next · Theme 07Medicine treats the chronic, earlier

Medicine treats the chronic, earlier

Question: How will GLP-1 obesity drugs change healthcare?

Obesity drugs, robotic surgery and gene editing shift spending from treating late disease to managing it early.

Where we are today
  • The FDA approved Lilly’s orforglipron (Foundayo), a once-daily GLP-1 pill for obesity, on April 1, 2026. Healio, Apr 2026
  • Morgan Stanley expects the GLP-1 market to reach about $190B by 2035 from $79B in 2025, with the treated share of eligible US patients rising from about 6% to nearly 30%. Morgan Stanley, Apr 2026
  1. 2031
    5 years

    GLP-1 drugs come as pills as well as injections, coverage widens, and price competition starts as semaglutide patents lapse in major markets. Robotic surgery faces real rivals in its core procedures.

  2. 2036
    10 years

    Obesity treatment is a mass-market category with several makers and nearly a third of eligible US patients on it. The leaders keep their lead through manufacturing scale and newer molecules, not through the original patents.

  3. 2046
    20 years

    Preventive and genetic medicine is a normal part of care. The moats are in data, manufacturing and the surgeon or physician habit, not a single drug.

We are wrong if

Long-term outcome data on GLP-1s disappoints and payers pull back coverage, or a rival robotic system wins a large share of new installs in general surgery by 2031.

GLP-1 drug sales, $B a year
2025: $79B$79B20252035: $190B (range $170B to $240B), forecast$190B2035
Reported ForecastMorgan Stanley
Checkpoints
  • 2031

    An oral GLP-1 is approved for obesity in the US.

    Hit

    Orforglipron was approved on April 1, 2026.

  • 2035

    Nearly 30% of eligible US patients are on a GLP-1.

    Watching

    About 6% in 2025, per Morgan Stanley.

NextMoats on the move

The themes are the big picture. Underneath, every stock file says which parts of its moat we expect to widen or narrow over the next 5 to 10 years, and what would show it. This list is built from those files, so it changes as soon as an analysis does. Of the 25 IM25 holdings, 18 hold, 3 widen and 4 narrow.

Moat widening over 5–10 yearsWidening7

  • ASMLNetwork effects

    Signal to watch: High-NA EUV moves into production at Intel 14A and at TSMC's A14P or Samsung's equivalent node, not only pilot lines; the trend stops if TSMC carries High-NA past the A14 family on low-NA multipatterning.

  • CCJTransaction embedding

    Signal to watch: AP1000 units financed under the conditional DOE long-lead commitment (up to 10 reactors) reach construction with Westinghouse as fuel and services supplier; the trend stops if that financing lapses without a single US AP1000 construction start.

  • MPTransaction embedding

    Signal to watch: Independence ships GM's roughly 1,000 t a year from Q4 2026 and Apple's recycled-magnet supply starts in 2027, putting MP magnets into qualified motor designs; the trend stops if GM or Apple qualifies a second magnet source for the same programmes.

  • ANTTransaction embedding

    Signal to watch: Anthropic is trading discounted token rates for multi-year minimum-spend contracts. The trend is confirmed if remaining performance obligations grow faster than revenue for four quarters after the $10.8B of June 2026; it stops if RPO stays below one quarter of revenue.

  • DESystem of record

    Signal to watch: Operations Center keeps adding monthly active users and engaged acres beyond the >450K MAU and >520M acres on file, and AGCO and CNH precision hardware keeps feeding it as the default record; the trend stops if MAU stalls or growers move records to FieldView or CNH FieldOps.

  • NXETransaction embedding

    Signal to watch: NexGen keeps adding long-term utility sales agreements as Rook I construction advances, taking the 10M lbs committed (2M lbs a year over the first five years) toward a majority of planned early output; the trend stops if the contract book stays near 10M lbs into first production.

  • PLTRSystem of record

    Signal to watch: More defence programmes of record move onto Maven after the first one, and the Army's NGC2 data layer on Foundry spreads under the enterprise agreement worth up to $10B over 10 years; the trend stops if Maven stays at a single programme of record and NGC2 does not extend past the Army.

Moat narrowing over 5–10 yearsNarrowing24

  • AVGOSystem of record

    Signal to watch: VMware workloads keep moving off vCenter to Nutanix, OpenShift, Hyper-V and the public clouds, as Gartner forecasts for more than a third of workloads by 2028; the trend stops if VMware ARR keeps growing at double digits (17% YoY now) without customer-count attrition.

  • NVDATransaction embedding

    Signal to watch: NVIDIA's share of AI accelerator spend (roughly 80% or more today) keeps falling as Google TPU, Amazon Trainium and Broadcom-built custom XPUs take inference volume; the trend stops if that share holds near 80% through the Rubin generation.

  • SPCXLearned interfaces

    Signal to watch: Cursor's share of AI-coding spend keeps sliding against Claude Code and GitHub Copilot; the trend stops if Cursor's paid seats grow faster than the category for a full year.

  • UBERNetwork effects

    Signal to watch: Waymo opens its own app in Austin and Atlanta from January 2028 and Tesla keeps scaling its own Texas ride-hailing network, so Uber's densest robotaxi routes are served off-platform; the trend stops if those operators keep supplying through Uber and autonomous trips on Uber keep growing from about 300K a week across 30+ partners.

  • 000660.KSScale economics

    Signal to watch: SK hynix's HBM revenue share (about 56% in Q1 2026, IDC) keeps falling toward Samsung and Micron as both ramp HBM4, as TrendForce's call for the bit lead to narrow from 59% to roughly 50% in 2026 implies; the trend stops if it holds above half of HBM revenue through the HBM4E generation at NVIDIA.

  • 6268.TScale economics

    Signal to watch: Chinese RV makers led by Zhejiang Shuanghuan Driveline and Ningbo Zhongda Leader keep taking RV sockets on robot lines built in China and start qualifying on FANUC, Yaskawa, ABB or KUKA arms outside China; the trend stops if Nabtesco holds RV price and precision-gear order growth through the robot recovery.

  • ADBEBundling

    Signal to watch: Creative Cloud price increases stay deferred while Canva and Figma keep adding paid seats inside Creative Cloud customers; the trend stops when Adobe lands a price increase without a rise in churn.

  • BERegulatory lock in

    Signal to watch: EPA turns its 2025-26 data-centre relief (the January 2026 turbine NSPS subcategory for temporary units and the proposal to let construction start before air permits) into final rules and gas turbine lead times fall from about three years; the trend stops if those proposals are withdrawn or struck down and Bloom keeps winning sites on permitting speed.

  • BIDUNetwork effects

    Signal to watch: Online Marketing revenue keeps falling year on year (−19% in Q2 2026); the trend stops if it returns to growth, or if Apollo's ride network grows large enough to rate on its own.

  • COINTransaction embedding

    Signal to watch: ETF sponsors keep adding second custodians beside Coinbase (BlackRock added Anchorage to IBIT in April 2025; Morgan Stanley's proposed Bitcoin Trust names BNY with Coinbase; Grayscale's amended Hyperliquid filing replaced Coinbase with Anchorage); the trend completes if a top-three US spot BTC or ETH ETF moves primary custody off Coinbase, and stops if Coinbase still holds primary custody of most US spot crypto ETF assets in 2028.

  • FICONetwork effects

    Signal to watch: VantageScore keeps gaining GSE originations from 5.5% in September 2026 (KBW) as originators follow Rocket; the trend completes if a second top-10 originator defaults to VantageScore and its GSE share passes 20%, and stops if that share stays under 10% through the end of TransUnion's $0.99 pricing in 2028.

  • FIGLearned interfaces

    Signal to watch: Claude Design and Google's prompt-to-prototype tools keep taking zero-to-one work from the non-designers who were two-thirds of Figma's roughly 13 million monthly users at IPO; the trend stops if that non-designer share holds and net dollar retention stays above 130% (136% in Q2 2026).

  • HSYDFScale economics

    Signal to watch: Chinese strain-wave makers led by Leaderdrive keep adding capacity and win reducer supply for volume humanoid and industrial-robot programmes, pushing reducer pricing down; the trend stops if Harmonic Drive's reducer pricing and share hold at those programmes as humanoid volumes ramp.

  • ISRGRegulatory lock in

    Signal to watch: Medtronic Hugo adds US general-surgery and gynecology clearances to its December 2025 urology clearance (510(k)s filed June 2026) and J&J Ottava expands from its July 2026 De Novo for upper-abdominal general surgery into hernia; the trend stops if those filings are not cleared by end-2027 or both stay in limited launch.

  • LEURegulatory lock in

    Signal to watch: Centrus stays the only NRC-licensed Western HALEU enricher; the trend completes if General Matter's Paducah plant (DOE $900M task order January 2026, NRC Part 70 application filed September 2026 with a HALEU amendment planned for 2027) or Urenco USA gains NRC authorisation above 10% enrichment and delivers HALEU, and it stops if no second producer holds a HALEU licence by 2031.

  • LULUBrand

    Signal to watch: Americas comparable sales keep declining (-12% in Q2 FY2026) through FY2027 while Alo and Vuori keep taking US premium athleisure share; the trend stops when Americas comps turn positive without deeper markdowns.

  • MDBBusiness logic

    Signal to watch: MongoDB-compatible engines built on PostgreSQL, led by the Linux Foundation's DocumentDB project backed by AWS, Google and Microsoft, keep adding managed services and production MongoDB workloads; the trend stops if Atlas growth holds in the high 20s or better and no hyperscaler ships a migration path that runs MongoDB applications unchanged at scale.

  • NOWBusiness logic

    Signal to watch: Salesforce Agentforce IT Service keeps adding accounts beyond the 180+ it had in February 2026 (against ServiceNow's 8,600 ITSM customers) and starts winning named ITSM replacements at large ServiceNow customers; the trend stops if ServiceNow's 98% renewal rate holds and Agentforce IT Service stays under a few percent of that base.

  • NVORegulatory lock in

    Signal to watch: Semaglutide's US compound protection runs out inside the decade (generic Ozempic expected around 2032 under Novo's confidential 2024 settlements) while no successor molecule is approved; the trend stops if amycretin or another Novo molecule with its own patent life carries most of Novo's US GLP-1 prescriptions before generic entry.

  • RBLXLearned interfaces

    Signal to watch: Build, Roblox's prompt-to-game tab in the mobile app (limited alpha in New Zealand from 28 July 2026), rolls out to more countries and a growing share of newly published experiences start in Build rather than Studio; the trend stops if Build stays a prototyping tool and the top experiences keep being built by Studio and Luau creators.

  • RDDTNetwork effects

    Signal to watch: US DAUq turned negative in Q3 2026 (−2%) as Google search referrals fell, and Reddit warns of a low-double-digit DAUq decline in Q4. The narrowing is confirmed if US logged-in DAUq falls below 22.5M in a reported quarter; it stops if the logged-in-only test lifts US logged-in DAUq above 24M.

  • REGNRegulatory lock in

    Signal to watch: Dupixent's US compound patent expires in March 2031 (Sanofi investor materials) and biosimilar dupilumab programmes file for US approval ahead of it; the trend stops if Sanofi's longer-interval Q4 formulation or follow-on patents convert most US Dupixent patients before 2031, as Eylea HD is doing for Eylea.

  • SNAPNetwork effects

    Signal to watch: North American daily users keep falling year on year (down 7% to 92M in Q2 2026) as the 13 to 34 cohort spends its time on Instagram and TikTok; the trend stops if North American DAU returns to flat or growth for two quarters running.

  • XAGIndustrial utility

    Signal to watch: Silver Institute photovoltaic silver demand keeps falling year on year (forecast down 19% to about 151 Moz in 2026, after 197.5 to 186.6 Moz in 2025) as cell makers thrift loadings and move to high-copper and copper-plated pastes while cell output grows; the trend stops if PV silver demand turns back up for two consecutive years.

  1. Om Malik, hyperscaler 2026 capex guidance · April 30, 2026
  2. McKinsey, The future of AI workloads · February 24, 2026
  3. TrendForce, advanced process capacity by region · March 5, 2025
  4. IEA, Energy and AI · April 10, 2025
  5. E&E News on the DOE lab data-center report · June 22, 2026
  6. World Nuclear News, US nuclear executive orders · May 24, 2025
  7. American Nuclear Society, Westinghouse $80B contract · October 28, 2025
  8. Gartner via WFTV, agentic AI and work decisions · September 30, 2026
  9. Gartner via SD Times, agentic AI project cancellations · June 30, 2025
  10. Noqta, Waymo weekly rides · March 30, 2026
  11. IFR, World Robotics 2025 · September 25, 2025
  12. Humanoid Guide on Goldman Sachs humanoid forecast · September 15, 2026
  13. Humanoid Guide on Morgan Stanley humanoid forecast · April 29, 2025
  14. The Robot Report on the 1X NEO launch · October 30, 2025
  15. The AI Insider on 1X NEO production · May 1, 2026
  16. The Block, Bessent on stablecoins · June 11, 2025
  17. The Block, Visa stablecoin settlement · September 8, 2026
  18. MP Materials, Department of Defense partnership · July 10, 2025
  19. SatNews, Starlink direct-to-cell plans · February 25, 2026
  20. Skillings on the S&P Global copper outlook · April 29, 2026
  21. Healio, FDA approves orforglipron · April 1, 2026
  22. Morgan Stanley, obesity drugs are scaling fast · April 22, 2026

These are InvestMoat's views, written to be checked, not financial advice. Calls are dated to a year and reviewed when the coverage they rest on changes. The IM25 is chosen by its published rules, not by this page, though moat trajectory is part of the moat score those rules use.

Start from the coverage universe or the research notes to see the underwriting behind each view.