The US Energy Information Administration’s national average price for on-highway diesel reached $6.285 a gallon for the week of September 14, 2026, a record in the weekly series; the last print before the Iran war began on February 28, 2026 was $3.809 (week of February 23). California diesel averaged $8.039. Over the same period the Henry Hub natural gas spot price was $2.79 per million Btu for the week of September 11, 2026, and has sat between $2.62 and $2.91 since late July. On an energy basis (diesel at roughly 138,000 Btu per gallon) diesel is costing about $45 per million Btu, sixteen times the wholesale gas price; that ratio is arithmetic from the two EIA series, not a published figure.

On September 15, 2026, the nonprofit Energy Vision published an update to its 2025 report on replacing older diesel trucks, re-run for the new fuel prices. As of early September it put compressed natural gas about $2 per diesel-gallon-equivalent below diesel nationally and about $3.50 below in California. It estimates a new Class 8 tractor with a Cummins X15N natural-gas engine costs about $75,000 more than a diesel equivalent, and that at 100,000 miles a year and 6 miles per gallon-equivalent the premium is recovered in 1.3 years at a $3.50 advantage or 2.3 years at $2; at 80,000 miles, 1.6 to 2.8 years. The estimates are fuel-savings only and do not include financing, maintenance, fueling infrastructure or residual value.

The installed base is tiny. Energy Vision counted slightly more than 1,000 trucks with the X15N in operation as of July 2026, against roughly 5.2 million Class 7 and 8 trucks in the United States, and said manufacturers could not supply the 65,000 tractors in its replacement scenario; it cites a Cummins projection of 20,000 to 25,000 X15N trucks a year by around 2030. There were about 1,400 CNG fueling stations nationally as of August 2026, 767 offering renewable natural gas, up from 527 in January 2025; renewable gas was 94% of natural-gas vehicle fuel in 2025. For comparison, 2,895 heavy-duty battery-electric trucks had been deployed as of December 2025.

Prices verified September 21, 2026. Clean Energy Fuels, the largest natural-gas truck-fueling network, closed at $1.72, near its one-year low of $1.57 (August 19, 2026) and 44% below its one-year high of $3.06 (October 24, 2025). Cummins closed at $534, 27% below its June 25, 2026 high of $728. Westport Fuel Systems was $1.92, OPAL Fuels $1.97 (one-year low $1.78 on June 22), and Hexagon Composites, which makes the CNG tank systems, NOK 12.80 against an August 14 high of NOK 14.50. Renewable-gas producer Montauk Renewables, by contrast, was $2.35 against a one-year high of $2.64 set September 11. Valero ($404) and Marathon Petroleum ($414) were within 3% of their highs. Brent futures fell toward $95 on September 21 after Saudi Arabia moved to restore its East-West pipeline and President Trump said he might meet Iran’s president at the United Nations this week.

Opportunity

The obvious readings of a diesel spike are the refiners, which have repriced, and the trucking carriers with surcharge protection, which have been argued elsewhere. The reading that has not moved anything is substitution: the relative price of the two fuels a heavy truck can burn has changed by more than the industry has ever seen, and the switching product — a 15-litre natural-gas engine with diesel-like performance, on sale only since 2024 — exists but is built in the hundreds.

The mechanism that makes the gap durable is that the two fuels are priced in different markets. Diesel is a globally traded product and has followed the Hormuz disruption; US natural gas is trapped behind fixed LNG export capacity, which is why Henry Hub has not risen even with a fifth of global LNG supply disrupted and Qatar’s exports curtailed. As long as the disruption persists, the spread is structural rather than a spot anomaly. A fleet that buys a truck for five to seven years and gets its money back on the premium in two is not making a bet on the spread lasting; it is making a bet that it does not fully close.

Hypothesis: the market is treating the diesel–gas gap as a war artifact that will not survive a settlement, and has therefore priced the renewable-gas molecule (Montauk near its high, on credit values) but not the vehicle and fueling channel that converts the gap into volume. If natural-gas truck orders inflect over the next two quarters, the beneficiaries are a small, thinly covered set — the fueling network, the tank maker, the engine maker’s share of Class 8 — that are priced today for the pre-2026 world in which the fuel saving was roughly $1 a gallon-equivalent. This is an inference about positioning; the sourced fact is the spread and the payback math.

A second inference: the binding constraint is not the fleets’ willingness but the supply of engines and tanks, which means the first place any switch shows up is in order backlogs and supplier capacity announcements, not in fuel volumes.

How it could play out

The diesel-to-gas gap holds through the autumn while the Hormuz disruption continues, even with crude off its highs, because the distillate crack rather than the crude price carries most of the premium. Fleets that could not pass the cost through — the owner-operators and small carriers that make up most US truckload capacity — either exit or, at the next replacement decision, specify the cheaper fuel; larger fleets running fixed lanes near existing stations add natural-gas tractors to their 2027 orders. Cummins’ X15N build rate, Hexagon’s tank capacity and the station count become the constraint, and each announces expansion; Clean Energy Fuels’ and OPAL’s fuel volumes step up on a base that has been flat for years. The possible investment implication is a re-rating of the small fueling and component names from an ex-growth, credit-dependent profile to a volume-growth profile, while Cummins’ Class 8 franchise gets a mix benefit. If a settlement collapses the spread back toward $1 a gallon-equivalent before orders inflect, none of this happens and the fueling names stay where they are.

Questions worth asking

• Have Class 8 natural-gas truck orders actually inflected since diesel crossed $5 in March 2026, or has seven months of record spreads produced nothing? ACT Research and FTR publish monthly Class 8 orders; Cummins reports X15N in its quarterly commentary. If orders have not moved, the thesis is wrong regardless of the fuel math.
• What is the diesel–gas gap at $70 Brent? The pre-war diesel price was $3.81; if the gap reverts to roughly $1 a gallon-equivalent, does a 2.3-year payback become 5 years, and does that still clear a fleet’s hurdle?
• Where is the bottleneck — engine (Cummins), tanks (Hexagon Agility), truck OEM slots (Freightliner, Kenworth, Peterbilt, Volvo) or stations? Whoever holds it captures the pricing.
• How much of Clean Energy Fuels’ and OPAL’s margin is the fuel spread versus the D3 RIN and California LCFS credits? If credits dominate, a wider spread helps volume but not necessarily margin; if the spread flows through, the operating leverage is larger than the share price implies.
• Is there a used-truck angle: natural-gas tractors have historically had poor residual values, which is the hidden cost in the payback math — and a wider spread should lift residuals, which would show up first at the auctions.
• Who loses if fleets switch: renewable-diesel producers, whose product is priced off diesel and has followed it up, and diesel-engine aftertreatment suppliers?

Where to look

• Clean Energy Fuels (CLNE) — the largest CNG/RNG fueling network for heavy trucks, at a one-year low while the fuel it sells has never been cheaper relative to diesel
• Cummins (CMI) — sole maker of the 15-litre X15N natural-gas engine, down 27% from its June high on the diesel truck cycle
• Hexagon Composites (HEX.OL) — Hexagon Agility supplies the composite CNG tank systems on natural-gas trucks; the most direct component exposure
• OPAL Fuels (OPAL) — renewable-gas producer that also owns and builds truck fueling stations, so it has both the molecule and the channel
• Westport Fuel Systems (WPRT) — high-pressure direct-injection natural-gas systems for heavy trucks through its Volvo joint venture
• Montauk Renewables (MNTK) — the renewable-gas producer that has already repriced, useful as the reference for what the market has and has not credited

Thesis check

The chain is strong at the front: the fuel-price gap is primary-sourced from EIA data, the payback arithmetic is published and reproducible, the installed base of about 1,000 trucks means any real switch is growth from near zero, and the fueling and component companies are priced at or near lows rather than for the opportunity. The weak link is that the gap is a product of the war and the market visibly doubts its persistence — on September 21, 2026 Brent fell toward $95 as Saudi Arabia moved to restore its East-West pipeline and Washington floated talks with Iran — and the deciding evidence, an inflection in natural-gas truck orders, does not yet exist in public data; Cummins can build only a small number of X15N engines this year, so even a genuine switch would take two years to appear in Clean Energy Fuels’ volumes.

Timing

September 2026 Class 8 truck-order data due in early October 2026; EIA prints retail diesel every Monday

Sources

EIA Gasoline and Diesel Fuel Update, Sep 14 2026 · EIA Henry Hub spot price (weekly), Sep 11 2026 · Heavy Duty Trucking, Sep 15 2026 · Energy Vision report update, Sep 2026 · CNBC, Sep 17 2026 · Trucking Dive, Sep 2026 · CNBC, Sep 21 2026

Nscale Ltd, the London-based AI cloud and data-centre developer founded in May 2024, filed a public Form S-1 with the SEC on September 18, 2026 for an initial public offering on the New York Stock Exchange, with Goldman Sachs, J.P. Morgan and Morgan Stanley leading. Bloomberg reported in August that the offering could raise as much as US$3 billion; Fortune, citing the Financial Times, reported on September 21 that the company is seeking a US$35 billion valuation. The Series C round in March 2026, US$2 billion led by Aker ASA and 8090 Industries with Nvidia, Dell, Lenovo, Nokia, Citadel, Jane Street and Point72 participating, valued the company at US$14.6 billion.

The prospectus shows revenue of US$140.6 million for the six months to June 30, 2026, up from US$10.4 million a year earlier, against a net loss of US$1.02 billion; cost of revenue of US$189.6 million exceeded revenue. Remaining performance obligations were US$56.4 billion as of August 31, 2026, within a contracted order book the company puts at roughly US$103 billion, anchored by compute agreements with Anthropic (reported at US$45 billion), OpenAI and Microsoft. Commitments to purchase technology equipment not yet delivered stood at US$24 billion at June 30, with a further US$3.5 billion of data-centre build commitments. The filing discloses that management initially identified substantial doubt about the company’s ability to continue as a going concern because its funding plan relied on uncommitted debt and equity, concluded that deferrable capital expenditure alleviated it, and that on September 15, 2026 Nvidia provided US$3.1 billion through unsecured convertible loan notes. A separate pre-IPO financing has been reported to include up to US$1.5 billion of convertible notes led by Third Point with a conversion cap at a US$30 billion valuation.

Aker ASA’s second-quarter report, published July 16, 2026, states that following the Series C and Nscale’s acquisition of American Intelligence & Power Corporation, Aker’s ownership in Nscale was 22.76% on a fully diluted basis, making Nscale Aker’s second-largest asset at NOK 32 billion, or 26% of gross asset value of NOK 121.3 billion. Aker’s net asset value was NOK 106.1 billion, NOK 1,429 per share, against a share price of NOK 1,148 — a 20% discount. Aker had also agreed to sell Cognite to Schneider Electric at a US$3.1 billion enterprise value for about NOK 14.7 billion in cash, taking its liquidity reserve above NOK 20 billion; Aker’s CEO Øyvind Eriksen sits on Nscale’s board. In the quarter Nscale signed two Microsoft contracts in Portugal and Norway for more than 96,000 Nvidia Rubin GPUs combined and raised US$790 million for its Narvik, Norway build.

Prices verified September 21, 2026: Aker ASA closed at NOK 1,554, up 35% from NOK 1,148 on June 30, 2026, against a six-month high of NOK 1,624 and a six-month low of NOK 1,026. The US dollar traded at about NOK 9.44. On those figures Aker has roughly 74 million shares, so the NOK 32 billion Nscale mark is about NOK 430 per Aker share; a 20% Aker stake at a US$35 billion valuation would be worth about US$7 billion, or NOK 66 billion — roughly NOK 890 per share. Those last two numbers are arithmetic from the published figures, not company disclosures, and assume dilution from the pre-IPO notes and the offering takes Aker’s stake to about a fifth.

Opportunity

The obvious reading is that this is a hot AI-infrastructure IPO with a large, well-covered anchor shareholder, that Aker’s 35% rise since June already reflects it, and that anyone who wants Nscale can simply buy the IPO. The first two are true. Holding the rest of Aker’s portfolio at its June 30 values, an IPO at US$35 billion lifts Aker’s net asset value to roughly NOK 140 billion, about NOK 1,885 per share, which puts today’s NOK 1,554 at an 18% discount — in other words the share price already embeds the reported IPO valuation with Aker’s customary discount left in place.

That is where the question becomes interesting rather than closed. Aker’s discount has historically been applied to a portfolio dominated by an oil producer (Aker BP) and unlisted industrial holdings; after the IPO, close to half of Aker’s gross assets would be a daily-marked, liquid, US-listed AI infrastructure stock, and Aker would be sitting on more than NOK 20 billion of cash from the Cognite sale. Holding companies whose largest asset becomes a liquid listed stake behave in two ways: the discount persists as a permanent tax on indirect ownership, or the parent realises value — sell-down, distribution in specie, or a special dividend — and the discount closes on the announcement. Aker’s own record over the past year is realisation: the Cognite sale, the Aker BioMarine take-private offer, and Akastor’s IPO and sell-down of HMH.

Hypothesis: the market has done the first-order arithmetic on Aker but not the second-order question of what Aker does with a stake that becomes worth more than its oil business, and the Oslo listing means the investor base that can buy Nscale directly is not the one that is pricing Aker. The asymmetry runs both ways: the Third Point conversion cap at US$30 billion suggests sophisticated pre-IPO money negotiated its entry below the reported target, so an IPO priced at US$25 to 30 billion would leave Aker’s share price with little support from the Nscale mark alone. This is an inference about how the two share registers relate, not a sourced fact.

How it could play out

Nscale prices its IPO, possibly within weeks, and Aker’s stake becomes a daily-marked line worth two to three times its June carrying value. Aker’s third-quarter report restates net asset value on the listed price, and the stock either holds its 20% discount — in which case the research question is whether that discount is now too wide for a holding company that is half cash and listed AI infrastructure — or Aker signals a sell-down or distribution after the lock-up, and the discount compresses. Nscale itself then has to deliver: the order book converts into revenue only as sites in Norway, Portugal, the UK and the US come online against US$24 billion of equipment commitments, and the going-concern language in the filing shows what happens if the equity market shuts. The possible investment implication is a listed, discounted, currency-diversified way to hold the IPO with a catalyst calendar (pricing, first quarterly NAV, lock-up expiry) attached — or, in the other direction, a way to see an over-valued IPO reflected in a stock whose other assets are easier to value.

Questions worth asking

• What Nscale valuation does Aker’s current share price imply once the rest of the portfolio is marked to market and Aker’s historical discount range is applied? If the answer is already US$35 billion or more, the proxy is not cheap and the work stops there.
• What is Aker’s stake after the Nvidia notes, the Third Point notes and the primary offering convert, and at what discounts? The prospectus’s capitalisation table answers this and the 22.76% figure will be wrong by then.
• What does Aker intend to do with the stake — hold, sell down after lock-up, or distribute? Its actions on Cognite, HMH and Aker BioMarine in the past year are the evidence.
• Do holding companies re-rate when their largest asset becomes a liquid listed stake? The comparators are Kinnevik and Investor AB in Stockholm, Prosus with Tencent, and Exor with Ferrari — the discount narrowed in some cases and widened in others.
• Is there operating exposure beyond the stake: Nscale’s Norwegian sites need power and construction, and Aker’s group companies are Norwegian industrial contractors.
• What do CoreWeave and Nebius trade at on remaining performance obligations, and where would Nscale’s US$56.4 billion put it — because that multiple, not the Series C, sets the mark.

Where to look

• Aker ASA (AKER.OL) — the listed proxy: 22.76% fully diluted owner of Nscale at June 30, 2026, with the stake carried at NOK 32 billion and the stock at a discount to net asset value
• Nscale (NYSE: NSCL, pending) — the direct expression once listed; the filing carries the going-concern discussion and the capitalisation table that resolves the ownership question
• CoreWeave (CRWV) and Nebius (NBIS) — the listed AI-cloud comparables whose multiples of contracted backlog will price the IPO
• Nokia (NOKIA.HE) — a Series C investor and network supplier to Nscale; a much smaller and less direct exposure

Thesis check

The chain is well documented: the stake, the carrying value and the discount are in Aker’s own report, the IPO terms and losses are in Nscale’s SEC filing, and the catalyst is weeks away. The weak link is that Aker’s share price has already risen 35% since June 30, roughly what a US$35 billion listing with an unchanged holding-company discount would justify, so the opportunity depends on either the discount behaving differently once the stake is listed or the IPO pricing away from the reported target — and Nscale’s own filing, with a negative gross margin, a US$1 billion half-year loss and going-concern language, means the mark can fall as easily as rise.

Timing

Nscale IPO reported for as early as late September 2026; Aker ASA reports third-quarter net asset value in the autumn

Sources

Nscale Form S-1, SEC, Sep 18 2026 · Fortune, Sep 21 2026 · Bloomberg, Sep 18 2026 · SiliconANGLE, Sep 18 2026 · Aker ASA Q2 2026 results release, Jul 16 2026 · Datacenter Forum, Mar 10 2026 · Yahoo Finance, Sep 2026