Tesla CEO Elon Musk told investors on the company’s Q3 2026 earnings call that Tesla is “almost sold out of all Megapacks for 2027, and we’re ramping production as quickly as possible to meet demand,” according to Yahoo Finance. The call took place on October 21, 2026, and the remark is now circulating well beyond the energy trade press. The reason is simple. The same hunger for electricity that powers every new AI training cluster is now showing up as a sold-out signal on a battery product.
Musk’s line landed in a quarter where Tesla’s car business looked soft and its storage business looked strong. Our Tesla Q3 2026 earnings breakdown covers the headline numbers. This piece goes narrower and deeper on one question: what does a nearly sold-out Megapack order book tell us about the real constraints on AI infrastructure, and how far can the story be pushed before it outruns the evidence?
One housekeeping note before the analysis. Social posts about this story tend to say Tesla is building Megapacks at the absolute limit of what is humanly possible and that nearly all of 2027 output is gone. The on-record wording is more measured. Tesla said it is ramping “as quickly as possible” and “almost sold out.” We stick to that wording throughout and flag where a popular claim goes beyond what executives actually said. For wider context on the machine-learning infrastructure race, see our AI and machine learning coverage.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Tesla Actually Said on the October 21 Call
Three on-record statements anchor the story. Musk said Tesla is “almost sold out of all Megapacks for 2027.” He also said the company is “ramping production as quickly as possible to meet demand.” In a separate remark reported by multiple outlets, he said, “We are increasing production of Megapacks as much as possible.” All three came from the Q3 2026 earnings call on October 21.
Tesla CFO Vaibhav Taneja added the demand detail. He said the energy business is witnessing extreme demand from data centers. That sentence carries the AI link in this story, because it comes from the executive who answers to investors on revenue and margin, not from a marketing page.
Two things are missing from these statements. Neither executive gave a percentage of 2027 capacity that remains unsold, and neither named a price per unit. Readers who see a headline claiming a precise sell-through figure are reading someone’s extrapolation. The honest summary is that Tesla’s 2027 Megapack book is nearly full, demand is described as extreme in the data center segment, and production is being raised.
Reading “Almost Sold Out”: What the Wording Supports
“Almost sold out” is a sales phrase, and sales phrases deserve a careful read. It tells us the order book for next year is close to full. It does not tell us how many gigawatt-hours that represents, who the buyers are, or how firm the contracts are. Companies often count signed supply agreements, framework deals, and deposits differently when they talk to investors.
The table below separates what the sourced material confirms from what the viral version of the story adds. We built it to keep the reporting clean, and it doubles as a quick checklist for anyone citing this story.
| Claim in circulation | Status | What the sources support |
|---|---|---|
| Tesla is almost sold out of Megapacks for 2027 | Confirmed | Musk’s on-record wording on the Oct. 21 call |
| Production is rising | Confirmed | “As quickly as possible” and “as much as possible” |
| Data centers drive extreme demand | Confirmed | Taneja’s statement on the same call |
| Building at the absolute human limit | Not confirmed | Not the reported wording |
| An exact share of 2027 output is sold | Not confirmed | Only “almost sold out” is on record |
| A specific Megapack price | Not confirmed | No confirmed price in sourced material |
| AI data centers are the only reason for demand | Not confirmed | Reports also point to utilities and renewable developers |
The last row matters most for the framing of this story. Data centers are a major demand source, and Taneja said so plainly. But utilities and renewable-energy developers also buy Megapacks, and reports on the call mention them. Anyone who writes that AI alone sold out Tesla’s 2027 storage book is claiming more than the evidence shows.
Why AI Data Centers Want Batteries Next to Their Racks
An AI training campus behaves differently from the data centers of a decade ago. Load swings can be sharp, because thousands of accelerators ramp up and down together as jobs start, checkpoint, and finish. Utilities dislike that kind of load, and a grid connection that cannot flex around it becomes a bottleneck. A battery system sitting beside the campus can absorb spikes, fill short gaps, and keep the facility running through brief disturbances.
There is a second pressure that has nothing to do with chips. Getting a large new grid connection approved and built takes time, and local politics can slow it further. Our reporting on a Maryland data center warning tied to $215M shows how state-level resistance is hardening in some regions. When the grid cannot deliver everything a campus wants on day one, on-site storage becomes a way to start earlier and ramp more smoothly.
Power and heat travel together, too. The same buildings that need more electricity also need more cooling, which is why we tracked AI cooling water use reaching 222 billion liters. Storage does not solve the cooling problem, but it belongs to the same cluster of constraints. Operators who plan power, water, and land together will ship capacity faster than those who plan them one at a time.
Behind-the-meter storage in plain terms
Behind-the-meter means the battery sits on the customer’s side of the utility meter. The owner can charge it when power is cheap or plentiful and discharge it when demand peaks or the grid wobbles. For a data center, that turns an unpredictable load into one the utility can manage. For a utility, it reduces the size of the connection it must build.
The 110 GWh Factory Footprint Behind the Order Book
Tesla’s stated annual capacity for Megapack manufacturing spans three sites. Megafactory Texas is listed at 50 GWh, the California factory at 40 GWh, and Shanghai at 20 GWh. Added together, that is 110 GWh of stated annual capacity. These are the company’s capacity figures, not shipment totals, and a factory rarely runs at its stated maximum from the first day.
| Site | Stated annual capacity | Share of 110 GWh total |
|---|---|---|
| Megafactory Texas | 50 GWh | 45.5% |
| California | 40 GWh | 36.4% |
| Shanghai | 20 GWh | 18.2% |
| Combined | 110 GWh | 100% |
The shares are our own arithmetic on Tesla’s stated numbers. The Texas site alone accounts for nearly half the total, which explains why so much of the near-term story hangs on how quickly that factory ramps. Reports tie Megapack 3 and Megablock to the newer Texas facility, though the sourced material does not give a delivery schedule for either.
A sold-out year against 110 GWh of stated capacity is a striking pairing. It suggests either that demand has outrun even a very large build-out, or that real output sits well below the stated ceiling, or both. Tesla has not published the split, so we flag it as an open question rather than pick a side.
Apollo Power’s 260-Unit Order Shows the Deal Shape
One reported transaction gives the demand story a concrete face. Apollo Power agreed to buy 260 Megapack 2XL units, described as approximately 1 GWh, with deliveries running through the end of 2027. That timing lines up neatly with Musk’s remark about 2027 being almost sold out. A single order of this size, delivered across the same window, is exactly the kind of commitment that fills a production calendar.
The order also shows how storage contracts work. Deliveries arrive in batches over many months, so the customer locks in supply long before the last unit ships. That structure rewards buyers who commit early and punishes late arrivals, which is part of why a “sold out” signal can propagate quickly. Once a few large buyers reserve a year of output, everyone else faces a longer queue.
Caution applies here too. Reports on the Apollo deal do not confirm a price, and the sourced material does not establish how much of the order maps to AI data centers versus other uses. Treat the 1 GWh figure as the reported scale of a single deal, not as a proxy for the whole market.
Megapack 2XL, Megapack 3 and Megablock Explained
Three product names keep surfacing in coverage of this story, and readers often mix them up. The Megapack 2XL is the unit in the Apollo order. Megapack 3 and Megablock are newer products linked to the Texas factory. Beyond those names, the sourced material gives no confirmed specifications, so we do not quote capacities, dimensions, or prices for them.
| Item | What is confirmed | What is not confirmed |
|---|---|---|
| Megapack 2XL | 260 units in the Apollo Power deal, about 1 GWh | Price per unit |
| Megapack 3 | Product name linked to the newer Texas factory | Specs, price, delivery dates |
| Megablock | Product name linked to the newer Texas factory | Specs, price, delivery dates |
| Megafactory Texas | 50 GWh stated annual capacity | Actual output rate |
| California factory | 40 GWh stated annual capacity | Actual output rate |
| Shanghai factory | 20 GWh stated annual capacity | Actual output rate |
The pattern is worth watching. Tesla is selling a current product while preparing its successors, which means 2027 capacity may be a mix of generations. If Megapack 3 and Megablock arrive in volume, they could change how many gigawatt-hours each factory can ship. That is a reason to treat the 110 GWh figure as a moving target.
Margins: What Musk Said About Profitability
Demand alone would be a good story. Demand with improving profitability is a better one. After describing the sold-out position, Musk said Tesla expects “gross margins on that to increase meaningfully,” as Tesla North reported the day after the call. He did not attach a number or a timeline to the statement.
The logic behind such a claim is easy to reconstruct. When a product is nearly sold out, the seller holds pricing power and can choose which orders to prioritize. Scale also helps, because a factory running closer to its design rate spreads fixed costs over more units. Neither mechanism guarantees a margin gain, and Tesla has not shown the math. Investors should wait for the next quarter’s energy segment numbers before treating the comment as a forecast they can model.
The comment also fits a broader theme from the quarter. Our earnings coverage noted that storage hardware offset a 3% auto drop, so the energy business is carrying more weight in Tesla’s story than it did a year ago. A margin lift on top of rising volume would reinforce that shift.
Data Center Demand in the CFO’s Words
The cleanest statement on AI-driven demand came from the finance chief. EV Wire quoted Taneja saying “The business is witnessing extreme demand from data centers, and we are doing our best to increase production.” The second half of that sentence is as telling as the first. It reads as an admission that supply, not demand, is the limiting factor.
For AI builders, that is a planning problem. A company that wants to energize a campus in 2027 and has not yet reserved storage may find the queue already long. Procurement teams that treated batteries as a late-stage line item will need to move them earlier in the schedule. The signal from Tesla is that the cheap, easy slots for 2027 are gone.
The statement also deserves an honest limit. Taneja did not say what share of storage orders come from data centers, and the sourced material does not provide that split. “Extreme demand” describes intensity, not proportion.
How Megapacks Compare With Other Ways to Power AI Campuses
Batteries are one option among several for a developer racing to add compute. We lack verified cost or performance data for the alternatives, so this section compares approaches qualitatively and labels the judgments as analysis. No figures here come from Tesla or its rivals.
Grid upgrades
A bigger utility connection remains the default answer, and it has the lowest operating complexity. Its weakness is timing. Interconnection work moves at the pace of utilities and regulators, not at the pace of chip deliveries. Batteries do not replace the connection, but they can shrink the size of the connection needed on opening day.
On-site generation
Some operators add their own generation to bypass grid delays. That route brings fuel supply, permitting, and emissions questions. Storage pairs with generation rather than competing with it, since a battery smooths the output of a generator and covers short gaps. In practice, many campuses will use both.
The competitive point for Tesla is that storage suppliers now sell into a market where buyers are short on time, not just on budget. That favors vendors who can promise delivery dates and who already run large factories. Tesla’s three-site footprint is a real advantage on that score, though rivals are also expanding. Our look at the AI infrastructure race and its historical echoes argues that suppliers of scarce inputs have often captured outsized gains in earlier build-outs. The same logic may apply here, with the caveat that batteries are easier to copy than leading-edge chips.
Historical Context: Storage Moves From Sideline to Core
For years, investors treated Tesla’s energy business as a footnote to its car sales. The Megapack line grew steadily out of that background, serving utilities and project developers who wanted large, factory-built battery blocks instead of custom-engineered installations. The shift now underway is that the buyer list is widening to include the companies building AI campuses.
Capital allocation tells a related story. Tesla has been trimming overall spending, as we reported when the company moved to cut capex to $20B, yet it keeps expanding Megapack production. That contrast suggests management sees storage as one of the better returns on each dollar spent. Whether the spending mix holds through 2027 will depend on how durable the demand proves to be.
A longer view helps, too. Every infrastructure boom produces a phase in which the supplier of a bottleneck input sells out in advance. Those phases can last years, and they can also end abruptly when demand normalizes or competitors catch up. A sold-out year is a strong data point, but it is one data point.
Market Impact: What Buyers, Investors and Rivals Should Watch
For AI developers, the practical impact is a longer lead time on storage. Teams planning 2027 campuses should confirm battery supply early, ask suppliers for firm delivery windows, and consider alternates in case a preferred vendor cannot deliver. The Tesla comment does not mean batteries are unavailable. It means the best slots are going first.
For investors, the story supports the view that AI spending flows well beyond chips. Hyperscaler budgets are huge, and we covered how Meta lifted AI capex to $145B while Tesla pulled back its own spending. Those budgets do not stop at servers. They also pay for land, power, cooling, and storage, and each of those categories can become a bottleneck with its own winners. Tesla’s energy segment has now put its name on that list.
For rivals, a sold-out Tesla is an opening. Customers turned away or quoted long lead times will talk to other storage suppliers. Expect competitors to emphasize availability and delivery speed in their pitches over the coming quarters. Tesla’s challenge is to convert its head start into long-term contracts before that window closes.
Risks and Open Questions
Several uncertainties sit under the headline. First, “almost sold out” is not audited. Order books can include contracts that slip, shrink, or cancel. Second, factory ramps rarely go perfectly, and Megafactory Texas is the largest piece of the stated capacity. A slow ramp there would push deliveries later and stretch the sold-out period.
Third, the pricing picture is blank. Without a confirmed price, nobody outside Tesla can size the revenue implied by a sold-out year. Fourth, demand concentration is unknown. If a few large buyers dominate the book, a single delay on their side would ripple through the schedule. Finally, policy matters. Utility rules, data center permitting fights, and trade rules can all shift the economics, and none of them are in Tesla’s control.
None of this undermines the core fact. It does argue for treating the story as a strong early signal rather than a settled outcome.
Five Predictions for Tesla Megapacks and AI Power
These are our own analytical judgments, not company guidance. Each rests on the confirmed facts above plus ordinary reasoning about how supply-constrained markets behave.
- Tesla will give more detail on the energy order book within two quarters. Investors now know the headline, and they will press for a gigawatt-hour figure, a customer mix, or a backlog number on upcoming calls.
- Lead times for 2028 storage will become the next talking point. Once 2027 is nearly full, buyers and analysts will shift attention to the following year, and early reservations will start to look standard.
- Megapack 3 and Megablock will show up in more 2027 conversations. If these products ship in volume from Texas, they will shape how much capacity each site can deliver.
- Rival storage suppliers will market availability as a selling point. Expect more announcements built around delivery speed and fixed dates.
- Data center operators will pair storage with other power strategies instead of relying on one. The mix of grid, on-site generation, and batteries will become a normal part of campus design.
Frequently Asked Questions
Is Tesla really sold out of Megapacks for 2027?
Not entirely. Musk said Tesla is almost sold out of all Megapacks for 2027. That wording leaves room for some unsold capacity, and Tesla has not published an exact percentage.
When did Tesla make the statement?
The remarks came from Tesla’s Q3 2026 earnings call, held on October 21, 2026. Outlets reported them over the following days.
Are AI data centers the only reason for the demand?
No. CFO Vaibhav Taneja described extreme demand from data centers, but reports on the call also point to utilities and renewable-energy developers as buyers. Data centers are a major driver, not the sole one.
How much Megapack capacity can Tesla build each year?
Stated annual capacity is 50 GWh at Megafactory Texas, 40 GWh in California, and 20 GWh in Shanghai. That adds up to 110 GWh, though actual output depends on how well each factory ramps.
What is the Apollo Power order?
Apollo Power agreed to buy 260 Megapack 2XL units, described as approximately 1 GWh, with deliveries through the end of 2027. Reports do not confirm a price.
How much does a Megapack cost?
No confirmed price appears in the sourced material, so we do not quote one. Any figure you see online should be treated as unverified until Tesla or a buyer discloses it.
What are Megapack 3 and Megablock?
They are newer Tesla storage products tied to the Texas factory. The sourced material does not provide confirmed specifications or delivery dates for either.
Will Megapack margins really rise?
Musk said Tesla expects gross margins on the product to increase meaningfully. He gave no figure or timeline, so the next few quarterly reports will show whether the claim holds.
