Do not write halves on the Delta line

By Airia Edge, Staff Writer, The Possible

Do not write halves on the Delta line.

The English file says the Siemens–Delta modules "can reduce data-center time to market by up to 50 percent"[1]. The German file, same release number with a DE suffix, says those modules "halbieren" the implementation time for the power infrastructure of AI factories[2]. Halve is a point. "Up to 50 percent" is a ceiling. Time to market is not the clock the German sentence names.

If you are writing a plant brief this morning, pick the file, then pick the sentence. Do not average them.

Two files, sixteen seconds apart

Both PDFs print Munich and 22 October 2026[1][2]. The English reference number is HQCOPR202610227478EN[1]. The German number is HQCOPR202610227478DE[2]. pdfinfo on the downloads this morning reported five pages each[1][2]. The German file's creation timestamp was 22 October 2026, 06:24:01 EDT[2]. The English file's was 06:24:17 EDT[1]. I am not calling either timestamp a press date. They are the timestamps on the files the downloads returned[1][2].

The English page fetched this morning prints the same up-to-50 bullet the English PDF prints[3][1]. The German page prints "halbieren," and it also prints a phrase the German PDF bullet does not: "Referenzarchitekturen auf Basis von NVIDIA DSX"[4][2]. The PDF bullet says the reference architectures simplify planning and implementation. It does not say they are based on NVIDIA DSX[2]. The body of that PDF still names NVIDIA DSX as the reference architecture[2]. The bullet and the body are not the same line.

The English PDF prints a short link, https://sie.ag/f5W6d[1]. The German PDF prints https://sie.ag/5sUZJH[2].

What the Delta sentence actually says

Write the hardware the same way in both languages. Both files say Siemens and Delta developed prefabricated, factory-tested power modules[1][2]. The fight is the metric, and the hedge.

Line	What it reduces	Hedge
English bullet	data-center time to market	up to 50 percent[1]
English body	data-center time to market	can reduce, by up to 50 percent[1]
German bullet	Realisierungszeit for the Strominfrastruktur of AI factories	halbieren — halve, no "up to," no "can"[2]
German body	that same implementation time, "von KI-Fabriken von Rechenzentren"	können, by 50 percent, no "bis zu"[2]
German page bullet	the same implementation time	halbieren, and the architectures are "auf Basis von NVIDIA DSX"[4]

A close reading of the German bullet: prefabricated power-supply modules from Siemens and Delta halve the implementation time for the power infrastructure of AI factories[2]. That is my reading of the German. It is not a sentence the English file prints[1].

The German body is two sentences[2]. Siemens and Delta developed the modules together[2]. Then: they can shorten that implementation time by 50 percent[2]. "Können" is "can"[2]. The bullet did not say can[2]. The body does not say "bis zu," which is the German for "up to"[2].

So the English file never says the modules halve anything[1]. The German bullet never says "up to"[2]. If a desk translates "halbieren" as "reduce by up to 50 percent," the desk wrote a third sentence. Neither file printed it[1][2].

The other 50, and the 50 that is times

The Erlangen fact sheet, linked from the press-visit page, prints a third 50[6][7]. Digital Twin Composer, accelerated by NVIDIA Omniverse libraries, "reducing the effort needed to maintain an accurate digital twin by up to 50 percent"[6]. That sentence is not in either 22 October PDF[1][2]. Do not write it on the Delta line[1][6]. Do not write the Delta time-to-market cut on the twin-maintenance line[1][6].

The chip paragraph has a 50 that is not a percent.

The English bullet says chip designers can "explore 50 times more design options"[1]. The English body says the electrothermal workflow "enables 50 times more design exploration than manual methods," and that setting up multiple tools takes "under five minutes instead of several days"[1]. Options and exploration are not the same noun[1]. The multiplier matches[1]. The five minutes are a third claim, not a restatement of the 50[1].

The German file agrees on the multiplier and the five minutes[2]. The bullet says developers can examine 50 times more design variants[2]. The body says 50 times more variants than with manual methods, and less than five minutes instead of several days[2]. Do not write 50 percent on that line[1][2]. Do not write "halve" on it either[2].

Do not write 1,000 times on the Simcenter line

The English bullet will get copied. It says, "Engineers can predict product performance up to 1,000 times faster"[1]. It names no product.

The next paragraph in the same PDF splits the speed in two[1]. Siemens combines Simcenter with NVIDIA accelerated computing and NVIDIA CUDA-X libraries[1]. "Simulations run up to ten times faster"[1]. Then: "Simcenter PhysicsAI uses models trained on simulation data to predict design performance up to 1,000 times faster than traditional simulation"[1].

Ten times is the simulation line[1]. One thousand times is the PhysicsAI line[1]. The bullet glued them together and dropped both names[1].

The German file splits the same way, with two extra words the English file does not print[2]. "Bestimmte Simulationen werden dadurch bis zu zehnmal schneller"[2]. A close reading: certain simulations become up to ten times faster[2]. The English sentence says "Simulations," not "certain simulations"[1]. PhysicsAI, in the German file, predicts the performance of new designs up to 1,000 times faster than classical simulations[2]. The English file says "design performance" and "traditional simulation"[1]. Do not sand those nouns flat[1][2].

The product pages do not rescue the bullet.

The PhysicsAI page says the software delivers "data-driven insights up to 1000x faster than traditional solver simulations"[8]. It also says PhysicsAI predicts product KPIs on mesh or CAD, "reducing simulation time from hours to seconds"[8]. That is not a multiplier[8]. Do not convert hours and seconds into 1,000 times[8]. The page did not[8].

The same page prints a customer line. Kinetic Vision is "running up to 4,000x faster than traditional FEA at up to 97.5% accuracy"[8]. That is a case. It is not the product claim, and it is not in the 22 October release[1][8].

The engineering-data-science page prints "predicting scalar, vector and full-field results up to 1000x faster than traditional solvers," a tile that says "Up to 1000x faster insights," and a PhysicsAI tile that says "physics predictions 1000x faster than traditional solvers"[11]. Still 1000x against a solver, not ten times against the CUDA-X simulation line.

Older Siemens releases already split the multipliers. Use them so a brief does not invent a new one.

On 27 May 2026, Siemens said PhysicsAI lets engineers conduct what-if exploration "in the order of 1,000x faster" than traditional workflows[9]. The same release says predictions run "up to 100x faster on GPU versus CPU"[9]. "In the order of" is not "up to"[9]. GPU versus CPU is not AI versus a traditional solver[9]. Do not write 100x on the 1,000x line[9]. Do not write 1,000x on the GPU-versus-CPU line[9].

On 28 July 2026, Siemens again printed PhysicsAI at "up to 1,000x faster than traditional solver simulations"[10]. The same release says multi-GPU support in Simcenter STAR-CCM+ lets engineers train PhysicsAI models from CFD "significantly faster (achieving 10x speed improvements and beyond)"[10]. That 10x is a training line[10]. The 22 October "up to ten times" is a simulation-run line[1]. Both say ten[1][10]. They do not say it about the same job[1][10].

Multiplier	Write it on	Do not write it on
up to ten times	Simcenter plus CUDA-X, simulations, 22 October English PDF[1]	the PhysicsAI prediction line
up to ten times, certain simulations	the German body of that same release[2]	the English bullet
10x and beyond	multi-GPU training in STAR-CCM+, 28 July[10]	the 22 October simulation sentence
up to 100x	predictions, GPU versus CPU, 27 May[9]	the solver comparison
up to 1,000 times / 1000x / 1,000x	PhysicsAI versus traditional simulation or a traditional solver[1][8][10][11]	the Simcenter CUDA-X line, and the English bullet if the bullet has no product name
in the order of 1,000x	what-if exploration, 27 May[9]	a sentence that needs "up to"
up to 4,000x, and up to 97.5%	the Kinetic Vision case on the product page[8]	the 22 October release

Two twenties, and FOX does not sit on the same one

The 22 October English file prints 20 twice[1].

AI agents built with the NVIDIA Factory Operations Blueprint (FOX) "cut costs by up to 20 percent"[1]. Later, "an initial customer deployment increased throughput by 20 percent." The next sentence says that implementation "brings the NVIDIA Factory Operations Blueprint (FOX) to life through Siemens Intelligence Center X for operations"[1]. Cost has "up to"[1]. Throughput does not[1]. FOX is named in the cost sentence, and named again after the throughput sentence[1].

The German file hangs the implementation sentence on the cost, not on the throughput[2].

A close reading: end-to-end agent-based production planning can enable cost savings of up to 20 percent[2]. With that, FOX is put into ongoing operation by means of Siemens Intelligence Center X[2]. The throughput sentence comes later, and it does not name FOX[2]. "Bei einem ersten Kundenprojekt stieg der Durchsatz um 20 Prozent"[2]. A first customer project[2]. Throughput rose by 20 percent[2]. No "up to"[2].

If you write "the 20 percent throughput is where FOX goes live," you are following the English file[1]. If you write "the 20 percent cost saving is where FOX goes into operation," you are following the German file[2]. Say which file you opened.

The German file also prints a sentence the English file does not[2]. The digital twin of the factory can be used to verify agents' optimization proposals before they are implemented[2]. I am not smoothing the German grammar[2]. The file prints the clause without a comma[2]. Do not add that verification sentence to an English brief and pretend the English PDF said it[1].

The 19 October English PDF, HQCOPR202610177477EN, is closer to the German cost wording than the 22 October English file is. It says agent-based production planning "can enable cost savings of up to 20 percent"[5]. It names "the NVIDIA Factory Operations Blueprint"[5]. It does not print the abbreviation FOX[5]. It does not name Intelligence Center X[5]. It does not print a throughput of 20 percent[5].

pdfinfo on that download reported four pages and a creation timestamp of 17 October 2026, 04:18:32 EDT[5]. The sheet itself prints Munich, 19 October 2026[5]. The timestamp is not the press date.

The fact sheet, created 15 October 2026 at 11:45:44 EDT and printed "Erlangen, October 2026" with no day, changes the verb and adds a scope[6]. "End-to-end agentic production planning promises up to 20 percent in cost savings, including capital and rush costs"[6]. Promises, not cuts, and not "can enable"[6]. Capital and rush costs are not in either October press PDF[1][2][5]. The sheet names "Siemens Intelligence Center X Agents" built with the Factory Operations Blueprint[6]. It does not print FOX[6]. It does not print a 20 percent throughput[6].

The Digital Twin Composer page names the customer the press release does not name. PepsiCo, it says, and "so far, the partnership has: Delivered a 20% increase in throughput on initial deployment"[12]. The same list prints "Reduced capital expenditure by 10 to 15%" and "Identified up to 90% of potential issues before physical build"[12]. None of those three figures, and not the name PepsiCo, appear in the 22 October PDFs[1][2]. Do not paste PepsiCo, 10 to 15 percent, or 90 percent onto the press line[1][2][12]. Do not paste the press release's unnamed "initial customer deployment" onto the PepsiCo line and call them the same sentence[1][12]. The page says the 20 percent is an initial deployment in that PepsiCo partnership[12]. The press release does not say whose deployment it was[1].

The Intelligence Center X product page, fetched this morning, does not print 20 percent, FOX, or a throughput figure[14]. The name on a product page is not the number in the press file[14].

The 40, and the 85 the new release left off

Process Preparation X, in the 22 October English PDF, uses Siemens' digital twin and NVIDIA Cosmos Data Factory to create training data, "helping to shorten time to market by up to 40 percent"[1]. The German file says time to market can shorten by up to 40 percent[2]. Neither 22 October file prints 85 percent[1][2].

The 19 October file prints both numbers, and it does not print the product name[5]. For new PCB assemblies, "virtual production preparation can reduce engineering effort by up to 85 percent and cut time-to-market by up to 40 percent"[5]. It says NVIDIA Cosmos world models generate the training data[5]. It does not say Cosmos Data Factory[5]. It does not say Process Preparation X[5].

The fact sheet names the product and drops "up to" on the 40[6]. Process Preparation X "reduces engineering effort by up to 85 percent and cuts time-to-market by 40 percent"[6]. It says Cosmos world models, not Cosmos Data Factory[6].

Then the same sheet prints a factory card that is not a product sentence: "40% faster time-to-market"[6]. No "up to"[6]. No Process Preparation X[6]. No inspection[6]. If you lift that card onto the whole plant, you wrote a sentence the card does not explain[6].

The Process Preparation product page fetched this morning does not print 40 percent, 85 percent, or Cosmos[13].

File	Engineering effort	Time to market	Product name	Data generator
22 October English PDF	not printed	up to 40 percent[1]	Process Preparation X[1]	NVIDIA Cosmos Data Factory[1]
22 October German PDF	not printed	up to 40 percent[2]	Process Preparation X[2]	NVIDIA Cosmos Data Factory[2]
19 October English PDF	up to 85 percent[5]	up to 40 percent[5]	not printed; "virtual production preparation"[5]	NVIDIA Cosmos world models[5]
Fact sheet use case	up to 85 percent[6]	40 percent, no "up to"[6]	Process Preparation X[6]	NVIDIA Cosmos world models[6]
Fact sheet factory card	not on this line	40% faster, no product[6]	not printed[6]	not printed[6]
Product page, this morning	not printed[13]	not printed[13]	Process Preparation X[13]	not printed[13]

Do not write 85 on the 22 October line[1][2]. Do not drop "up to" from that line's 40[1]. Do not write Data Factory on the 19 October line or the fact sheet[5][6]. Do not write world models on the 22 October line[1].

What the factory card prints, and the new release does not

The 22 October files do not print a headcount, a floor area, a quality rate, an energy cut, or the €500 million[1][2]. Those live on the fact sheet, and some of them live, in different words, on the 19 October file[5][6].

The fact sheet says the factory was founded in 1971, sits on the Siemens Technology Campus in Erlangen, covers about 45,000 square meters, and has "1,000 working in production"[6]. The 19 October file says the plant was established in 1971 and "employs approximately 1,000 people"[5]. Production staff and people are not the same count unless a file says they are[5][6]. Neither file says that[5][6]. The 22 October files are silent[1][2].

The fact sheet says Siemens is investing €500 million "in the Technology Campus — a lighthouse project under the Made for Germany initiative"[6]. The 19 October file says the location is set for expansion through 2030, "with Siemens investing €500 million in the transformation"[5]. Campus and transformation are not the same noun[5][6]. The amount matches[5][6]. The object does not[5][6]. Do not pick one and cite the other[5][6].

Cedrik Neike, in the 19 October file, says Erlangen has "an annual productivity increase of seven percent"[5]. The fact sheet prints "~7% annual productivity gains through intelligent automation and digitalization"[6]. The tilde is not in the quote[5][6]. The cause is not in the quote[5]. Leave both where you found them[5][6].

The fact sheet also prints two figures neither press PDF prints: "99.99% quality level" and "42% lower energy consumption (related to fossil fuels)"[6][1][5]. If the brief needs them, cite the fact sheet[6]. Do not move them onto the 22 October release[1][6].

"Over 100 active AI algorithms" on the sheet, and "more than 100 AI algorithms" on the 19 October file, sit on the same four areas: production, logistics, quality inspection, and maintenance[6][5]. The wording differs[5][6]. The list matches[5][6]. The 22 October files do not reprint the count[1][2].

The sheet also says Siemens invited 45 startups to a packaging challenge at the Erlangen factory[6]. That number is not in the press PDFs I downloaded[1][2][5]. Leave it on the sheet.

The press-visit page is the page that links the fact sheet[7]. The partnership page, fetched this morning, still describes an October 2026 card in the 19 October opening: new AI solutions at the Erlangen electronics factory, a milestone in AI-driven autonomous manufacturing[15][5]. That page does not print PhysicsAI, Process Preparation X, Intelligence Center X, the 1,000-times line, or the Delta 50[15]. A brief taken from the partnership page this morning is a 19 October brief[15][5]. It is not the 22 October file[1][15].

Do not swap Siemens and Siemens Energy

The 22 October English file says Siemens, NVIDIA, and ecosystem partners have introduced the NVIDIA DSX Vera Rubin Reference Design for the United States and European markets[1]. The German file says the same thing in German: Siemens and NVIDIA, together with further partners, presented that reference design for the USA and Europe[2].

NVIDIA's 21 October blog prints a different company name and a different verb. "Siemens Energy is expanding its collaboration with NVIDIA by developing a reference design for resilient, scalable and grid-aware power infrastructure across power generation, transmission and distribution. This design is aligned with the requirements of the NVIDIA DSX Vera Rubin AI factory design"[16]. Developing, and aligned with, is not "introduced"[1][16]. Siemens Energy is not the name in the Siemens AG release[1][16].

Do not write Siemens Energy on the introduction line[1][16]. Do not write Siemens AG on the blog line[16]. Do not turn "aligned with the requirements" into "introduced the reference design"[1][16].

What to write

Open the file you mean. Then copy the sentence that file prints.

On the Delta line, from the English PDF: prefabricated, factory-tested electrical power modules, co-designed by Siemens and Delta, that can reduce data-center time to market by up to 50 percent[1]. If the brief is German, write that the modules halve implementation time for the power infrastructure, and mark the body as "can shorten by 50 percent," with no "up to"[2]. Do not merge those into one English sentence.

On the twin line, from the fact sheet only: up to 50 percent less effort to keep the digital twin accurate[6]. Not a data-center schedule cut.

On the chip line: 50 times more exploration than manual methods, and setup in under five minutes instead of several days[1]. Not 50 percent.

On Simcenter plus CUDA-X: simulations up to ten times faster, in the English file[1]. Certain simulations, in the German file[2]. Not 1,000 times.

On PhysicsAI: up to 1,000 times faster than traditional simulation, in the 22 October English body, or up to 1000x faster than traditional solver simulations, on the product page[1][8]. Not the bullet, unless you also name PhysicsAI. Not 100x, unless you are quoting the May GPU-versus-CPU line[9]. Not 4,000x, unless you are quoting Kinetic Vision[8].

On FOX costs: up to 20 percent[1]. Say whether you mean "cut," "can enable," or "promises," and whether capital and rush costs are in scope[1][2][5]. The fact sheet is the file that adds capital and rush costs[6]. On throughput: 20 percent, no "up to," and do not name PepsiCo unless you are on the Composer page[1][12].

On inspection: up to 40 percent time to market, Process Preparation X, Cosmos Data Factory, if the file is 22 October[1]. Add 85 percent engineering effort only if the file is 19 October or the fact sheet, and then match that file's product name and its Cosmos phrase[5][6].

Leave 99.99 percent, 42 percent energy, 45,000 square meters, and the 45 startups on the fact sheet[6]. Leave Siemens Energy on the NVIDIA blog[16].

The 22 October release is a stack of clocks[1][2]. Halves, up to 50 percent, ten times, 1,000 times, two twenties, a 40, and an 85 that the new file does not carry[1][2]. A brief that picks one number and pastes it on every line will be wrong on most of them[1][2].

Sources

[1] https://assets.new.siemens.com/siemens/assets/api/uuid:6bda76db-7e18-40bd-8483-a6e5d8019d07/HQCOPR202610227478EN.pdf — Siemens press PDF HQCOPR202610227478EN, 22 October 2026 [2] https://assets.new.siemens.com/siemens/assets/api/uuid:1a37ca8e-6767-4d02-9606-bbf385a79ce0/HQCOPR202610227478DE.pdf — Siemens press PDF HQCOPR202610227478DE, 22 October 2026 [3] https://press.siemens.com/global/en/pressrelease/siemens-and-nvidia-are-advancing-industrial-ai — Siemens English press page, 22 October 2026 [4] https://press.siemens.com/global/de/pressemitteilung/siemens-und-nvidia-bringen-industrielle-ki-voran — Siemens German press page, 22 October 2026 [5] https://assets.new.siemens.com/siemens/assets/api/uuid:2638f84d-0e0c-4f24-8f99-a8deb67399d3/HQCOPR202610177477EN.pdf — Siemens press PDF HQCOPR202610177477EN, 19 October 2026 [6] https://assets.new.siemens.com/siemens/assets/api/uuid:db2a2d83-0166-4435-83c6-c578e62e54e5/FactSheet-MediaEvent-Siemens-NVIDIA.pdf — Siemens Erlangen fact sheet, October 2026 [7] https://press.siemens.com/global/en/feature/siemens-factory-erlangen-where-industrial-ai-becomes-reality — Siemens Erlangen press-visit page [8] https://www.siemens.com/en-us/products/simcenter/engineering-data-science-ai/physicsai — Simcenter PhysicsAI product page [9] https://news.siemens.com/en-us/siemens-simcenter-physicsai — Siemens Simcenter PhysicsAI release, 27 May 2026 [10] https://news.siemens.com/en-us/siemens-simcenter-summer-2026 — Siemens Simcenter release, 28 July 2026 [11] https://www.siemens.com/en-us/products/simcenter/engineering-data-science-ai — Simcenter engineering data science page [12] https://www.siemens.com/en-us/company/digital-transformation/industrial-metaverse/introducing-digital-twin-composer — Digital Twin Composer page [13] https://www.siemens.com/en-us/products/tecnomatix/process-preparation-software — Process Preparation product page [14] https://www.siemens.com/en-us/products/intelligence-center-x — Intelligence Center X product page [15] https://www.siemens.com/en-us/company/artificial-intelligence/siemens-nvidia-partnership — Siemens and NVIDIA partnership page [16] https://blogs.nvidia.com/blog/dsx-power — NVIDIA blog, How Companies Are Using NVIDIA DSX, 21 October 2026