2025: A technology forecast for the year ahead

Image may be NSFW.
Clik here to view.

As has been the case the last couple of years, we’re once again flip-flopping what might otherwise seemingly be the logical ordering of this and its companion 2024 look-back piece. I’m writing this 2025 look-ahead in November for December publication, with the 2024 revisit to follow, targeting a January 2025 EDN unveil. While a lot can happen between now and the end of 2024, potentially affecting my 2025 forecasting in the process, this reordering also means that my 2024 retrospective will be more comprehensive than might otherwise be the case.

That all said, I did intentionally wait until after the November 5 United States elections to begin writing this piece. Speaking of which…

The 2024 United States election (outcome, that is)

Image may be NSFW.
Clik here to view.

Yes, I know I covered this same topic a year ago. But that was pre-election. Now, we know that there’s been a dominant political party transition both in the Executive Branch (the President and Vice President) and the Legislative Branch (the Senate, to be specific). And the other half of the Legislative Branch, the House of Representatives, will retain a (thin) Republican Party ongoing majority, final House results having been determined just as I type these words a bit more than a week post-election. As I wrote a year ago:

Trump aspires to fundamentally transform the U.S. government if he and his allies return to power in the executive branch, moves which would undoubtedly also have myriad impacts big and small on technology and broader economies around the world.

That said, a year ago I also wrote:

I have not (and will not) reveal personal opinions on any of this.

and I will be “staying the course” this year. So then why do I mention it at all? Another requote:

Americans are accused of inappropriately acting as if their country and its citizens are the “center of the world”. That said, the United States’ policies, economy, events, and trends inarguably do notably affect those of its allies, foes and other countries and entities, as well as the world at large, which is why I’m including this particular entry in my list.

Given that I’m clearly not going to be diving into other hot-button topics like immigration here, what are some of the potential technology impacts to come in 2025 and beyond? Glad you asked. Here goes, solely in the order in which they’ve streamed out of my noggin:

• Network Neutrality: Support for net neutrality, which Wikipedia describes as “the principle that Internet service providers (ISPs) must treat all Internet communications equally, offering users and online content providers consistent transfer rates regardless of content, website, platform, application, type of equipment, source address, destination address, or method of communication (i.e., without price discrimination)” predictably waxes or wanes depending on which US political party—Democratic or Republican, respectively —is in power at any point in time. As such, it’s likely that any momentum that’s built up toward ISP regulation over the past four years will fade and likely even reverse course at the Federal Communications Commission (FCC) in the four-year Presidential term to come, along with course reversals of other technology issues over which the FCC holds responsibility. Note that the “ISP” acronym, traditionally applied to copper, coax and fiber wired Internet suppliers, has now expanded to include cellular and satellite service providers, too.
• Tariffs: Wikipedia defines tariffs on imports, which is what I’m primarily focusing on here, as “designed to raise the price of imported goods and services to discourage consumption. The intention is for citizens to buy local products instead, thereby stimulating their country’s economy. Tariffs therefore provide an incentive to develop production and replace imports with domestic products. Tariffs are meant to reduce pressure from foreign competition and reduce the trade deficit.” The Trump administration, during his first term from 2016-2020, activated import tariffs on countries—notably China—and products determined to be running a trade surplus with the United States (tariffs which, in fairness, the subsequent Biden administration kept in place in some cases and to some degrees). And Trump has emphatically stated his intent to redouble his efforts here in the coming term, ranging up to 60%. The potential resultant “squeeze” problem for US domestic suppliers is multifold:
  • Tariff-penalized countries are likely to respond in kind with import tariffs of their own, hampering US companies’ abilities to compete in broader global markets
  • Those countries are likely to also tariff-tax exports (to the United States, specifically) of both product “building blocks” designed and manufactured outside the US—such as semiconductors and lithium batteries—and products built by subcontractors in other countries—like smartphones.
  • And broader supply-constraint retaliation, beyond fiscal encumbrance, is also likely to occur in areas where other countries already have global market share dominance due to supply abundance and high-volume manufacturing capacity: China once again, with solar cells, for example, along with rare earth minerals.

Perhaps this is why Wikipedia also notes that “There is near unanimous consensus among economists that tariffs are self-defeating and have a negative effect on economic growth and economic welfare, while free trade and the reduction of trade barriers has a positive effect on economic growth…Often intended to protect specific industries, tariffs can end up backfiring and harming the industries they were intended to protect through rising input costs and retaliatory tariffs.” Much will likely depend on if the tariffs to be applied will be selective and scalpel-like versus broadly wielded as blunt instruments.

• Elon Musk (and his various companies): Musk spent an estimated $200M financially backing Trump’s campaign, not to mention the multiple rallies he spoke at and the formidable virtual megaphone of his numerous posts on X, the social media site formerly known as Twitter, which he owns. A week post-election, the return on his investment is already starting to become evident. What forms could it take?
  • Electric vehicle (EV) manufacturer Tesla is a particularly lucrative revenue and profit generator for Musk. On one hand, Trump’s stated plan to eliminate EV rebates included in the Biden administration’s Inflation Reduction Act, as part of Trump’s broader wind-down of environmental regulations and increased protectionism for domestic suppliers of petrochemical-powered autos, might be interpreted as a bad thing for Tesla. On the other, already-mentioned planned tariffs on China, whose EV manufacturers are Tesla’s primary competitors at the moment, will likely be a good thing.
  • Further to the Tesla prognostication, the company has stated its aspirations to aggressively move into the robotaxi business, which necessitates full vehicle autonomy versus today’s advanced human-driver assistance (ADAS). Fast-tracking National Highway Traffic Safety Administration (NHTSA) and other agency approvals would be beneficial in actualizing Tesla’s robocar aspirations, as would more broadly be Department of Justice dismissals of existing and pending “Autopilot”-related lawsuits (along with those to inevitably come).
  • Starlink is another increasingly lucrative corporate entity. Regarding the low Earth orbit (LEO) satellite constellation (and the current Falcon 9 rockets that launch new members of it into orbit), Musk would undoubtedly welcome favorable FCC treatment both standalone and relative to satellite, cellular, and other scarce-spectrum competitors. And then there’s the next-generation Starship, queuing up for its pending sixth test launch as I type these words (and potentially launched by the time you read them). In the near term, Starlink and Musk aspire to use Starship to loft larger per-launch LEO payloads into orbit. And down the road, NASA and other large-payoff contracts also await.
  • And then there’s the Neuralink brain computer interface and the Food and Drug Administration (FDA)…
• Asia-based foundries: Taiwan, the birthplace of TSMC, and South Korea, headquarters of Samsung, are among the world’s largest semiconductor suppliers. Of particular note, as foundries they manufacture ICs for fabless chip companies, large and small alike. And although both companies are aggressively expanding their fab networks elsewhere in the world, their original home-country locations remain critical to their ongoing viability. Unfortunately, those locations are also rife with ongoing political tensions and invasion threats, whether from the People’s Republic of China (Taiwan) or North Korea (South Korea). All of which will make the Trump administration’s upcoming actions critical. Last summer, during an interview with Bloomberg, then-candidate Trump indicated that Taiwan should be paying the United States to defend it, that in this regard the US was “no different than an insurance company”, and that Taiwan “doesn’t give us anything”, accusing it of taking “almost 100%” of the US’s semiconductor industry. And during his first term, Trump also cultivated a relationship with North Korean dictator Kim Jong Un.
• Ongoing CHIPS funding: Shortly before the election, and in seeming contradiction to Republican party leader Trump’s earlier noted expressed regret about lost US semiconductor dominance, then (and likely again) House of Representatives Speaker (and fellow Republican) Mike Johnson indicated that the legislative body he led would likely repeal the $280B CHIPS and Science Act funding bill if his party again won a majority in Congress. Shortly thereafter, he backpedaled, switching his wording choice from “repeal” to “streamline”. Which will it actually be? We’ll have to wait and see.
• DJI and TikTok: Back in September, I mentioned that the US government was considering banning ongoing sales of DJI drones, citing the company’s China headquarters and claimed links to that country’s military and other government entities, resulting in US security concerns. Going forward, given Trump’s longstanding economic-and-other animosity toward China, it wouldn’t surprise me to see the proposed ban become a reality, which US-based drone competitors like Skydio would seemingly welcome (no matter that, to my earlier comments, China is already proactively reacting to the political pressure by cutting off battery shipments to Skydio). Conversely, although Trump championed a proposed ban of social media platform TikTok (a far more obvious security concern, IMHO) at the end of his first term, he’s now seemingly doing an about-face.
• Etc.: What have I overlooked or left on the cutting room floor in the interest of reasonable wordcount constraint, folks? Sound off in the comments.

Ongoing unpredictable geopolitical tensions

Image may be NSFW.
Clik here to view.

This was the first topic on my 2024 look-ahead list. And I’m mentioning it here just to reassure you that it hasn’t fallen off my radar. But as for predictions? Aside from comments I’ve already made regarding semiconductor powerhouses Taiwan and S. Korea, along with up-and-comer China, I’m going to avoid prognosticating any further on Asia, or on Europe or the Middle East, for that matter. Instead, I’ll just reiterate and slightly update two comments I made a year ago:

I’m not going to attempt to hazard a guess as to how the situations in Europe, Asia, and the Middle East (and anywhere else where conflict might flare up between now and the end of 2024, for that matter) will play out in the year to come.

and, regarding the US election:

Who has ended up in power, not only in the presidency but also controlling both branches of Congress, and not only at the federal but also states’ levels, will heavily influence other issues, such as support (or not) for Ukraine, Taiwan, and Israel, and sanctions and other policies against Russia and China.

That’s all, at least on this topic, folks! To clarify, if necessary, please don’t incorrectly interpret my reduced comparative wordcount for this section versus the previous one as indicative of perceived lower importance in my mind, or heaven forbid, of “inappropriately acting as if my country and its citizens are the center of the world,” to requote an earlier…umm…requote. It’s just that a year and a month after the October 7, 2023 attack that initiated the latest iteration of armed conflict between Israel and Iran’s Hamas and Hezbollah proxies, nearly three years into Russia’s latest and most significant occupation of Ukraine sovereign territory, and a few weeks shy of three quarters of a century (as I write these words) since the Republic of China (ROC) fled the mainland for the island of Taiwan…I’ve given up trying to figure out the end game for any of this mess. And echoing the Serenity Prayer, I realize there’s only so much that I can personally do about it. Speaking of prayer, though, one thing I can do is to pray for peace. So, I shall, as ceaselessly as possible. I welcome any of you out there who are similarly inclined to join me.

AI: Will transformation counteract diminishing ROI?

Image may be NSFW.
Clik here to view.

In next month’s 2024 look-back summary, I plan to dive into detail about why I feel the bloom is starting to fade from the rose of AI. Briefly, the ever-increasing resource investments:

• Processing hardware, both for training (in particular) and subsequent inference
• Memory and mass storage
• Interconnect and other system infrastructure
• Money to pay for all this stuff
• And energy and water (with associated environmental impacts) to power and keep cool all this stuff

are translating into diminishing capability, accuracy and other improvement “returns” on these investments, most recently noted in coverage appearing as I was preparing to write this section:

OpenAI’s next flagship model might not represent as big a leap forward as its predecessors, according to a new report in The Information. Employees who tested the new model, code-named Orion, reportedly found that even though its performance exceeds OpenAI’s existing models, there was less improvement than they’d seen in the jump from GPT-3 to GPT-4. In other words, the rate of improvement seems to be slowing down. In fact, Orion might not be reliably better than previous models in some areas, such as coding.

What can be done to re-boost the improvement trajectory seen initially? Thanks for asking:

• Synthetic data: This one is, I’ll admit upfront, tricky. Conceptually, it would seem, the more training data you feed a model with, the more robust its resulting inference performance will be. And such an approach is particularly appealing when, for example, databases of real-life images of various objects are absent perspectives from certain vantage points, of certain colors and shapes, and captured under certain lighting conditions. Similarly, a training algorithm’s ability to access the entirety of the world’s literature is practically limited by copyright constraints. But that said, keep in mind that both the quantity and quality of training data are critical. A synthetic image of an object that has notable flaws compared to its real-life counterpart, for example, would be counterproductive. Same goes for the slang and gibberish (not to mention extremist language and other garbage) that pervades social media nowadays. And while on the one hand you want your training data set to be comprehensive (to prevent bias, for example), proportionality to real life is also important in guiding the model to the most likely subsequent inference interpretation of an input. After all, there’s a fundamental reason why pruning to reduce sparsity is key to optimizing both model size and accuracy.
• Multimodal models: Large language models (LLMs), which I rightly showcased at the very top of my 2023 retrospective list, are increasingly impressive in their capabilities. But they’re also, admittedly somewhat simplistically speaking, “one-trick ponies”. As their name implies, they’re language-based from both input (typed) and output (displayed) standpoints. If you want to speak to one, you need to first run the audio through a separate speech-to-text model (or standalone algorithm); the same goes for spitting a response back at you through a set of speakers. Analogies to images and video clips, and other sensory and output data, are apt. Granted, this approach is at least somewhat analogous to human beings’ cerebral cortexes, which are roughly subdivided into areas optimized for language, vision and other processing functions. Still, given that humans are fundamentally multisensory in both input and output schema, any AI model that undershoots this reality will be inherently limited. That’s where newer multimodal models come in. Vision language models (VLMs), for example, augment language with equally innate still and video image perception and generation capabilities. And large multimodal models (LMMs) are even more input- and output-diverse. Think of them as the deep learning analogies to the legacy sensor fusion techniques applied to traditional processing algorithms, which I ironically alluded to in my 2022 retrospective.
• Continued (albeit modified) transition from the cloud to the edge: Reiterating what I initially wrote a couple of years ago:

  One common way to reduce a device bill-of-materials cost (BOM) is to offload as much of the total required processing, memory and other required resources to other connected devices. A “cloud” server is one common approach, but it has notable downsides that also beg for consideration from the device supplier and purchaser alike, such as:

  • Sending raw data up to the “cloud” for processing, with the server subsequently sending results back to the device, can involve substantial roundtrip latency. There’s a reason why self-driving vehicles do all their processing locally, for example!
  • Sending data up to the “cloud” can also engender privacy concerns, depending on exactly what that data is (consider a “baby cam”, for example) and how well (or not) the data is encrypted and otherwise protected from unintended access by others.
  • Taking latency to the extreme, if the “cloud” connection goes down, the device can turn into a paperweight, and
  • You’re trading a one-time fixed BOM cost for ongoing variable “cloud” costs, encompassing both server usage fees (think AWS, for example) and connectivity bandwidth expenses. Both of those costs also scale with both the number of customers and the per-customer amount of use (both of each device and cumulatively for all devices owned by each customer).
• Another popular BOM-slimming approach involves leveraging a wired or (more commonly) wireless tethered local device with abundant processing, storage, imaging, and other resources, such as a smartphone or tablet. This technique has the convenient advantage of employing a device already in the consumer’s possession, which he or she has already paid for, and for which any remaining “cloud” processing bandwidth involved in implementing the complete solution he or she will also bankroll. The latency is also notably less than with the pure “cloud” approach, privacy worries are lessened if not fully alleviated, and although the smartphone’s connection to the “cloud” may periodically go down, the connection between it and the device generally remains intact.
• For these and other reasons, in recent years I’ve seen a gradually accelerating transition from cloud- to edge-based processing architectures. That said, an in-parallel transition from traditionally coded algorithms to deep learning-based implementations has also occurred. And of late, this latter shift has complicated the former cloud-to-edge move, due specifically to the aforementioned high processing, memory, and mass storage requirements required to run inference on locally housed deep learning models. New system architecture variants to address both transitions’ merits are therefore gaining prominence. In one, the hybrid exemplified by Apple Intelligence along with Google’s Pixel phones’ conceptually equivalent approach, a base level of inference occurs locally, with cloud resources tapped as-needed for beefier-function requirements. And in the other, whereas “edge” might have previously meant a network of “smart” standalone edge cameras in a store, now it’s a network of less “smart” cameras all connected to an edge server at each store (still, versus a “cloud” server at retail headquarters).

• Deep learning architectures beyond transformers (and deep learning models beyond LLMs and their variants): The transformer, initially developed for language translation, quickly expanded into broader natural language processing and now also finds use for audio, still and video images, and various other applications. Similarly, usage of the LLM and its previously mentioned multimodal relatives is pervasive nowadays. However, when Yann LeCun, one of the “godfathers” of AI (and chief scientist at Meta), suggested earlier this year that the next generation of researchers should look beyond today’s LLM approaches and their associated limitations, accompanied by Meta’s public rollout of one such next-generation approach, and then more recently stated that today’s AI is as “dumb as a cat”, it caught a lot of industry attention. A recently published arXiv paper goes into detail on transformers’ limitations, along with the inherent strengths and shortcomings, current status and evolution potential of other “novel, alternative potentially disruptive approaches”. And I also commend to your attention a recent episode of Nova on AI. The entire near-hour is fascinating, and it specifically showcases an emerging revolutionary architecture alternative called the liquid neural network.
• New hardware approaches: Today’s various convolutional neural network (CNN), recurrent neural network (RNN) and transformer-based deep learning network architectures are well-matched to the GPU-derived massively parallel processing hardware architectures championed for training by companies such as NVIDIA, today’s dominant market leader (and also one of the leading suppliers for inference processing, although architectural diversity is more common here). That said, any one chip supplier can only satisfy a subset of total market demand, and the resultant de facto monopoly also leads to higher prices, all of which act to constrain AI’s evolutionary cadence. And that said, the emerging revolutionary network architectures and models I’ve just discussed, should they gain traction, will also open the doors to new hardware approaches, along with new companies supplying products that implement those approaches. To be clear, I don’t envision this emergent hardware, or the new network architectures and models that it supports, to become dominant in 2025 (or, realistically, even before the end of this decade). That said, I feel strongly that such revolutionary transformation is essential to, as I said earlier, re-boosting AI’s initial trajectory.

Merry Christmas (and broader happy holidays) to all, and to all a good night

I wrote the following words a year ago and couldn’t think of anything better (or even different) to say a year later, given my apparent constancy of emotion, thought and resultant output. So, with upfront apologies for the repetition, a reflection of my ongoing sentiment, not laziness:

I’ll close with a thank-you to all of you for your encouragement, candid feedback and other manifestations of support again this year, which have enabled me to once again derive an honest income from one of the most enjoyable hobbies I could imagine: playing with and writing about various tech “toys” and the foundation technologies on which they’re based. I hope that the end of 2024 finds you and yours in good health and happiness, and I wish you even more abundance in all its myriad forms in the year to come. Let there be Peace on Earth.

 —Brian Dipert is the Editor-in-Chief of the Edge AI and Vision Alliance, and a Senior Analyst at BDTI and Editor-in-Chief of InsideDSP, the company’s online newsletter.

 Related Content

• 2024: A technology forecast for the year ahead
• 2023: A technology forecast for the year ahead
• 2022 tech themes: A look ahead
• A tech look back at 2022: We can’t go back (and why would we want to?)
• A 2021 technology retrospective: Strange days indeed

The post 2025: A technology forecast for the year ahead appeared first on EDN.