We’ve got a stacked AMA for you today.
Kicking things off is a question on the status of semiconductor fabrication plants in the U.S. and the progress already being made to reshore the manufacturing of high-tech, bleeding-edge chips in the States.
We’ll also discuss the importance of ensuring AI training methodologies are grounded in factual, evidence-based, unbiased data when it comes to building large language models… the potential to power data centers by recycling the heat they generate… and the real problem with the U.S. biotech industry right now.
We’ve got a lot to cover today, so let’s get right to it…
Thank you for all your amazing research and for sharing that information with your subscribers so we can have some knowledge of what is going on in the tech world.
You have talked several times about onshoring semiconductor fabrication plants here in the U.S. Can you share what the status is of those facilities and when they might start production?
Thanks again.
– Greg L.
Hi Greg,
Thank you for writing in about this.
The good news is that there’s so much happening with U.S. onshoring initiatives right now, it would take pages and pages of information to thoroughly answer your question.
Something extraordinary is happening right now with onshoring production that I didn’t think I’d see in my lifetime. And it’s not just in the semiconductor industry, but that’s what we’ll focus on right now.
Here’s a map of the U.S. showing where onshoring is taking place in the semiconductor industry.
Red dots are the semiconductor manufacturing plants. Light blue dots are for semiconductor packaging projects, yellow shows semiconductor manufacturing equipment plants, green represents material suppliers, and the dark blue dot is a research and development facility.
(Click here to expand) | Source: Semiconductor Industry Association
What’s important to note is that to build a resilient, onshore semiconductor industry, it’s not just about manufacturing the chips.
Those chips have to be packaged into forms that can then be tested and eventually attached to a printed circuit board destined for some kind of electronics product.
To see all the companies with reshoring projects in the U.S., you can go here for a full list from the Semiconductor Industry Association.
To date, more than 100 projects have been announced across 28 states, encompassing $540 billion in investment. Even more exciting is that these investments will result in more than 500,000 new jobs.
One of the most high-profile reshoring projects – which is actually a foreign direct investment (FDI) project – has been TSMC’s investment to build out semiconductor manufacturing in Arizona.
This has already been an exciting year because TSMC started production of 4-nanometer (4nm) chips this January, which puts that manufacturing facility almost on the bleeding edge of mass production (right now, bleeding edge for production is 3nm).
Most of these projects will go live within two years, and the rest within three years. It’s use it or lose it, so the incentives are there to build and manufacture as quickly as possible.
These developments will transform supply chain resiliency and represent a major secular trend towards reshoring and building a resilient, more decentralized manufacturing architecture.
And it will be more efficient and ecological due to manufacturing being close to end customers.
Hey Jeff –
Had a question on LLMs, as I am nervous about deteriorating first-party data as LLMs advance.
My concern is that, historically, consumers have flocked to the internet to post their reviews and other experiences in such a way that these data sets are useful to the first-generation LLMs for training data.
But over time, as humans get more accustomed to leaning on LLMs, when seeking answers about products and services online, won’t we eventually reach a point where fresh, genuine, human-contributed data will become more and more scarce, such that the AIs will have no choice but to create synthetic data? Essentially, the potential for a feedback loop where the very success and convenience of LLMs could inadvertently lead to a decline in the rich human-generated data that initially fueled their learning?
And does this then give credence to SLMs being far more effective than LLMs in the long run, and where one should disproportionately invest? Regards.
– Kim D.
Hi Kim,
Excellent question. This is precisely why the training methodology of an LLM is so critically important. If the information provided in training is incorrect, so will be the output, which could be disastrous.
This is why what companies like Alphabet/Google, Meta, and Microsoft are doing is so terrible. They are training their LLMs on factually incorrect information to push ideology.
Doing so results in inherent risks to society and a false logic developed by the LLM that could make catastrophic mistakes in the future.
This is precisely why xAI’s primary goal is to create a maximum-truth-seeking AI in its quest to develop an artificial general intelligence (AGI). This is completely possible. Here are some key techniques that xAI uses to accomplish that end:
I don’t believe that the real challenge will be building a truth-seeking AGI. The real challenge will be educating the public that some AGIs have been intentionally programmed with bias and false information and are therefore not trustworthy.
A perfect analogy today is mainstream media. It is not journalism, it is propaganda that has been paid for by the U.S. government. We have been manipulated as a society. There are those who are trying to do the same with LLMs now, as mainstream media is suffering a quick death and is consumed by very few.
AGI will become somewhat of an operating system for society. Not just for our lives, but for our businesses, our source of news, and even for our governments. That’s why developing an evidence-based, truth-seeking AGI is so critically important.
Jeff,
There have been recent articles about thermoelectric generator improvements, such that they could be incorporated into a vehicle’s exhaust system as a significant boost to the vehicle’s electrical demands. Particularly useful, I suppose, with regard to hybrid vehicles.
Are you aware of any plans to upscale that concept for the utilization of the tremendous amount of waste heat that is a by-product of data centers’ operations?
That heat is an expense, since they had to buy the electricity that creates it, as well as pay for the cooling systems to mitigate it. If the hot air flow could be routed through a sufficiently efficient thermoelectric generator, it would simultaneously reduce the heat load while creating “free” electricity to somewhat reduce the electric expense overhead. Thanks again for your stimulating articles.
– Brent R.
Hi Brent,
This subject has been an area of study for decades. The idea is simple: there is a difference in temperature between two thermoelectric materials that creates a voltage difference, which can be captured and used to create electricity. That’s the “free electricity” that you referred to. This is known as the Seebeck effect.
And it does work. The real question is whether or not it makes sense economically.
Hybrid electric vehicles seem ideal for this, as you mentioned, because they produce some exhaust. Hybrid and conventional internal combustion vehicles have high-temperature exhaust. High temperatures result in a larger difference between the two thermoelectric materials, which means a higher voltage produced.
I remember reading a report about GMZ Energy – a company that attached a thermoelectric generator (TEG) to the exhaust of a Honda Accord and generated about 270 watts of electricity.
The reality is that the amount of electricity produced is so small, it is probably not enough to even support the added weight of the TEG to the car. This is why we haven’t seen this kind of approach in the automotive industry. Economically, it just doesn’t make sense.
As for data centers, power generators produce a lot more exhaust, but ironically, the temperature difference is much smaller because the exhaust of the generator isn’t as hot as what comes out of the exhaust of a car.
High-entropy materials like bismuth telluride (Bi₂Te₃) are showing promise with the theoretical potential to achieve 20% efficiency, compared to today’s 1–5% efficiency.
At such low levels of efficiency, a TEG doesn’t justify the expense of employing it. And bismuth telluride can cost as much as 15 times the cost of materials that produce the low-efficiency performance.
A much more promising short-term solution to better managing heat in data centers is to use liquid cooling techniques whereby the GPUs are submerged in a liquid that carries the heat away from the circuit boards.
The heat from the heated liquids can be recaptured and turned into electricity, sending the cooled liquids back through the system.
As perhaps an indication of the economics of TEGs applied to data centers… Coherent (COHR) – an excellent company with a strong data center-related business – does make TEGs, but for industrial and oil/gas applications.
If there were a return on investment for TEGs in data center applications right now, I’m confident Coherent would have a product for its data center customers.
While nothing has really been commercialized for automotive or data centers, there is still hope. One of the areas where generative AI will create breakthroughs is in materials technology, which may lead to higher efficiencies and TEGs that make economic sense.
[There was a recent] Boston Globe lead story regarding the possibility that Boston may lose its lead in biotech and that China is taking over the field.
The story went on to say that Real Estate and Higher Education will suffer with the shrinking of the Bio Tech Industry in Massachusetts and within the U.S., as funding for research shrinks and we are left behind by China’s rising Bio Tech Industry.
As we know, the biotech sector has been in what you call “The Biotech Winter.” Your thoughts on how all this will shake out for the Industry going forward?
– Richard D.
Hi Richard,
This is a multifaceted story that was not objectively covered by the Boston Globe. Understanding the drivers of what’s happening in biotech in the U.S. can’t be done in isolation, just looking from one perspective.
First and foremost, the biotech sector is still growing in the Boston area. There was 2.5% job growth in 2023, which is down from an average of 7.8% from 2020 through 2022.
Of course, there have been large layoffs in companies like Moderna (MRNA), which experienced a rapid decline in revenues as people stopped taking its experimental mRNA drugs for COVID-19.
Moderna’s share price is down 85% from all-time highs, and revenues have collapsed from $19.3 billion in 2022 to an estimated $2.1 billion this year. Naturally, Moderna will have to adjust to its new reality as a company in trouble.
Most journalists tend to focus on government-related grants from the National Institutes of Health (NIH) and government-funded biotech grants distributed to academic institutions. And the U.S. does lag compared to China in that particular regard.
This overlooks the vibrant venture capital funding of the biotech industry that is far more substantial than the U.S. government grants and any federal funding by China for its own domestic industry.
About 70% of all private capital that invests in the biotech industry invests in the U.S. And the Boston area is the largest concentration of investments by VCs in biotech.
Specifically, biopharmaceutical companies in Massachusetts received $7.98 billion in venture capital funding in 2024 – an increase of $220 million from VC funding in 2023. Below we can see the trend since 2013 for biopharma funding in Massachusetts from venture capital sources.
Massachusetts Biopharma Funding 2013 – 2024 | Source: MassBio
2020 and 2021 were banner years for VC funding in the Boston area, which is why the commercial developers fell hook, line, and sinker for building more laboratory space. But the reality of the biotech winter caused by failed pandemic policies, irresponsible fiscal policy, and runaway inflation resulted in the ugly stock market and rotation out of the biotech sector.
As we can see above, 2023 was the bottom for VC funding in Massachusetts, 2024 showed a small improvement, and I do believe that interest rates will fall by at least 100 basis points this year, and VC funding will be up compared to 2025.
Harvard, MIT, and Boston University – three academic institutions that were definitively on the wrong side of history with respect to the pandemic – all have massive endowments, some of the largest in the world. It is within their power to provide additional funding to their biotech-related programs and research.
And the venture capital community in biotech is very robust and on an upswing right now. I remain very bullish. The VC investment trend tends to lead the market. Interest rates will come down. And we’ll enter the biotech golden age with the Boston area at the epicenter of some of the biggest breakthroughs.
The Boston area’s larger problem is that it overbuilt commercial real estate during the 2020–2021 run-up in the biotech market.
That was a major mistake made by commercial developers. There is currently 16.1 million square feet of unleased lab space for the biopharmaceutical industry in the Boston area.
The real problem with biotech right now is actually interest rates. Fed Chair Jay Powell, again, ridiculously, refused to lower the Fed Funds rate from 4.5%.
Despite inflation being tamed, a healthy economy, and trade agreements getting hammered out (most recently India and the UK), Powell is holding rates at artificially high levels.
This also negatively impacts the real estate market as 30-year mortgage rates remain extremely high.
The biotech industry always suffers in high-interest-rate environments. And it thrives in low-interest-rate environments when institutional capital is in a “risk on” mode.
I wrote about this – and how the dangerous game Powell is playing is directly contributing to this enduring biotech winter – recently in The Bleeding Edge – A Dangerous Game…
It may sound odd, but I’m finding it hard not to think about interest rates right now. What Jerome Powell and the Federal Reserve are doing right now makes no sense at all.
[…]
Not only does the Fed have the “cover” and the data to significantly reduce interest rates, but the government really needs to. The U.S. government is facing a wall of $9.2 trillion of U.S. Treasuries that mature or need to be refinanced this year.
Some of it has happened already, but $8.55 trillion still needs to be dealt with. And lower interest rates result in lower yields for the Treasuries, which means less interest expense for the U.S. government.
This is a dangerous political game that Powell is playing right now to the detriment of the entire country. The U.S. is now paying more than $1 trillion in interest payments a year on its debt. It cannot continue.
Inflation is finally back under control. Rates need to come down… not just for the U.S. debt, but for the housing market, the automotive market, and of course the stock markets.
We won’t see a broad-based shift of institutional capital into high-growth small-cap stocks until this happens. That’s why we have so many attractive small-cap stocks right now to choose from.
The same goes for the early-stage biotech sector, which suffers for the same reason. We explored this dynamic yesterday in The Bleeding Edge – Beam Us Up, Biotech.
Enough with the political games. It’s time to bring rates down.
So, yeah, Powell needs to stop playing political games and pull interest rates down. That’s what we need to free up investment capital for the biotech industry.
That’s it for this week’s AMA. As always, you can reach me and the team right here.
Have a great weekend, everyone.
Jeff
The Bleeding Edge is the only free newsletter that delivers daily insights and information from the high-tech world as well as topics and trends relevant to investments.
The Bleeding Edge is the only free newsletter that delivers daily insights and information from the high-tech world as well as topics and trends relevant to investments.
